International Conference on Digital Transformation and Applications (ICDXA) Proceeding

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 0.43). This shows that product information is important in making purchase decision yet it will not stop people from buying it even if the information is not sufficient. This might be due to the availability and ease of access of information in the Internet and customer does not merely depend on one platform to get the full details of a product before purchase from one platform. Online Advertising shows the lowest agreement (Mean = 3.37) but it affects the shopping behaviour with high effect size (f = 0.57). This shows that customers pay attention to the advertisement and get influenced unconsciously in their shopping behaviour. As for the Information Security and Delivery Risk, students have high awareness and both has high impact on the shopping behaviour. As a conclusion, students nowadays are smart customers who are aware of many considerations during online shopping instead of focusing in advertisement only. This is the development of new digital culture among youngster due to the advancement of the Internet and technologies. Additional analysis is conducted to test whether gender shows different result in each pair OA → SB, PR → SB, DR → SB, IS → SB and the result is shown in Table 3. The result shows that there is not much difference between male and female customers in the shopping behaviour except for Delivery Risk which shows significant effects (F = 2.71; Sig = 0.004). Table 3. Two-Way ANOVA analysis. Two-way ANOVA F Sig. Online Advertising (OA) x Gender 0.48 0.924 Product Risk (PR) x Gender 0.15 0.997 Delivery Risk (DR) x Gender 2.71 0.004 Information Security (IS) x Gender 0.66 0.762 Dependent Variable – Shopping Behavior (SB) Independent Variables: OAxGender; PRxGender; DRxGender; ISxGender 6.0 CONCLUSIONS Factors that affect online shopping behavior are considered an important issue in e- commerce. This research proves that online advertisement, product risk, delivery risk and information security significantly affect college students’ online shopping behavior. To reduce the negative aspect of the associated factors that affect online shopping behavior among college students and to increase the possibility of experience among college students, e-marketers and e-retailers should examine into these concerns. For example, improving advertisement information, product information, delivery service, and secured information would definitely increase customers’ trust and thus reduce its negative impact on the online shopping behavior. Therefore, this study is important in contributing to the e-commerce development especially among e-retailers who are targeting on younger generation. There are several future works or recommendations that can be done in the future to improve the research study. Firstly, other associated variables/factors also can be analyzed in future research such as health risk, quality risk, social risk, time risk and etc. In addition, this research shows the need for further research to examine the influence of respondents' characteristics such as gender and experience on the composition of various associated factors dimensions and analyses their various influences on online customers' purchasing decision makings. 142 of 225 ICDXA/2021/14 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 REFERENCES Ahmad T (2017) Perceived usefulness and perceived ease of use of electronic health records among nurses: Application of Technology Acceptance Model. Informatics for Health and Social Care 43(4):379-389, https://doi.org/10.1080/17538157.2017.1363761. Ankita P and Smita M (2015) Factors of perceived risk affecting online purchase decisions of customers. Pacific Business Review International 8(2): 49-58. Anupkumar D and Snehal G (2019) A descriptive study of the effectiveness of Internet Advertising on customer buying behavior in Nagpur city. International Journal of Latest Engineering and Management Research 3(5). Anurag P and Jitesh P (2019) Factors affecting customer’s online shopping buying behavior. Proceedings of 10th International Conference on Digital Strategies for Organizational Success, http://dx.doi.org/10.2139/ssrn.3308689. Chakraborty D (2016) Factors affecting customer purchase decision towards online shopping: a study conducted in Gangtok, Sikkim. Adarsh Business Review 3(1), 11-18. Chiu HC, Hsieh YC, Li YC and Lee M (2005) Relationship marketing and customer switching behavior. Journal of Business Research 58(12): 1681-1689. Chong SX and Siti AS (2020) The relationship between security and online shopping intention. Research in Management of Technology and Business 1(1): 28-40. Cohen J (1988) Statistical power analysis for the social sciences. Hillsdale, New Jersey, Lawrence Erlbaum Associates. Emad YM (2013) The effect of perceived risk on online shopping in Jordan. European Journal of Business and Management 5(6): 76-88. Ganapathi P and Abu-Shanab EA (2020) Customer satisfaction with online food ordering portals in Qatar. International Journal of E-Services and Mobile Application 12(1):23, https://doi.org/10.4018/IJESMA.2020010104. Haider T and Shakib S (2017) A study on the influences of advertisement on consumer buying behavior. Business Studies Journal 9(1). ITA (2019) Malaysia e-commerce. ITA, Washington, DS, USA. Khandare AU and Suryawanshi PB (2016) Studying the impact of internet advertising on customer buying behaviour. International Journal of Basic and Applied Sciences 1(1): 28-33. Kok W, Omkar D, Zainudin J and Nurlida I (2020) Perceived risk factors affecting customers' online shopping behaviour. The Journal of Asian Finance, Economics and Business 6(4): 246-260. Lee JE and Jessie HCY (2018) Effects of price discount on customers’ perceptions of savings, quality, and value for apparel products: mediating effect of price discount affect. Fashion and Textiles 5(13), https://doi.org/10.1186/s40691-018-0128-2. Lu MH, Wan FWZ and Nurul HH (2016) The impact of perceived risks towards customer attitude in online shopping. International Journal of Accounting, Finance and Business 1(2): 13-21. Mae SA (2019) Factors influencing Japanese customers’ purchase intention of subscription streaming services. Ritsumeikan Asia Pacific University. Beppu, Ōita, Japan. Meskaran F (2015) The effect of perceived trust, perceived security and attitude on online purchase intention. Proceedings of the International Conference on Electronic Business: 417–426. Mohammad HMJ, Hossein RD, Mojtaba N, Amir P and Ahmad RA (2012) An analysis of factors affecting on online shopping behavior of customers. International Journal of Marketing Studies 4(5). 143 of 225 ICDXA/2021/14 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Neeraj G and Veena B (2016) Application of decomposed theory of planned behavior for m- commerce adoption in India. Proceedings of the 18th International Conference on Enterprise Information Systems 2: 357-367. Nur N, Azlinah M and Syaripah RSA (2020) Online advertising on customer purchasing behaviour: effective elements and its impact. Proceedings of the 3rd International Conference on Networking, Information Systems & Security: 1-7, https://doi.org/10.1145 /3386723.3387875. Onewo TT, Morakinyo DA and Akintan AA (2020) Effect of online advertising on customer buying behaviour of Internet users in Lagos state. Ilorin Journal of Human Resource Management 4(1): 1. Prasetyo YT, Tanto H, Mariyanto M, Hanjaya, C, Young MN, Persada SF, Miraja BA and Redi AANP (2021) Factors affecting customer satisfaction and loyalty in online food delivery service during the Covid-19 pandemic: its relation with open innovation. Journal of Open Innovation: Technology, Market, and Complexity 7(1): 76. Rasool A, Gupta V, Slathia B, Mahajan G (2017) Online shopping adoption and influencing factors: a study in Karnataka. Journal of Management Sciences and Technology: 29–40. Shahzad KA, Liang Y and Sumaira S (2015) An empirical study of perceived factors affecting customer satisfaction to re-purchase intention in online stores in China. Journal of Service Science and Management 8(3): 291–305. Sirkka LJ, Noam T and Lauri S (1999) Customer trust in an Internet store: a cross-cultural validation. Journal of Computer-Mediated Communication 5(2), https://doi.org/10.1111/ j.1083-6101.1999.tb00337.x. Statista (2020) Impacts of Covid-19 pandemic on the online purchase behavior among consumers in Malaysia as of May 2020, by age group. Hamburg, Germany. Vincent O, Andrew N, Alfa H (2018) Effects of online advertising on customer buying behaviour: study of Nigeria police academy cadets. Polac International Journal of Economics and Management Science 4(1). Yu D, Dong T and Liu R (2007) Study of types, resources and their influential factors of perceived risks in purchase online. Journal of Dalian University of Technology 28(2): 13-19. Zamzuri NH, Kassim ES and Shahrom M (2018) Entertainment gratification, informative gratification, web irritation and self-efficacy as motivational factors to online shopping intention. Management & Accounting Review 17(3): 95-108. Zhiyong JZ, Ke-Hai Y, Yujiao M, Meghan C, Han D, Ge J, Haiyan L, Agung S, Miao Y and Xinyi W (2015) Effect size calculator for one-way Anova. Notre Dame, Indiana. 144 of 225 ICDXA/2021/14 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 5G CYBERSECURITY: RISK ASSESSMENT AND INCIDENT RESPONSE IN THE HEALTHCARE INDUSTRY Yit Loong Teh1*, Yao Tong Tan1, and Siaw Lang Wong1 1 Faculty of Computing and Information Technology, Tunku Abdul Rahman University College, Kampus Utama, Jalan Genting Kelang, 53300, Wilayah Persekutuan Kuala Lumpur, Malaysia *Corresponding author: [email protected] ABSTRACT Rapid proliferation and developments in the field of telecommunication have led to a renewed interest in the fifth-generation technology standard for broadband cellular networks. As compared to the previous generations, 5G is capable of connecting numerous nodes simultaneously while offering pathbreaking implementations in enabling time-critical services such as autonomous vehicles, real-time drone manoeuvres, Industry 4.0, virtual reality, augmented reality, etc. Nevertheless, these conveniences may turn out to be security threats or vulnerabilities susceptible to cyberattacks. Apparently, thwarting cyberattacks is a never-ending journey, especially in the major economic sector – healthcare. It is an undeniable fact that healthcare is among the industries underpinning a nation’s economy. Hence, this paper will first forecast the trend of PHI breaches with Holt Winter’s model to emphasise the rise of cyber threats. Next, evaluating the risks of the chosen asset with ISO/IEC 27001 risk matrix. Then, compares the National Institute of Standards and Technology (NIST) SP 800-61 Incident Response Lifecycle and John Boyd’s OODA Loop; the latter offers significant advantages over the former concerning the adoption of situational awareness concepts, philosophy, and more. As a result, the number of PHI breaches increase tremendously due to the growth in the number of IT incidents and insider attacks. Keywords: 5G network, Incident response, Agile response, Cyberattack, Information security List of notations Lt is the value of Level Tt is the value of Trend ������������ is the value of Seasonality ������������ is the last period value 1.0 INTRODUCTION From the top, a mobile network, or otherwise known as the cellular network is a type of wireless network utilising radio waves to transmit data (Ramachandran et al., 2021). To date, there are five generations of cellular networks, with each providing different capabilities. Nonetheless, the fifth-generation cellular network is a game-changer as it offers lots of new features and opportunities like never before. Such technology otherwise known as 5G, introduces a ground-breaking kind of network that is neither confined by the context of only improved access to the internet services, nor only better phone call quality. But it is capable of connecting everyone and everything as today’s world is packed with the Internet of Things (IoT) devices (Liyanage et al., 2018). 5G unleashes its potentials by promoting earth- shattering benefits such as delivering sky-high peak data speeds, massive bandwidth, 145 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 extremely low latency, supporting millions of IoT devices, enabling smart cities, and so on (Bendale and Rajesh Prasad, 2018). However, it is tremendously difficult to achieve a perfect balance between rapid technological advancement and the risks they pose to the users. Hence, these advantages offered by 5G may turn into security risks that will threaten all involved aspects and elements. Additionally, the preponderant influence of 5G has left its security concerns and risks behind. This causes numerous governments and industries to be in a hurry to set up and deploying 5G-related technologies just to keep themselves up to par with the new national broadband standard without considering its security concerns thoughtfully (Arias et al., 2020). Significantly, 5G itself may be susceptible to various cyberattacks and may indirectly facilitate other cyberattacks to be carried out. Since cybersecurity is a never-ending journey, users have to always be prepared to brace for impacts. For the time being, there are only a meagre number of cybersecurity incident handling strategies designed for 5G-related incidents. Thus, this paper addresses the issues by evaluating two incident response (IR) models to determine if they are still feasible in containing evolved cyberattacks facilitated by 5G, including the NIST SP 800-61 Incident Response Lifecycle and John Boyd’s OODA Loop. Particularly, the assessment and discussion will be encompassing the context of two industries: healthcare and online retail since they are the potential heavy users of the 5G network (Gohar and Nencioni, 2021). These industries are also playing important roles in pushing smart cities forward. Nonetheless, such that there is a cyberattack capable of bringing down the entire smart city including the power grid, internet connection, and more, cutting-edge IoTs such as autonomous drones will no longer work. Therefore, an agile incident management plan must be there ensuring the continuity of service. An agile cybersecurity incident handling strategy is different from the traditional ones. The latter will usually be a static framework that is designed to contain pre-emergent cyberattacks. They are designed without the consideration of a network that is dealing with an array of new technologies such as network slicing, fog and cloud computing, IoT, etc. On the other hand, the former considers surrounding data gathered before a decision is made. An agile model is also known as a ‘data-driven model’ that gives the IR team some insights into the causes of the incident (Sun et al., 2019). For instance, the team may come up with a hypothesis where the alerts generated by an Intrusion Detection System (IDS) indicating there is an incident happening on the network. The rest of the paper is organised as follows. In Section 2, the background study and related work in 5G security are reported. Next, the methodology of the study is discussed in Section 3. Section 4 presents the results and discussions. Section 5 concludes the paper. 2.0 BACKGROUND AND RELATED WORKS This section consists of four sub-sections. Sub-section 2.1 wrapped up the potential applications of the 5G and its associated technologies in the healthcare industry. Then, sub- sections 2.2 and 2.3 emphasised the increased PHI breaches over the past ten years due to hacking, IT incidents, and unauthorised disclosure and introduced several cyberattacks that may pose security threats to the facilities, respectively. Sub-section 2.4 indicated that the current IR strategies need improvements to cope with the 5G environment. 2.1 5G and Its Application in the Healthcare Industry In light of the 5G network’s deployment, industries began to transform or upgrade their business operations, especially the healthcare industry. With reference to Li (2019), the potential applications are remote diagnosis, autonomous ambulance, smart hospital, Augmented Reality (AR) and Virtual Reality (VR) training, telemedicine, etc. Logically, 146 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 when the use of technologies has been widened, it expands the attack surface. Furthermore, one of the benefits of the 5G network offers is the enormous bandwidth. Gurusamy et al. (2019) have indicated that the Massive IoT (MIoT) concept will increase the size of an ordinary botnet tremendously. Thus, it pushes further the immediate need for a superlative IR strategy and response plan tailored for the healthcare industry. Medical practitioners are leveraging decent technologies associated with 5G like AR and VR to draw on the outcomes from their medical researches and also provisioning pre-operation simulation to get them prepared for the actual operation. 2.2 Cyber Threats Posed to the Healthcare Industry Protected Health Information (PHI) is an alluring target for cybercriminals as this information can be used in various ways to make a profit illegally (Kasperbauer, 2020). Not to mention that more information can be extracted with the aid of big data analysis. Moreover, Bai et al. (2017) have found out that 53% of 1150 PHI leakages reported are caused by internal attacks, either advertently or unwittingly. The remaining 47% comprise of external attacks like hackers and other cybercriminals. Therefore, it is an undeniable fact that the roll out of 5G and its associated technologies may worsen the situation, especially when the cybercriminals started to utilise such technology to facilitate cyberattacks. Referring to HIPAA Journal (2020), the major causes of PHI breaches are hacking, Information Technology (IT) incidents, and unauthorised disclosures, refer to Figure 1. Unsurprisingly, the number of PHI breaches shot up apace with the emergence of technologies. Number of PHI Breaches Due to Hacking/IT Number of PHI Breaches Due to Unauthorised Incidents between 2010 and 2020 Disclosure between 2010 and 2020 500 429 160 130 128 140 141 143 140 400 312 300 200 115 148 164 100 8 17 17 29 39 56 0 Number of PHI Breaches 120 103 Number of PHI Breaches 100 87 80 64 60 40 29 28 20 10 0 Year Year (a) (b) Figure 1. Number of reported data breaches in the healthcare industry due to (a) hacking/IT incidents and (b) unauthorised disclosure. Image source: HIPAA Journal (2020) 2.3 5G Cybersecurity Concerns Because none of the technologies was made perfect, Hussain et al. (2019) have pointed out that the 5G radio network is susceptible to Tracking via Paging Message Distribution (ToRPEDO) attack. This attack allows a cybercriminal to inject fabricated paging messages and launch a Denial-of-Service (DoS) attack by spamming unnecessary paging messages. Not only that, Oliver (2021) has mentioned that most of the existing security vulnerabilities that exist in 3G and 4G will remain in 5G such as locating users, intercepting SMSs, initiating DoS attacks, etc. 2.4 5G Incident Management According to Barona López et al. (2017), the existing IR strategies are not specifically designed for incidents in the context of the 5G network. This is due to the architectural design 147 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 of the 5G its associated technologies were not considered. The 5G network connects the infrastructure, hardware, software, data, people, and other components as a whole. However, these ordinary general-purpose models can be tailored to fit various organisational needs so long that the core structure of the response architecture is not affected. 3.0 METHODOLOGY In this paper, Holt Winter’s model was being used to forecast the trend of PHI breach incidents in the next five years, refer to equations (1) - (3). This model is feasible since such incidents happen periodically. The weekly seasonality was set to yearly seasonality, m = 365. Missing points were filled with interpolation method while duplicates were aggregated with the average value. Secondary data was gathered from HIPAA Journal that is available for public access online. (1) Lt= α(Yt - St - m) + (1 - α)[Lt - 1 + Tt - 1] (1) (2) Tt + 1 = β(Lt + 1 - Lt) + (1 - β)Tt (2) (3) St = γ(Yt - Lt) + (1 - γ)St - m (3) Next, a comparative, one asset will be chosen from the healthcare industry and to be assessed with the ISO/IEC 27001 risk matrix. In Table 1, the degree of impact is significant when the cyberattack threatens an asset’s confidentiality disclosure, serious when there is an integrity breach, and mild when it challenges its availability. Then, a comparative analysis was conducted between an ordinary cybersecurity IR model and a dynamic IR model with several metrics/criteria: number of steps, philosophy, situational awareness, hypothesis making, etc. In this context, the former is referring to the NIST SP 800-61 while the latter is the OODA Loop. These criteria were chosen because they are playing an influential role in facilitating the IR team to make a precise decision for containing an incident. Table 1. ISO/IEC 27001 risk matrix. Degree of Impact Threat Likelihood High Significant (High) Low Moderate 3 23 Serious (Moderate) 1 23 Mild (Low) 1 12 4.0 RESULTS AND DISCUSSIONS 4.1 Forecast of PHI Breach Trend With reference to Figure 2, the solid line in orange is referring to the predicted values. The trend of PHI breaches was predicted to increase 138.69% and 65.03% by 2025, respectively. In general, both types of PHI breaches are showing steep and almost consistent increments each year. At this point, there is a high likelihood of severe PHI breaches to happen in the coming years and it is also indicating that the commercialisation of the 5G and its associated technologies can also be one of the main factors that drive the figures up. As mentioned in the introduction, the 5G is still a new technology embeds with hidden vulnerabilities. Therefore, these forecasts have signified the need for a dynamic IR strategy. 148 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Forecast of PHI Breaches Due to Hacking/IT Forecast of PHI Breaches Due to Incidents by 2025 Unauthorised Disclosure by 2025 250 1200 200 1000 150 800 100 600 50 400 200 Number of PHI Breaches Number of PHI Breaches 0 0 20102012201420162018202020222024 2010 2012 2014 2016 2018 2020 2022 2024 Year Year (a) (b) Figure 2. Forecast of reported data breaches in the healthcare industry due to (a) hacking/IT incidents and (b) unauthorised disclosure by 2025. 4.2 Cyber Risk Evaluation on PHI With reference to Figure 3, PHI was adopted as the most anticipated asset for security protection by the healthcare industry players (Pool et al., 2019). According to the researchers, PHI is the most difficult asset to protect against cyberattacks as it involves multiple parties in the use of these data such as medical practitioners, registrars, etc. Furthermore, this data is valuable to hackers as they are able to make ends meet by disclosing it to the public, trade it to a third party through the dark web, or just simply make it inaccessible for a certain reason. Also, they may illegally modify the data. From the findings, both threats are equally posing a very critical cyber risk to PHI, especially when it comes to confidentiality and integrity breaches. On top of that, Figure 2 have shown that the plausibility for a cybercriminal to target PHI is very promising. As a result, hackers and careless personnel are probable to impact PHI. Healthcare businesses should be prepared to brace for cyber impacts in future. Figure 3. Evaluating cyber risk posed on PHI. 4.3 Comparison of IR Models In the previous discussions, the need for a new IR model is a must when healthcare businesses started to adopt and deploy 5G technologies in their organisations. As mentioned and discussed in the literature review, the 5G networks connect including but not limited to people, data, infrastructure, hardware, and software. And the incident can happen on any one 149 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 of these elements. Hence, the new IR strategy must show a high degree of agility than ever before. John Boyd’s OODA Loop is a dynamic IR model which is different from the ordinary procedural NIST SP 800-61 Incident Response Lifecycle. The summary of differences as discussed in the literature review and the comparison are tabulated in Table 2. Table 2. Summary of comparison between IR models. Number of Phases NIST SP 800-61 Boyd’s OODA Loop 4 4 Philosophy Procedural and evidence-based Data-driven Situational Awareness No Yes Unfolding No Yes Circumstances Hypothesis Making No Yes It is an undeniable fact where the world today is full of IoTs and surrounded by 5G- associated technologies such as smart cities, smart homes, smart factories, and more. In the healthcare context, As technology advances from day to day, the same goes for the evolvement of cyberattacks. Plus, the literature reviews have also raised some of the issues of the 5G network. At this point, the need for a dynamic IR model has been pushed further. Both IR models have four phases. One point that is noteworthy to mention is that the criteria of proceeding to the next phase are different and it varies from case to case. The philosophy of the NIST SP 800-61 Incident Response Lifecycle in identifying an incident is based on solid evidence provided. To further detail this, a good example would be an information espionage case where a suspicious individual extracted PHI data from a medical workstation and escaped from the building. From here, the evidence will be the workstation’s system logs and the surveillance system’s recordings. The lifecycle emphasises the analysis of evidence while collection and this will eventually cost more time to get to the eradication phase. In contrast, John Boyd’s OODA Loop will be based on data gathered from the environment before a hypothesis is made. For example, the hypothesis can be the leakage of a patient’s health information alongside other registered credentials. The formulation of the hypothesis will usually make use of experience and training knowledge. External key factors including culture and previous experience are to be taken into consideration. Besides that, both internal and external key factors are accountable for the formulation of a hypothesis of what activities are involved in an incident. Correlation of anomaly activities can be done in the Orient phase or analysis phase of the NIST SP 800-61 Incident Response Lifecycle. Whenever any ambiguities are found in the phases, the team shall loop back to the Observation phase. Not to mention that situational awareness helps the IR team to learn the attacker’s activities, attribution, capabilities, and intelligence (Van Os, 2021). Another factor is the flexibility of IR models. John Boyd’s OODA Loop has both Observe and Orient phases to gather and synthesis data and information from the incident environment. Such that the environment consists of millions of IoT devices including autonomous drones, cars, manufacturing facilities, smartwatches, smartphones, smart grids, and other relevant technologies, John Boyd’s OODA Loop is certainly playing an influential role in halting a cyberattack. The NIST SP 800-61 Incident Response Lifecycle emphasises the procedural steps taken to ensure the completeness of the handling process while John Boyd’s OODA Loop aims to minimise the impacts as quickly as possible. 150 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Furthermore, the Plan-Do-Check-Act (PDCA) methodology used to formulate various business continuity strategies that have been included in ISO 22301 standard is a Boyd’s OODA Loop lookalike. Nonetheless, Boyd’s OODA Loop depends heavily on unfolding circumstances whereas both PDCA and NIST SP800-61 Incident Response Lifecycle do not. Boyd’s OODA Loop has another unique characteristic called ‘Survive with own terms’. Such a characteristic enables the model to focus on synthesising an action out of an incomplete data set since it is almost impossible to identify every variable in the environment that the IR team is forced to deal with (Sharp, 2020). Therefore, that contributes to the reason of making a hypothesis in which the IR team believes that a particular decision will result in the highest probability for success while reducing any potential security risks. The gist here is the formulation of action from the Observe and the Orient phases given that the IR team is placed in a complex and mysterious environment which subject to regular and unpredictable change. From here, it is related to the 5G and its associated technologies as the environment are still embedded with leashed potentials or even a potential superb cyber threat. Since the NIST SP 800-61 is offering a procedural guideline in outlining cybersecurity frameworks, it primarily deduces the course of actions through thorough data analysis and step-to-step best practices. However, such a methodology will find it difficult to act proactively as it is a set of pre-planned procedures waiting to be executed. The 5G environment is consistently changing and truth to be told that fluidly changing plans and responses must be emphasised by businesses involved in the fourth industrial revolution like healthcare organisations. 5.0 CONCLUSIONS All in all, this paper has discussed the future of PHI breaches alongside the implementation of the 5G and its associated technologies in the healthcare industry. This paper has foreseen that more and more PHI will be leaked due to the growth of IT incidents and unauthorised disclosure. With the aid of Holt Winter’s seasonal model, healthcare businesses should be equipped with appropriate cyber defence strategies. In addition, this paper used ISO/IEC 27001 risk matrix to assess the level of risk posed to the PHI since it is playing an influential role in bridging medical practitioners and their patients. As a result, it shows that PHI is at risk in face of the worsening situation. Hence, the gradual need for a dynamic IR model alongside the emergence of 5G-associated technologies has been signified through the findings. Moreover, the comparison between IR models in terms of their philosophies, implementation of situational awareness concept, etc. have shown that John Boyd’s OODA Loop is the potential IR model that offers significant advantages over the NIST’s SP 800-61 Incident Response Lifecycle given that it has a higher degree of dynamicity in inferring actions to be taken in the face of cybersecurity incidents. 6.0 ACKNOWLEDGMENTS The authors would like to thank the Faculty of Computing and Information Technology, Tunku Abdul Rahman University College for the financial support and resources to carry out this study. Besides that, the authors would also like to thank the anonymous reviewers for their valuable comments and suggestions. REFERENCES Arias, R. et al., 2020. The Impact of 5G: Creating New Value across Industries and Society, The World Economic Forum. Bai, G., Jiang, J. (Xuefeng) and Flasher, R., 2017. Hospital Risk of Data Breaches. JAMA Internal Medicine, 177(6), p.878. Available at: http://annals.org/article.aspx?doi=10.7326/M15-2886. 151 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Barona López, L. et al., 2017. Towards Incidence Management in 5G Based on Situational Awareness. Future Internet, 9(1), p.3. Available at: http://www.mdpi.com/1999- 5903/9/1/3. Bendale, S.P. and Rajesh Prasad, J., 2018. Security Threats and Challenges in Future Mobile Wireless Networks. 2018 IEEE Global Conference on Wireless Computing and Networking (GCWCN). November 2018 IEEE, pp. 146–150. Gohar, A. and Nencioni, G., 2021. The role of 5g technologies in a smart city: The case for intelligent transportation system. Sustainability (Switzerland), 13(9), pp.1–24. Gurusamy, D., Deva Priya, M., Yibgeta, B. and Bekalu, A., 2019. DDoS risk in 5G enabled iot and solutions. International Journal of Engineering and Advanced Technology, 8(5), pp.1574–1578. HIPAA Journal, 2020, Healthcare Data Breach Statistics [Online]. Available at: https://www.hipaajournal.com/healthcare-data-breach-statistics/ [Accessed: 10 October 2021]. Hussain, S.R. et al., 2019. Privacy Attacks to the 4G and 5G Cellular Paging Protocols Using Side Channel Information. Proceedings 2019 Network and Distributed System Security Symposium. 2019 Internet Society, Reston, VA. Kasperbauer, T.J., 2020. Protecting health privacy even when privacy is lost. Journal of Medical Ethics, 46(11), pp.768–772. Available at: https://jme.bmj.com/lookup/doi/10.1136/medethics-2019-105880. Li, D., 2019. 5G and intelligence medicine—how the next generation of wireless technology will reconstruct healthcare? Precision Clinical Medicine, 2(4), pp.205–208. Liyanage, M. et al., 2018. 5G Privacy: Scenarios and Solutions. 2018 IEEE 5G World Forum (5GWF). July 2018 IEEE, pp. 197–203. Oliver, D., 2021, 5G security: everything you need to know about the security of 5G networks [Online]. Available at: https://www.5gradar.com/features/5g-security-5g-networks- contain-security-flaws-from-day-one [Accessed: 14 October 2021]. Van Os, R., 2021, The added value of the OODA loop to cyber security - part 3/3 [Online]. Available at: https://www.linkedin.com/pulse/added-value-ooda-loop-cyber-security- part-33-rob-van-os?trk=public_profile_article_view [Accessed: 9 September 2021]. Pool, J.K., Akhlaghpour, S., Fatehi, F. and Burton-Jones, A., 2019. Causes and Impacts of personal health Information (PHI) Breaches: A scoping review and thematic analysis. Proceedings of the 23rd Pacific Asia Conference on Information Systems: Secure ICT Platform for the 4th Industrial Revolution, PACIS 2019, (April 2020). Ramachandran, T., Faruque, M.R.I., Siddiky, A.M. and Islam, M.T., 2021. Reduction of 5G cellular network radiation in wireless mobile phone using an asymmetric square shaped passive metamaterial design. Scientific Reports, 11(1), p.2619. Available at: https://doi.org/10.1038/s41598-021-82105-7. Sharp, D., 2020, Planning ahead: The OODA vs PDCA methodology  [Online]. Available at: https://www.ideagen.com/thought-leadership/blog/planning-ahead-ooda-vs-pdca- methodology [Accessed: 12 October 2021]. Sun, N. et al., 2019. Data-Driven Cybersecurity Incident Prediction: A Survey. IEEE Communications Surveys & Tutorials, 21(2), pp.1744–1772. Available at: https://ieeexplore.ieee.org/document/8567980/. 152 of 225 ICDXA/2021/15 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 REVIEW ON LIGHT VERB CONSTRUCTIONS IN COMPUTATIONAL LINGUISTICS Kathleen Swee Neo Tan1*, Tong Ming Lim2, Chi Wee Tan1 and Wei Wei Chew3 1 Faculty of Computing and Information Technology 2 Centre for Business Incubation and Entrepreneurial Ventures 3 Faculty of Social Science and Humanities, Tunku Abdul Rahman University College, Kampus Utama, Jalan Genting Kelang, 53300, Wilayah Persekutuan Kuala Lumpur, Malaysia *Corresponding author: [email protected] ABSTRACT Light verb constructions (LVC) are an interesting phenomenon that occurs in many languages. It is a category of verbal Multiword Expressions (MWE) and has the canonical form of verb+noun (Constant et al., 2017; Cordeiro and Candito, 2019; Nagy T., Rácz and Vincze, 2020). Examples of LVCs include give help, make a decision, and take a walk. Identifying LVCs is essential for many natural language processing (NLP) applications which include summarization, machine translation, semantic parsing, question answering, and information extraction. The importance of LVC identification to these downstream applications has recently spurred a growing volume of work in both the field of linguistics as well as computational linguistics in various languages as it can potentially increase the performance of these tasks. This paper presents a review of existing work related to LVC identification by summarizing gaps identified and proposing some future work that could bring novel contributions. Keywords: Light verb constructions, Multiword expressions, Natural Language Processing 1.0 INTRODUCTION Multiword Expressions (MWE) are expressions that contain two or more words that are used together to convey a certain meaning. An MWE consists of a head word and one or more other words that are syntactically related to the head word. Recently, there has been a growing interest in verbal MWEs (Gantar, Krek and Kuzman, 2017; Savary et al., 2017; Ramisch et al., 2020) and Light Verb Constructions (LVC) (Vincze, Zsibrita and Nagy T., 2013; Chen, Bonial and Palmer, 2015; Cordeiro and Candito, 2019; Nagy T., Rácz and Vincze, 2020). A verbal MWE is an MWE which functions as a verbal phrase and has a head word which is a verb (Ramisch et al., 2018). Light verb constructions (LVC) is a category of verbal MWEs that plays an important role to the linguistic and natural language processing community for two reasons. Firstly, the verbal component of the LVC has a “light” meaning in the sense that it does not really contribute much to the meaning of the LVC. Secondly, it is more efficient to use the synthetic verbal counterpart of the verbs in the LVC (i.e., the verbs help, decide, and walk may be used to replace the LVC examples provided earlier). The interest in LVC has been demonstrated by work in the field of linguistics (Gilquin, 2019; Mehl, 2019; Bonial and Pollard, 2020) as well as computational linguistics in various languages (Taslimipoor, Fazly and Hamzeh, 2012; Cap et al., 2015; Nagy T., Rácz and Vincze, 2020). The importance of LVC identification in NLP applications can be observed from recent developments in verbal MWE annotations such as the enhancement of the annotation methodology to add a new LVC subcategory known as causative LVC which is applicable for 153 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 all languages (Ramisch et al., 2018). In a full LVC, its verb is semantically bleached completely. For example, in “give a speech”, the verb give does not at all imply the transfer of possession to the recipient. In causative LVCs, the verb is the cause or source of the event or state denoted by the noun. For example, in the LVC “give a headache”, the verb give means to cause a person to have a headache. In the work by Ramisch et al. (2018), corpora annotation for 20 languages were carried out. These annotated corpora are much needed not only for NLP, but also in the field of linguistics and translation studies. However, it is a very time-consuming and resource-intensive task to carry out the manual annotation that is required to create new annotated corpora, especially for under-resourced languages. In the remaining sections of the paper, we present an introduction to LVCs, followed by a discussion on existing work related to LVC identification. The final section summarizes gaps identified and proposes some important future work that could bring novel contributions. 2.0 WHAT ARE LIGHT VERB CONSTRUCTIONS? Multiword Expression (MWE) identification is important for natural language processing (NLP) tools such as part-of-speech (POS) taggers, semantic parsers and syntactic parsers, as well as downstream applications such as machine translation, emotion analysis, and question answering systems (Constant et al., 2017). MWEs have been categorized using various schemes and an overview of the subcategories of MWEs extracted and integrated from the work of (Constant et al., 2017) and (Ramisch et al., 2018) is shown in Figure 1. Figure 1. Overview of MWE categories. Light verb constructions (LVCs), also known as support verbs, are often described as being a complex predicate consisting of a verb+noun or verb+particle+noun that can be replaced by the corresponding verbal form of the nominal component in the LVC. For example, the LVC ‘make a review’ can actually be more efficiently replaced by its synthetic verb counterpart ‘review’ in a sentence. Table 1 illustrates the use of LVCs and their synthetic verb counterparts in sentences. 154 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Table 1. Examples of the use of LVCs and synthetic verbs in sentences. With LVC With synthetic verb counterpart We give a review of light verb constructions in We review light verb constructions in computational linguistics. computational linguistics. Jon took a brisk walk around the college this Jon walked briskly around the college this morning. morning. They will make a decision on the new product They will decide on the new product next week. next week. In linguistics study there remains differences in opinion on the definition of LVC (Bonial and Pollard, 2020). The main characteristic is that the verbs in LVCs are considered as ‘light’ as they contribute a lesser degree to the meaning of the complex predicate and the noun provides most of its meaning. In addition, these verbs are not interpreted in the literal sense. Consider the ‘heavy’ usages or literal meaning of the verbs used in these non-LVC examples: make implies the act of creating something (as in make a cake), take is an action that results in the possession of the object (e.g., take the plate from the cupboard), and give is the act of transferring an object to be in the possession of a subject (e.g., gave a bouquet of flowers). The heavy usage of these verbs is also known as productive verbs as their use indicate that they effect or produce some results (e.g., the creation of an object or transfer of possession). 3.0 LVC IDENTIFICATION LVC identification is the task of automatically detecting instances of LVCs in running text. As LVC is a subcategory of verbal MWEs, it shares the latter’s challenging characteristics of having complex structure, discontinuities, variability, and ambiguity (Savary et al., 2017). The task of automatic LVC identification is also complicated by the fact that the definition of LVCs across languages and even within a language remains ambiguous (Chen, Bonial and Palmer, 2015). For the English language, LVC identification is challenging as the syntax of its LVCs is generally the same as other verb+noun constructions (Bonial and Pollard, 2020). The main approaches to LVC identification can be broadly divided into rule-based, machine learning, and corpus-based statistical measures. 3.1 Rule-Based Rule-based approach for LVC identification uses rules which identifies syntactic patterns through the use of POS taggers. Vincze, Nagy T. and Berend (2011) identified LVCs using the following methods (both alone and combined): the syntactic rule method which considered n-grams that exhibited the verb+noun patterns using POS tagging; the stem method which extracted constructions consisting of a nominal component derived from a verb (e.g., make a decision) or was the same as its verbal form (e.g., give a review); and the most frequent method in which constructions were extracted if they contained a verb from a list of most frequent verbs used in LVCs (e.g., make, take, give, etc.). They evaluated these methods on their corpus which consisted of 50 Wikipedia articles in which 368 LVCs and 2929 nominal compound occurrences were marked. Their results showed that the combination of syntactic rules and the most frequent method achieved the best performance of 45.31% F1-score. They opined that their methods could be improved by extending the feature set and refining the syntactic rules. For under-resourced languages, the unavailability of reliable part-of-speech (POS) taggers would pose a problem in creating POS-rules for their language. 155 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 3.2 Machine Learning Chen, Bonial and Palmer (2015) proposed a model using Liblinear logistic regression with L2-regularization to perform binary classification of individual verb+noun pairs. The candidate dependency trees for training the classifier were selected based on a set of six most frequently used light verbs in the data and a list of eventive nouns (i.e., nouns that denote events) and stative nouns (i.e., nouns that denote states) that were curated from WordNet (Fellbaum, Grabowski and Landes, 1998). The features used included POS tags, dependency relation of the verb and noun, WordNet features, and OntoNotes (Pradhan et al., 2007) word sense features for word sense disambiguation (WSD). Their model obtained F1-scores of 88.90% when evaluated on the British National Corpus (BNC) LVC data (Tu and Roth, 2011), 64.20% on the OntoNotes 4.99 (Pradhan et al., 2007), and 80.68% on the OntoNotes Gold Standard. The ablation study that they carried out showed that both the WordNet and OntoNotes word sense features enabled better performance to be achieved. A method based on Conditional Random Fields (CRF) that made use of linguistic (contextual) features was proposed by Vincze, Nagy T. and Zsibrita (2013) to identify English and Hungarian LVCs in various domains (i.e., legal domain and news). Their results showed that their method outperformed the baseline methods (rule-based method and dictionary labelling) for both languages. For the English corpora, they obtained F1-scores of 64.09% and 59.41% for the English legal and news domains respectively. For the Hungarian corpora, they obtained F1-scores of 78.97% and 53.51% for the legal and news domains respectively. These results indicate that LVCs are domain specific. They also experimented with a simple domain adaptation approach which involved incrementally extending the training dataset with sentences from the target domain and their results showed that the gap between domains could be reduced by using this simple method. In addition, they carried out an ablation analysis and found that morphological features were the most useful among all the features used in the study. They also noted that syntactic information contributed lesser to the detection of LVCs in Hungarian compared to English and attributed this to the fact that Hungarian is a morphologically rich language. This implies that syntactic features (e.g., POS tagging) contributes more to languages with poor morphology such as English. Nagy T., Rácz and Vincze (2020) considered each verb+(preposition)+(article)+noun combination (the component in parentheses being optional components) as an LVC. Their proposed machine learning-based method used a decision-tree classifier for English, which was then adapted to German, Spanish and Hungarian using the 4FX parallel corpus (Rácz, Nagy T. and Vincze, 2014). Their language-independent feature set comprised of statistical (candidate LVCs and their frequencies), lexical (the most common verbs used in LVCs), morphological (whether the nominal component was derived from a verbal stem/was exactly the same as the verbal form), syntactic (the dependency labels for the nominal and verbal components of the candidate LVCs, and whether the nominal component of the candidate LVC was preceded by a determiner), and orthographic (whether the lemma of the nominal component ends in a given character bigram or trigram, and the number of constituent words of the candidate LVC) features. In addition, they included various language-specific features which included the auxiliary verbs (such as do, have), gender, agglutinative morphology, and so on. They obtained an average F1-score of 65.65% and found that the performance of their method depended on the quality of the dependency parsers used for LVC candidate extraction. The requirement for a reliable dependency parser would undoubtedly be a challenge for resource-poor languages. There has also been work focused on refining the annotation methodology for verbal MWEs (which include LVCs as a category) as well as the creation of annotated corpora. Ramisch et al., (2018) introduced a unified decision tree for joint verbal MWE identification and classification (PARSEME Annotation Shared Task edition 1.1) which is an enhanced 156 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 annotation methodology of the two-stage annotation process used in the PARSEME Shared Task edition 1.0 (Savary et al., 2017). The enhanced annotation guidelines will enable clearer distinctions to be made among the various verbal MWE categories, thereby making the annotation process more reliable and enabling higher-quality annotated corpora to be developed. This would then facilitate the development of rule-based and machine learning- based methods for LVC identification. The latest PARSEME edition 1.2 (Ramisch et al., 2020) focused on the identification of unseen verbal MWEs, i.e. verbal MWEs that were not found in the training set. A summary of the language coverage among the manually annotated corpus produced in the three PARSEME shared tasks is shown in Table 2. These annotated corpora will facilitate research and development that uses supervised learning approaches for LVC identification. Table 2. Language coverage of manually annotated corpora from PARSEME. Edition 1.0 Edition 1.1 Edition 1.2 Turkish (TR) Arabic (AR)a Basque (EU) Bulgarian (BG) Basque (EU) a French (FR) Czech (CS) Bulgarian (BG) Greek (EL) Farsi (FA) Croatian (HR) a Hebrew (HE) French (FR) English (EN) a Hindi (HI) German (DE) Farsi (FA) Italian (IT) Greek (EL) French (FR) Polish (PL) Hebrew (HE) Greek (EL) Portuguese (PT) Hungarian (HU) Hebrew (HE) Romanian (RO) Italian (IT) Hindi (HI) a Turkish (TR) Lithuanian (LT) Hungarian (HU) Chinese (ZH) a Maltese (MT) Italian (IT) German (DE) Polish (PL) Lithuanian (LT) Irish (GA) a Portuguese (PT) Polish (PL) Swedish (SV) b Romanian (RO) Portuguese (PT) Slovene (SL) Romanian (RO) Spanish (ES) Slovene (SL) Swedish (SV) Spanish (ES) Turkish (TR) German (DE) a Additional languages compared to the previous editions. b Corpora was substantially increased with respect to editions 1.0 and 1.1. 3.3 Corpus-Based Statistical Measures Askarian, Fazly and Hamzeh (2012) investigated the effectiveness of two groups of statistical measures in identifying Persian LVCs: association measures and fixedness measures. Association measures refer to statistical measures that provide an indication of the collocation behaviour of two words. In their study, the association measures used were Dice, Pointwise Mutual Inclusion (PMI) and Gravity (Gries, 2009). Their fixedness measures incorporated linguistic information that reflected the degree to which the LVC components were fixed. This was computed by obtaining frequency counts of nouns appearing in specified positions relative to the verbs. These frequencies were then used to compute the linear-combination fixedness measure and entropy-based fixedness measure. They evaluated the performance of these various statistical measures on a Persian POS-tagged corpus. Their results showed that the Gravity association measure and linear-combination fixedness measure achieved the best results. 157 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 4.0 DISCUSSION Several research gaps in LVC identification that should be addressed are listed below: Identification of all LVC types: There have been work focused on distinguishing among the different types of LVCs into true LVCs, idiom-like LVCs and productive-like LVCs by applying various tests such as passivization (e.g., conversion of the LVC made the decision into its passive form ‘the decision was made last week’), nominalization (decision-maker), wh-questioning (the noun in the LVC may be the answer to a question that starts with who/when/what), and others (Fazly, Stevenson and North, 2007). However, a comprehensive approach which includes all the different types of LVCs would actually be beneficial for downstream applications of computational linguistics (Nagy T., Rácz and Vincze, 2020) as this would enable all identified LVCs to be treated as a single semantic unit which would lead to more accurate processing of terms as well as feature set reduction. Methods for under-resourced languages: As LVCs are flexible, they may appear in a variety of forms due to morphosyntactic variations (e.g., the decision was made), the use of a complex noun phrase instead of a single noun in the nominal part of the LVC (e.g., adjective+noun such as he took a brisk walk and quantifier+adjective+noun as in she gave five interesting lectures), or may even be further discontinuous whereby the verbal and nominal components are separated by other words (e.g., the walk in the beautiful and dense rainforest that he had taken). This makes it particularly challenging especially because most existing work rely on the use of POS-tagging, dependency parsers, and annotated corpora. As indicated by Table 2, the languages for which annotated corpora for verbal MWEs are available is still very much limited. For under-resourced languages, this is a major obstacle due to the lack of such tools and resources. Therefore, there is a need to explore methods for LVC identification that do not need depend on the use of such tools or that require a smaller annotated corpus. Code-mixed LVCs: One growing challenge is the identification of code-mixed LVCs which are commonly used in social media posts and comments. Code-mixed text refers to text which includes a mixture of languages used, which is a common phenomenon in multilingual countries and is also increasingly more common in recent times as more people are able to speak in more than one language. There has been a growing interest in sentiment analysis of code-mixed text (Kaur and Mangat, 2017; Lo et al., 2017; Wang et al., 2017) and even in Malay-English code-mixed text (Abu Bakar et al., 2020; Tan, Lim and Lim, 2020). To our knowledge, there has not been work on code-mixed LVC identification and therefore, this needs to be addressed to enable LVCs occurring in text to be treated as a single semantic unit and avoid loss of contextual meaning that arise from the individual words of the LVC being considered as separate features. 5.0 CONCLUSION There has been an increasing interest in LVC identification as it plays an important role in downstream text processing tasks such as emotion analysis, machine translation, and question answer systems. LVCs are particularly challenging to identify due to their flexibility, complex structure, and discontinuity. This paper presented a review of methods that have been applied in LVC identification and a discussion on a number of research gaps. In addition, several promising future works were identified. 158 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 REFERENCES Abu Bakar, M. F. R. et al. (2020) ‘Sentiment Analysis of Noisy Malay Text: State of Art, Challenges and Future Work’, IEEE Access. IEEE, 8, pp. 24687–24696. doi: 10.1109/ACCESS.2020.2968955. Askarian, N., Fazly, A. and Hamzeh, A. (2012) ‘A comparison of statistical measures for the automatic identification of Persian light verb constructions’, in 16th CSI International Symposium on Artificial Intelligence and Signal Processing (AISP 2012). IEEE, pp. 479–483. doi: 10.1109/AISP.2012.6313795. Bonial, C. and Pollard, K. A. (2020) ‘Choosing an event description: What a PropBank study reveals about the contrast between light verb constructions and counterpart synthetic verbs’, Journal of Linguistics, 56(3), pp. 577–600. doi: 10.1017/S0022226720000109. Cap, F. et al. (2015) ‘How to Account for Idiomatic German Support Verb Constructions in Statistical Machine Translation’, in Proceedings of NAACL-HLT 2015, pp. 19–28. doi: 10.3115/v1/w15-0903. Chen, W. Te, Bonial, C. and Palmer, M. (2015) ‘English light verb construction identification using lexical knowledge’, Proceedings of the National Conference on Artificial Intelligence, 3, pp. 2375–2381. Constant, M. et al. (2017) ‘Multiword Expression Processing: A Survey’, Association for Computational Linguistics, 43(4), pp. 837–892. doi: 10.1162/COLI. Cordeiro, S. and Candito, M. (2019) ‘Syntax-based identification of light-verb constructions’, in The 22nd Nordic Conference on Computational Linguistics (NoDaLiDa 2019). Fazly, A., Stevenson, S. and North, R. (2007) ‘Automatically learning semantic knowledge about multiword predicates’, Language Resources and Evaluation, 41(1), pp. 61–89. doi: 10.1007/s10579-007-9017-9. Fellbaum, C., Grabowski, J. and Landes, S. (1998) ‘Performance and confidence in a semantic annotation task’, in Fellbaum, C. (ed.) WordNet: An Electronic Database. MIT Press. Gantar, P., Krek, S. and Kuzman, T. (2017) ‘Verbal multiword expressions in Slovene’, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10596 LNAI(March 2019), pp. 247– 259. doi: 10.1007/978-3-319-69805-2_18. Gilquin, G. (2019) ‘Light verb constructions in spoken L2 English An exploratory cross- sectional study’, International Journal of Learner Corpus Research, 5(2), pp. 181–206. doi: 10.1075/ijlcr.18003.gil. Gries, S. T. (2009) ‘Bigrams in registers, domains, and varieties: a bigram gravity approach to the homogeneity of corpora’, Corpus linguistics. Kaur, H. and Mangat, V. (2017) ‘Dictionary based Sentiment Analysis of Hinglish text’, International Journal of Advanced Research in Computer Science, 8(5), pp. 816–822. Available at: www.ijarcs.info. Lo, S. L. et al. (2017) ‘Multilingual sentiment analysis: from formal to informal and scarce resource languages’, Artificial Intelligence Review, 48(4), pp. 499–527. doi: 10.1007/s10462-016-9508-4. Mehl, S. (2019) ‘Light verb semantics in the International Corpus of English: onomasiological variation, identity evidence and degrees of lightness’, English Language and Linguistics, 23(1), pp. 55–80. doi: 10.1017/S1360674317000302. Nagy T., I., Rácz, A. and Vincze, V. (2020) ‘Detecting light verb constructions across languages’, Natural Language Engineering, 26(3), pp. 319–348. doi: 159 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 10.1017/S1351324919000330. Pradhan, S. S. et al. (2007) ‘OntoNotes: A unified relational semantic representation’, ICSC 2007 International Conference on Semantic Computing, pp. 517–524. doi: 10.1109/ICSC.2007.83. Rácz, A., Nagy T., I. and Vincze, V. (2014) ‘4FX: Light verb constructions in a multilingual parallel corpus’, Proceedings of the 9th International Conference on Language Resources and Evaluation, LREC 2014, (2010), pp. 710–715. Ramisch, C. et al. (2018) ‘Edition 1.1 of the PARSEME Shared Task on Automatic Identification of Verbal Multiword Expressions’, in Proceedings of the Joint Workshop on Linguistic Annotation, Multiword Expressions and Constructions (LAW-MWE-CxG 2018), pp. 222–240. Ramisch, C. et al. (2020) ‘Edition 1.2 of the PARSEME Shared Task on Semi-supervised Identification of Verbal Multiword Expressions’, Proceedings of the Joint Workshop on Multiword Expressions and Electronic Lexicons, pp. 107–118. Available at: https://aclanthology.org/2020.mwe-1.14. Savary, A. et al. (2017) ‘The PARSEME Shared Task on Automatic Identification of Verbal Multiword Expressions’, Proceedings of the 13th Workshop on Multiword Expressions (MWE 2017), (Mwe), pp. 31–47. Available at: http://multiword.sf.net/. Tan, K. S. N., Lim, T. M. and Lim, Y. M. (2020) ‘Emotion Analysis Using Self-Training on Malaysian Code-Mixed Twitter Data’, in Web Based Communities and Social Media 2020, pp. 181–188. Taslimipoor, S., Fazly, A. and Hamzeh, A. (2012) ‘A Class-based Acceptability Measure for Persian Light Verb Constructions’, in The 16th CSI International Symposium on Artificial Intelligence and Signal Processing (AISP 2012 ). IEEE, pp. 250–255. doi: 10.1109/AISP.2012.6313753. Tu, Y. and Roth, D. (2011) ‘Learning English Light Verb Constructions : Contextual or Statistical’, Computational Linguistics, 2011(June), pp. 31–39. Vincze, V., Nagy T., I. and Berend, G. (2011) ‘Detecting noun compounds and light verb constructions: a contrastive study’, in Proceedings of the Workshop on Multiword Expressions: from Parsing and Generation to the Real World (MWE 2011), pp. 116– 121. Vincze, V., Nagy T, I. and Zsibrita, J. (2013) ‘Learning to Detect English and Hungarian Light Verb Constructions’, ACM Transactions on Speech and Language Processing, 10(2). doi: 10.1145/2483691.2483695. Vincze, V., Zsibrita, J. and Nagy T., I. (2013) ‘Dependency Parsing for Identifying Hungarian Light Verb Constructions’, Proceedings of the Sixth International Joint Conference on Natural Language Processing, (October), pp. 207–215. Available at: http://aclweb.org/anthology/I13-1024. Wang, Z. et al. (2017) ‘Emotion Analysis in Code-Switching Text with Joint Factor Graph Model’, IEEE/ACM Transactions on Audio Speech and Language Processing, 25(3), pp. 469–480. doi: 10.1109/TASLP.2016.2637280. 160 of 225 ICDXA/2021/16 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 THE IMPACT OF SOCIAL MEDIA ON STUDENT’S ACADEMIC PERFORMANCE: A SURVEY ON TAR UC COMPUTING STUDENTS IN MALAYSIA DURING COVID-19 PANDEMIC Wai Ping Lim1, Jia Yin Loo1, Kylie Lee1, Hock Ming Pui1 and Tin Tin Ting2* 1 Faculty of Computing and Information Technology, Tunku Abdul Rahman University College, Kampus Utama, Jalan Genting Kelang, 53300, Wilayah Persekutuan Kuala Lumpur, Malaysia 2 Faculty of Information Technology, INTI International University, Negeri Sembilan, Malaysia *Corresponding author: [email protected] ABSTRACT Social media has become an inevitable tool in daily life for communication, entertainment, learning, and business. However, social media can cause addictive especially among university students. The aim of this study is to investigate if there is a relationship between social media usage and students’ academic performance focusing in factors: social media usage for academic purposes, time spent on social media and gender difference. A questionnaire was distributed in Google Form via social media platforms. PSPP is used to analyse the data collected utilizing one-way and two-way ANOVA together with Cohen’s f effect size. The findings show that there is a significant relationship between social media usage for academic purposes and students’ academic performance with large effect size (F=1.94, Sig=0.19, f=0.5). Time spent on social media is also significantly related to students’ academic performance with medium effect size (F=3.91, Sig=0.01, f=0.28). However, there is no significant relationship between gender’s social media usage and students’ academic performance (F=1.66, Sig=0.59). Keywords: Social Media, Academic Performance, Social Media Usage for Academic Purpose, Time Spent, Gender 1.0 INTRODUCTION Social media is becoming important in many lives especially the younger generation especially during COVID 19 pandemic when all the universities have transformed their educational method into online classes. The main attractions of Social Media that enticed many into the world of digital including boundless communications, unlimited information, and inexhaustible digital entertainments. According to NapoleonCat (2021) statistics, by August 2021, there are 29 millions of Facebook users (86.3% of the population) in Malaysia. People aged 18-24 is the second largest group of users with 22.1% with the increase of 6 million in 3 years from 2018 to 2021 (NapoleonCat, 2021). Due to this increased popularity, there are growing concerns over the possibility of the influences of social media on students’ academic performance. Thus, many researches were carried out to find out the factors that encourage social media usage (Raza et al., 2020; Teagen et al., 2020; Seounmi and Wonson, 2019; Yeunjae, 2020). Many researches also proves that there is a significant relationship between social media usage and academic performance (Bianca et al., 2021) with negative effects (Alam and Aktar, 2021; Ahmad et al., 2021; Beneyam, 2021) or positive effects 161 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 (Sandeep, 2018). However, there is limited research carried out in Malaysia on university students regarding the strength of social media effects on students’ academic performance. 2.0 LITERATURE REVIEW Social media is an application that allows users to interact with each other, create, edit and share new forms of textual, visual, and audio content, label and recommend existing forms of content. Social media, therefore, denotes the wide collection of internet-based and mobile services that connect people together to communicate, participate, interact, discussion and exchanges information on an online platform (Lemoine et al., 2016) The rapid growth of social media usage among tertiary students in academic usage has been positively associated with academic performance. As the usage of social media continues its constant growth, its application among tertiary students is inevitable (Boahene et al., 2019). There are many studies that reveal the positive effects and contribution of social media on the students’ academic performances (Bianca et al., 2021; Alam and Aktar, 2021; Ahmad et al., 2021; Beneyam, 2021; Lemoine et al., 2016; Al-Rahmi et al., 2015; Lambić, 2016; Lahiry et al., 2018). One of the researches done by Mushtaq and Benraghda (2018) found that most of the students use social media as informational and communicational tools for their education. Students are able to communicate effectively with each other, receive university- related information to improve their learning process. Other than that, Lemoine et al. (2016) examined how social media has impacted on students’ academic life and found that social media is widely used by students as a platform of discussions for their assignments and other course work. In short, most of the students support the idea that social media does contribute a significant quota to the development of their academic life. Additionally, Al-Rahmi et al. (2015) concluded that social media plays a very important role of input to students’ academic performance while collaborative learning is the mediator. They also hope that educational institutions will develop the use of social media in order to improve their students’ academic performance. A significant connection between social media usage for academic purposes and the students’ academic performance has been established in many studies. Boahene et al. (2019) found social media can be used effectively for academic purposes as an innovative tool to boost students’ CGPA. This is because the tertiary students who take part in this study use WhatsApp groups which aids them in interacting with peers. The increased interaction has a positive impact on their CGPA and gives students an alternate platform aside from the classroom. Lambić (2016) stated that there is a positive relationship between the frequency of use of Facebook for academic purposes and the academic performance of students during the course “Designing Educational Software and Media”. Lahiry et al. (2018) discovered 88.58% of students were using social media for academic purposes and suggested that it has a positive impact on their academic performance among tertiary medical, paramedical and nursing students in East India. All the findings of these research papers suggested that social media usage for academic purposes has a positive impact on the students’ academic performance. Jamil et al. (2020) carried out a study regarding the relationship between time spent on social media and students’ academic performance at Samuel Adegboyega University. From this study, if the student uses social media for academic purposes such as having a discussion forum for classwork this will positively impact their academic performance. Nevertheless, if the student spends too much time on social networking through the social media platform which is not related to academic pursuit will have a negative impact on their academic performance. Other than that, Young (2017) uses a series of interviews and questionnaires on each student for a week-long period, 24 hours each day. They found that over 83.33% of participants find social networking sites important and spend nearly 116 minutes on social 162 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 media every day. Young (2017) summarized that time spent on social media does bring adverse effects to academic performance. In addition, Alomari (2019) sought to probe the time spent using social media platforms by 971 students at a mid-south university. Findings discovered that the majority of the study sample usually spends between one to four, with an approximate average of three and a half hours on social media platforms hours (mean of hours spent in social media Academic Performance Alomari, Q1. What is your CGPA range (Below 2.0, 2.0 - 2.49, 2.5 - 2.99, 3.0 - 3.49, 3.5 - 2019 3.99, 4.0)? Social Media Usage for Academic Purpose Please express your frequency with the following statement on a scale of 1-5 Boateng and Q3. How often do you use social media to find information for academic purpose? Amankwaa, Q4. Do you agree that social media is important for academic purposes? 2019; Q5. Do you agree that social media contributes to academic achievement? Alomari, Q6. Do you agree that social media is used to discuss academic matters such as 2020; assignment with your university peers or professors? Q7. Do you agree that social media is used to exchange knowledge with your peers? Q8. Do you agree that social media is used to seek knowledge about specific academics? Time Spent on Social Media Q9. How many hours do you spend on social media per day (Below 1 hour, 1 hour Alomari, – 3 hours, 3 hours – 9 hours, more than 10 hours)? 2019 Figure 1. Questionnaire Items based on Alomari (2020) and Boateng and Amankwaa (2019) studies to test hypotheses. = 3.69). This study also used Pearson’s Correlation Coefficient to prove that there is a significant negative correlation between the time spent on social media and students’ academic performance regardless of the purpose of usage (r = -0.144, p = 0.000). Generally, it is common to see several researches regarding demographics such as age and gender’s social media usage on academic performances recently. For example, one of the researchers Alnjadat et al. (2019) had carried out research regarding gender’s social media usage on academic performance. Based on the result obtained by the research, it showed that the gender group who are more addicted to social media is men compared with women. Ali et al. (2021) carried out research on gender discrepancies in social media usage and found that there is a distinction of usage between the gender in which males use social media mainly for communication while females for academic purposes. This study has also revealed that there is a positive correlation between the academic performance of students and the gender’s social media usage. Besides that, the study from Wickramaratne et al. (2019) proved in the Chi-Square test that different gender shows different significant effects of Facebook in their academic performance. Not only that, the research from Shen (2019) discovered that girls use social media applications lesser than boys. In the light of the previous studies, the following hypotheses are constructed for this study: H1: There is a relationship between social media usage for academic purposes with the students’ academic performance. H2: There is a relationship between time spent on social media with the students’ academic performance. H3: There is a relationship between different gender’s social media usage for academic purposed with the students’ academic performance. 163 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 3.0 RESEARCH METHODOLOGY To investigate the relationship between the impact of social media and students’ academic performance, a questionnaire has been designed using Google Form to collect data as shown in Figure 1. These questionnaire items are extracted from the latest related studies by Alomari (2020), and Boateng and Amankwaa (2019). The questionnaire is sent to TAR UC students through WhatsApp groups of second year students in the Faculty of Computing and Information Technology. 152 samples are collected based on the student’s availability and PSPP is used in the data analysis. One-way ANOVA is used to analyze H1, H2 while two- way ANOVA for H3 together with its Cohen’s f effect size. 4.0 RESULTS AND DISCUSSION A total of 152 samples are collected from the students in the Faculty of Computing and Information Technology, TAR UC with 86 are male and 66 are female. Based on Table 1, Cronbach’s Alpha result shows that the items in questionnaire is valid and reliable in this research and therefore it can be used in the data analysis. Table 1. Cronbach’s Alpha of the items in questionnaire Cronbach’s Alpha .71 N of Items 9 Table 2. One-Way and Two-Way ANOVA analysis. F Sig. Effect Size, f Social Media Usage for Academic Purpose 1.94 0.19 0.50 Time Spent on Social Media 3.91 0.01 0.28 Gender x Social Media Usage for Academic Purpose 1.66 0.59 0.40 Dependant Variable: CGPA f = 0.10 = small effect; f = 0.25 = medium effect; f = 0.40 = large effect (Cohen, 1988) The result in Table 2 shows that there is a significant relationship between Social Medial Usage for Academic Purpose and students’ CGPA (F=1.94, sig=0.19) with f=0.50, large effect. Therefore, H1 is supported. This is the same with the previous studies in which the previous studies also showed that there is a significant relationship between Social Media usage for Academic Purpose and students’ CGPA (Bianca et al., 2021; Alam and Aktar, 2021; Ahmad et al., 2021; Beneyam, 2021; Lemoine et al., 2016; Al-Rahmi et al., 2015; Lambić, 2016; Lahiry et al., 2018). Most of the students agreed that social media is useful in assisting their academic activities (Mean=3.87). Therefore, although Social Media could affect one’s academic performance, it also depends on whether the effect is a positive or negative. This study clearly shown that if a student use Social Media for the academic purpose, there will be a positive effect on their academic performance. Besides that, Time Spent on Social Media also showed a significant relationship with CGPA with medium effect (F=3.91, sig=0.01, f=0.28). This is corresponding with the previous studies (Jamil et al., 2020; Young, 2017; Alomari, 2019). The average time spent in social media is about 3 hours per day (Mean=2.74). Therefore, H2 is supported. However, it should be further research into the total time spent in Social Media for academic purpose in order to prove that the time spent could affect their academic performance. One might spend most of the time in Social Media in accomplishing the assignment and definitely this will improve their academic 164 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 performance since there are a lot of information available and also ease the communication between team members. Two-way ANOVA analysis result for Gender effects on Social Media Usage for Academic Purpose and its relationship with CGPA showed that there is no significant difference between male and female (F=1.66, sig=0.59) with large effect (f=0.40). Therefore, H3 is rejected. Although this result does not correspond to most of the previous studies, it is interesting to further explore the reasons behind the insignificant difference between the two groups. Since this research is carried out on students from the faculty of information technology, it shown a phenomenon that both male and female of computing students are good in utilizing the social media technology in academic activities. This is supported by the time spent in social media by male and female are almost the same (male’s Mean=2.71; female’s Mean=2.77). In short, there is no difference of Social Media usage and time spent between different gender among computing students. Besides that, both gender shows the same respond in social media usage in academic purpose too (male’s Mean=3.87; female’s Mean=3.88). Additional analysis is carried out to further analyze on the gender’s difference in the academic performance. The mean for male’s CGPA is 4.37 while female is 4.64 which means the CGPA for both groups are between 3.0-3.99. Nevertheless, Table 3 shows the significant relationship between gender and academic performance but there is no significant difference between social media usage effects on CGPA among male and female. Table 3. One-Way ANOVA analysis. Gender F Sig. Dependant Variable: CGPA 4.58 0.34 5.0 CONCLUSIONS This study found that based on the survey conducted among TAR UC second year computing students, most of the students agree that social media is helpful and useful in their academic activities. Most of the respondents use social media as a platform to discuss their studies among friends and exchange knowledge to improve their academic performance. Students spent about 3 hours per day in social media daily. Other than that, they also agree that social media can help them to find information regarding their studies and assignments. Therefore, these two predictors are proven to be affecting students’ academic performance. Surprisingly, this study found that there is no relationship between different gender’s social media usage in academic purpose and students’ academic performance though gender does affect the CGPA. To better understand the result, it is recommended to further the study in analysing hours spent in academic versus non-academic purpose in order to gain further insight and more accurate relationship between social media and academic performance. This study could be extended to include other factors such as non-academic purposes usage, demographic statistics, psychological factors in relationship with students’ academic performance. REFERENCES Ahmad WMWP, Muhammad SM and Syazwan MZ (2021) Implications of social media addiction on academic performance among generation Z student-athletes during COVID-19 Lockdown. International Journal of Learning, Teaching and Educational Research 20(8), https://doi.org/10.26803/ijlter.20.8.12. 165 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Alam MS and Aktar H (2021) The effect of social media on student academic performance: a case study at the Islamic University of Bangladesh. International Journal on Transformations of Media, Journalism & Mass Communication 6(1). Ali S, Qamar A, Habes M and Al Adwan MN (2021) Gender discrepancies concerning social media usage and its influences on students academic performance. Utopía y Praxis Latinoamericana: Revista Internacional de Filosofía Iberoamericana y Teoría Social 26(1): 321–333, https://doi.org/https://doi.org/10.5281/zenodo.4556283. Alnjadat R, Hmaidi MM, Samha TE, Kilani MM and Hasswan AM (2019) Gender variations in social media usage and academic performance among the students of University of Sharjah. Journal of Taibah University Medical Sciences 14(4): 390–394, https://doi.org/ 10.1016/j.jtumed.2019.05.002. Alomari AA (2019) The impact of social media use on students’ academic performance: a field study at a mid-south university. Arkansas State University. ProQuest Dissertations Publishing. Al-Rahmi WM, Othman MS, Yusof LM and Musa MA (2015) Using social media as a tool for improving academic performance through collaborative learning in Malaysian higher education. Review of European Studies 7(3): 265–275, https://doi.org/10.5539/res. v7n3p265. Beneyam LY (2021) Social media usage, psychosocial wellbeing and academic performance. Community Health Equity Research & Policy, https://doi.org/10.1177/0272684X211 033482. Bianca AB, Katharine SA, Blaine LB, Meagan CAC (2021) The effects of social media usage on attention, motivation, and academic performance. Active Learning in Higher Education 22(1): 11-22, https://doi.org/10.1177/1469787418782817. Boahene KO, Fang J and Sampong F (2019) Social media usage and tertiary students’ academic performance: examining the influences of academic self-efficacy and innovation characteristics. Sustainability: 11(8): 1–18, https://doi.org/10.3390/su11082431. Cohen J (1988) Statistical power analysis for the social sciences. Hillsdale, New Jersey, Lawrence Erlbaum Associates. Jamil M, Ain Q, Batool S, Saadat S, Malik S, Arshad M, Nagra RN, Haider M, Shameem R and Latif B (2020) Impact of social media on academic performance. European Journal of Medical and Health Sciences 2(5): 1454–1462, https://doi.org/10.24018/ejmed. 2020.2.5.512. Lambić D (2016) Correlation between Facebook use for educational purposes and academic performance of students. Computers in Human Behavior 61: 313–320, https://doi.org/ 10.1016/j.chb.2016.03.052. Lemoine PA, Hackett PT and Richardson MD (2016) The impact of social media on instruction in higher education. Handbook of Research on Mobile Devices and Applications in Higher Education Settings February: 373–401, https://doi.org/10.4018/978-1-5225-0256-2. ch016. Mushtaq AJ and Benraghda A (2018) The effects of social media on the undergraduate students’ academic performances. Library Philosophy and Practice. NapoleonCat (2021) Facebook users in Malaysia [online] Available at: https://napoleoncat. com/stats/facebook-users-in-malaysia/2021/08. Raza SA, Qazi W, Nida S, Muhammad AQ, Shahzad Q and Ramsha A (2020) Drivers of intensive Facebook usage among university students: an implications of U&G and TPB 166 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 theories. Technology in Society 62(January): 101331, https://doi:10.1016/j.techsoc. 2020.101331. Sandeep L, Shouvik C and Suparna CAH (2018) Impact of social media on academic performance and interpersonal relation: a cross‑sectional study among students at a tertiary medical center in East India. January: 1–6, https://doi.org/10.4103/jehp.jehp. Seounmi Y and Wonson S (2019) Teens’ responses to Facebook newsfeed advertising: The effects of cognitive appraisal and social influence on privacy concerns and coping strategies. Telematics and Informatics 38(May): 30-45, https://doi.org/10.1016/j.tele. 2019.02.001. Shen J (2019) Social-media use and academic performance among undergraduates in biology. Biochemistry and Molecular Biology Education 47(6): 615–619, https://doi.org/10.1002/ bmb.21293. Teagen NG, Christy MKC and Jason BT (2020) Inside out and outside in: How the COVID- 19 pandemic affects self-disclosure on social media. International Journal of Information Management 55(December): 102188, https:// doi.org/10.1016/j.ijinfomgt.2020.102188. Wickramaratne PDVC, Abu Bakar DSHS and Phuoc PJC (2019) Gender as a moderator and its moderating effect on relationship between Facebook usage and the academic performance of government university undergraduates in Sri Lanka. Global Journal of Management and Business Research December: 15–20, https://doi.org/10.34257/gjmbravol19is15 pg15. Yeunjae L (2020) Motivations of employees’ communicative behaviors on social media: Individual, interpersonal, and organizational factors. Internet Research 30(3). Young RE (2017) An analysis discussing the impact of time spent on social media by current students through the operation of electronic time management. Rhodri Ellis Young Cardiff Metropolitan University. 167 of 225 ICDXA/2021/17 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 DIGITAL MUSIC: A STUDY OF FACTORS IN INFLUENCING ONLINE MUSIC STREAMING SERVICE PURCHASE Jian Yong Chai1*, Lee Kit Khen Ken1, Kah Him Chan1, Shao Xuan Wan1 and Tin Tin Ting2* 1 Faculty of Computing and Information Technology, Tunku Abdul Rahman University College, Kampus Utama, Jalan Genting Kelang, 53300, Wilayah Persekutuan Kuala Lumpur, Malaysia 2Faculty of Information Technology, INTI International University, Negeri Sembilan, Malaysia *Corresponding author: [email protected] ABSTRACT Digitalization have caused the online subscription service industry to be one of the fastest growing industries in recent years especially the online music streaming service. To cope with the changes of the market demand transitioning from physical to digital, many companies from the music industry have been focusing on the digitalization of music streaming service as it is the most profitable subscription-based service. Therefore, it is crucial to examine the factors that may influence the adoption of online music streaming service, the Purchase Intention in the fee-based subscription. The objective of this study is to identify the factors that will affect the willingness of the consumers to subscribe to online music streaming service (Purchase Intention). Online questionnaire is conducted and the result is analysed by Pearson Correlation and Cohen’s f2, effect size. The findings show that Perceived Value (r=0.513, f2=0.36), Tangibility Preference (r=-0.349, f2=0.14), Music Affinity (r=0.229, f2=0.06) and Music Piracy Awareness (r=-0.380, f2=0.17) are the factors that will significantly influence the consumers’ Purchase Intention in online music streaming service. Keywords: Online Music Streaming Service, Perceived Value, Tangibility Preference, Music Affinity, Online Music Piracy 1.0 INTRODUCTION The increasing technologies in the digital world introduce new business model among industries - subscription-based service. A subscription-based service is an online business model of selling access to content - ‘Content as a Service’ (CaaS). Specifically, the music segment of streaming services, or Online Music Streaming Services (OMSS), is of interest due to the high revenue in the music industry (Mae, 2019). Previously, people could only access digital music by purchasing or downloading music from music e-store (e.g., Apple’s iTunes). After the introduction of platforms such as Spotify and Pandora Radio, people started to change the way they listen to music from downloading files to “stream” songs through these platforms (Grannell, 2018). The technology development creates new business opportunities and rewarding revenue through subscription- based music streaming services. OMSS charges consumers a monthly fee for the unlimited access to music content. Spotify, Apple Music & YouTube Music are the several largest global OMSS subscription service providers (Mulligan, 2018). 168 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Although OMSS are common nowadays, there are still some questions about why people are willing to pay to subscribe for these online streaming services while others do not. Hence, this research is carried out to investigate the reasons why Malaysian people are willing to spend money on OMSS. 2.0 LITERATURE REVIEW Technology Acceptance Model (TAM) has been used by many researchers in studying consumers’ acceptance in certain technology (Davis et al., 1989). TAM suggested that perceived usefulness and perceived ease of use are two main determinants in people using computer. As technology is advancing and consumers are familiar with the technologies, therefore, perceived ease of use as determinant might not be as important as perceived usefulness as many OMSS providers are familiar with good system design. Thus, perceived usefulness or perceived value is the focus of this study. Perceived Value (PV) in this study refers to the net benefit from the OMSS after the cost trade-off (Lee & Lin, 2019). Some studies have shown that perceived value is a crucial reason that is affecting the consumer’s behaviour in their willingness to subscribe to OMSS (Purchase Intention) (Suki, 2011; Barros, 2017; João, 2019; Nik et al., 2020, Homnien and Pupat, 2020). Suki (2011) in his research found that women rely stronger on PV than men in Purchase Intention. Through investigating 924 music listeners in Portuguese, Barros (2017) found that PV has a positive influence in willingness to subscribe. From the findings of Teresa and João (2019) towards 318 people in Portuguese further suggested that PV of the OMSS is one of the main reasons that consumers are willing to subscribe to OMSS. The consumer-perceived value of downloadable (free) music, in terms of expected value, was found to be quite low as the Digital Rights Management (DRM) restrictions decrease value by making it difficult for consumers to use the product freely. The value of the online music subscription service could be increased by improving the most important benefits such as ease of use and search, a large music catalogue, and good sound quality, flexibility in use, exclusive contents, in addition to the basic functionality. PV could also be increased by decreasing privacy risk such as concerns of paying with credit card online, and most importantly is the reduction of price (Suki, 2011). Therefore, it is expected that OMSS with higher PV will attract higher subscription (Purchase Intention). Other than perceived value, there are some other external factors that could affect the purchase intention of a consumer according to many researches. In the music industry, Dörr’s research (2013) suggested reduced cost of search and prevention of moral scruples (conscience) as the main influence in OMSS subscription. Koh et al. (2014) in their research defined Online Music Piracy as the illegal duplication and distribution of sound or music recordings. It includes peer-to-peer file sharing and recording music from the Internet platforms such as YouTube or Pandora. Consumers utilizes the conveniency of the Internet to get variety of resources and believed that these are free. Meanwhile, consumers tend to engage more in the online music piracy due to their low moral equity, low ethical orientation and belief and low awareness of the social costs for downloading unpaid music (Hoon et al., 2001) However, the data from Korea also shows that piracy of digital music has led to a decline in physical music sales but not so much towards online music in United States, Germany, Italy and Belgium (Husin et al., 2021). Barros (2017) found that the attitude towards piracy will negatively influence the user to pay for the OMSS. This also means that consumer with low attitude towards piracy (low awareness on piracy) will likely to adopt digital music. 169 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Tangibility refers to the product’s physical properties and the extent to which it can be seen, felt, heard, smelled, and etc (Dörr et al., 2013). Music, a piece of information, is completely intangible but becomes tangible through its medium. With the on-going digitalization of mediums (vinyl → CD → MP3 → Music as a Service (MaaS)), music is becoming increasingly intangible due to its loss of haptic attributes, which in turn influences consumers’ purchasing behaviour. By comparing reviews on digital and physical music, it could be shown that digital music is consumed in a less involved and therefore less focused way than physical music (Dörr et al., 2013). Therefore, music consumption is in a transformation stage and music consumers are purchasing their music in different types of formats, physical (tangible) or non-physical (intangible). Tangibility Preference (TP) is the consumer’s preference of physical formats of music over the non-physical formats of music. Even though MaaS is a more convenient way of consuming music rather than CDs and vinyl, some consumers prefer tangible solutions. These consumers often feel proud to display their physical record collections. Research of Mac (2019) indicated that Japanese consumers are anticipated to have a high preference for ownership and Tangibility Preference, which could cause a lower number of users to be interested in paying for access to intangible streaming subscriptions. Therefore, it would seem that an individual with a high Tangibility Preference will be less inclined to subscribe to OMSS. Music Affinity (MA) refers to the individual's liking for music and affection of music in individual life (Zhou, 2019). The revenue shows that although sales of albums are subject to pirated copies, the popularity of the album is not a factor of piracy because entertainment is not as important as in the past. Consumers who still prefer listening to music will have the willingness to pay more. When music plays an important role in life, the willingness to download will be stronger (Zhou, 2019). From the research from Sinha and Mandel in 2008, music involvement significantly negatively influences the respondent’s willingness to pay for downloadable mp3. Another research shows that interest in music (Music Affinity) is considered a factor to influence the desire to use OMSS to access unlimited music services (Puspitasari et al., 2019; Zhou, 2019). Therefore, it is predicted that consumers with higher Music Affinity will be more willing to pay for OMSS. 3.0 CONCEPTUAL MODEL We choose a research model modified from Dörr’s research (2013) which will focus in four factors as shown in Figure 1. This model will extend TAM model and Dörr’s with additional important factor (main concern of music industry nowadays) – Music Piracy Awareness. It is crucial to identify whether this factor (MPA) will affect the OMSS market and provide an insight to the industry in dealing with this issue. Perceived Value Tangibility Preference Music Affinity Music Piracy (PV) H(2TP) H(3MA) Awareness (MPA) H1 Purchase Intention (PI) H4 Figure 1. Research Conceptual Model. Our hypotheses are as following: H1: Perceived Value of OMSS is positively related to the Purchase Intention in OMSS. 170 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 H2: Tangibility Preference is negatively related to the Purchase Intention in OMSS. H3: Music Affinity is positively related to the Purchase Intention in OMSS. H4: Music Piracy Awareness is negatively related to the Purchase Intention in OMSS. 4.0 RESEARCH METHODOLOGY To test the hypotheses, data are collected by distributing the online questionnaire prepared in Google Form (adopted from Barros’ research (2017) (Appendix Figure 3). Firstly, the question to collect data on the willingness of respondents to pay for the OMSS is designed as a nominal question for subscription: 0 = no; 1 = yes. Multiple choice questions are designed for collecting data of which OMSS the respondents subscribe to. It also has a blank option for the respondents to fill in if their option is not available or not subscribed to any OMSS. The data are then compiled and summarized. The next section is related to the hypotheses to collect the opinions on the OMSS from the view of Perceived Value, Tangibility Reference, Music Affinity and Music Piracy Awareness. The questions are designed using an ordinal scale which is 5-point Likert Scale, from 1: strongly disagree to 5: strongly agree. Data collection is carried out from 25 July to 19 September, 2021. PSPP is used in this research to analyse the data collected utilizing Pearson Coefficient and Cohen’s f, effect size of the relationships. Cronbach Alpha tests the validity of the questionnaire items. Path analysis is used to test the research model. In addition, effect size f2 is also calculated to show the strength of the relationship between variables (Dörr et al., 2013). 5.0 RESULTS AND DISCUSSION This study has accumulated data from a total 200 respondents who are all Malaysian. Most respondents are aged between 18 - 25 years old. The respondents are collected using the data sampling plan which is the Simply Random Sampling method. The respondents are randomly selected by sending the questionnaire via Google Form to the respondent through email or sharing in between the respondents. Based on Table 1, Cronbach’s Alpha result shows that the items in questionnaire is valid and reliable. Table 1. Cronbach’s Alpha of questionnaire items Cronbach’s Alpha .71 N of Items 15 Perceived Value Tangibility Preference Music Affinity Music Piracy (PV) (TP) (MA) (MP) -.349**(.14) .229** (.06) -.380** (.17) .513**(.36) Purchase Intention (PI) Values: Pearson Coefficient (Effect Size, f2) f2 ≥ .02 = small effect; f2 ≥ .15 = medium effect; f2 ≥ .35 = large effect (Cohen, 1988) ** sig < 0.05 Figure 2. Results of data analysis. 171 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Based on Figure 2, Perceived Value has a positive Pearson Correlation (0.513) with Sig = .000 which shows a positive relationship between PV and PI. This shows that consumers that perceived OMSS with high value will be willing to pay for the subscription. This is also supported by the statistic that 60.5% of the respondents agree that PV will affect willingness to subscribe OMSS with Mean of 3.38 (Table 2). The effect size, f2 of the path PV → PI is 0.35 which shows a large effect. Therefore, H1 is supported. The higher the perceived value of the online music streaming service, the higher the consumers' willingness to pay for OMSS. Tangibility Preference in Figure 2 has a negative Pearson Correlation (-0.349) with Sig = .000 which shows a negative relationship between TP and PI. This proves that consumers with low Tangibility Preference (or prefer intangible medium) is more willing to adopt OMSS. The effect size, f2 of the path TP → PI is 0.14 which shows a small effect. Therefore, H2 is supported. Consumer with less preference to the tangible music medium is more willing to pay for OMSS. Music Affinity in Figure 2 has a weak positive Pearson Correlation (0.229) with Sig = .001 which shows a positive relationship between MA and PI. This shows that consumers who like music might adopt OMSS. The effect size, f2 of the path MA → PI is 0.06 which shows a small effect. Therefore, H3 is supported. The higher Music Affinity will affect consumers to pay for OMSS. Most of the respondents have high music affinity with Mean 3.86 as shown in Table 2. Attitude towards Music Piracy shows a negative Pearson Correlation (-0.380, Figure 2) with Sig = 0.000 which indicates that consumers with low awareness regarding music piracy will likely to pay for OMSS. This effect size, f2 of the path MPA → PI is 0.17 which shows a medium effect. Therefore, H4 is supported. Consumer with low music piracy awareness will more likely to pay for OMSS. Table 2: Pearson Correlation and Mean. R Sig Mean PV 0.513 0.000 3.38 TP -0.349 0.000 3.13 MA 0.229 0.001 3.86 MPA -0.380 0.000 2.95 Independent variable: PV: Perceived Value of OMSS (Q2, Q3, Q4) TP: Tangibility Preference (Q5,Q6) MA: Music Affinity (Q7,Q8,Q9) MP: Music Piracy Awareness (Q10,Q11,Q12,Q13,Q14,Q15) Dependent variable: Consumer willingness to pay for the OMSS, PV 6.0 CONCLUSION AND FUTURE WORKS This research has some limitations. First and foremost, the research method that is used in this research is an online questionnaire. The benefit of the online questionnaire is that it could significantly reduce the time needed to collect data. As a downside, this method of data collection may only limit the questionnaire recipient to only Internet users and since most of the Internet users are students, their responses are confined into a single pattern which might causes biased results. Therefore, the findings of this study might not represent the entire population. The findings of this study can be further validated by including respondents of different age groups and ethnic to provide an even broader view of the study. 172 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 During this research, the main focus is the consumer Purchase Intention on Online Music Streaming Service (OMSS). However, the platform that is chosen to be used for the OMSS is not studied. Therefore, it would be better if further research can include the research on which platform is chosen by each respondent so that the weaknesses and strengths of each of the platforms is able to provide a clearer image on the Purchase Intention. Besides, this study does also not include every factor that might affect consumers’ purchase intention and only measures the declared intention. This is because it would be impracticable to include a large number of factors in one survey due to its increased complexity. Therefore, a different set of factors for predicting consumers’ Purchase Intention should be considered in the further research such as sense of fashion, economic status, age and gender. On the other hand, interesting research topics can be studied in the future such as research on the effect of advertising on the consumers’ decision and also factors of in-app purchases. REFERENCES Barros FF (2017) Online music streaming services: what motivates consumers' intention to go premium? A study of the factors that may influence the intention to adopt music as a service. University of Minho, Braga, Portugal. Cohen J (1988) Statistical power analysis for the social sciences. Hillsdale, New Jersey, Lawrence Erlbaum Associates. Davis FD, Bagozzi RP and Warshaw PR (1989) User acceptance of computer technology: a comparison of two theoretical models. Management science 35(8): 982–1003. Dörr J, Wagner T, Benlian A and Hess T (2013) Music as a Service as an Alternative to Music Piracy? Business & Information Systems Engineering 5(6): 383–396, http://dx.doi.org/ 10.1007/s12599-013-0294-0. Fernandes T and Guerra J (2019) Drivers and deterrents of music streaming services purchase intention. International Journal of Electronic Business 15(1): 21-42, http://dx.doi.org/ 10.1504/ijeb.2019.10020273. Grannell C (2018) A history of music streaming. Dynoaudio. Skanderborg, Denmark. Homniem C and Pupat N (2020) The factors influencing Thai passenger’s intention to reuse Grab car service in Bangkok. AU-GSB e-journal 13(1): 41-51. Hoon AS, Sim CP, Lim EAC and Kuan TS (2001) Spot the difference: consumer responses towards counterfeits. Journal of Consumer Marketing 18(3): 219–235, http://dx.doi.org/ 10.1108/07363760110392967. Husin N and Hidayanto AN (2018) Influence factors in the fall of music industry and the emergence of new digital music service player in the digital music transformation - a systematic literature review. Proceedings of 2018 3rd International Conference on Information Technology, Information System and Electrical Engineering, http://dx.doi.org/10.1109/icitisee.2018.8720993. Koh B, Murthi BPS and Raghunathan S (2014) Shifting demand: online music piracy, physical music sales, and digital music sales. Journal of Organizational Computing and Electronic Commerce 24(4): 366–387, http://dx.doi.org/10.1080/10919392.2014. 956592. Lee CW and Lin TC (2019) Purchase intention in subscription-based online music service from the perspective of push-pull-mooring model. Journal of Advanced Engineering 14(1): 45-53. Mae SA (2019) Factors influencing Japanese consumers’ purchase intention of subscription streaming services. Ritsumeikan Asia Pacific University. Beppu, Japan. Mulligan M (2018) Mid-year 2018 streaming market shares. Midia. 173 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Nik AANH, Wan FWZ, Roslizawati CA, Nor MMN, Zaimatul A, Nur HM, Abdullah MY (2020) Grab pay app: the factors influencing tourists’ behavioural intention-to-use. Talent Development & Excellence 12(3s): 820–828. Puspitasari NB, Susanty A, and Prakoso MFA (2019) Analysis of consumer behaviour factors on subscription-based music services. E3S Web of Conferences 125(2019) 21003, https://doi.org/10.1051/e3sconf/201912521003. Sinha RK and Mandel N (2008) Preventing digital music piracy: the carrot or the stick? Journal of Marketing 72(1):1-15, http://dx.doi.org/10.1509/jmkg.72.1.001. Suki NM (2011) Gender, age, and education: do they really moderate online music acceptance? Communications of the IBIMA 2011, http://dx.doi.org/10.5171/2011.959384. Neeraj G and Veena B (2016) Application of decomposed theory of planned behavior for m- commerce adoption in India. Proceedings of the 18th International Conference on Enterprise Information Systems 2: 357-367. Zhou Y (2019) Research on the factors affecting the willingness to pay for digital music. Journal of The Korea Society of Computer and Information 24(6): 81-88, http://dx.doi.org/10.9708/jksci.2019.24.06.081. APPENDIX Purchase Intention Q1. Do you subscribe to any online music subscription service (e.g. Apple Music, Spotify, Youtube music ,etc)? Perceive Value Please express your agreement with the following statements about the value of paid music streaming services on the scale of 1-5. Q2. Paid music streaming services have an excellent level of quality. Q3. The use of paid services of music streaming make me feel good Q4. Paid music streaming services worth the price they cost Tangibility Preference Please indicate your opinion regarding the following statements about Tangibility Preference of music on a scale of 1-5 Q5. It is important to me to have music in a physical format Q6. I feel that music in physical format is more “real” or genuine than digital music Music Affinity Please express your agreement with the following statements about your affinity with music on a scale of 1-5 Q7. I have a strong interest in music Q8. I value music as an important part of my lifestyle. Q9. Listening to music is one of the most important things I do day to day. Music Piracy Awareness Please indicate your opinion regarding the following statements about unpaid download of music over the internet on a scale of 1-5. Q10. Record companies protect their copyright just to exploit the consumers. Q11. Record labels make profits excessively loud 174 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Q12. It's culturally accepted. Q13. There is not a high risk of being caught up. Q14. The consequences are not severe even if it is caught up. Q15. I think pirating music is an excellent idea. Figure 3. Questionnaire Items for each variable based on Barros’ research (2017). 175 of 225 ICDXA/2021/18 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 FORECASTING FACEBOOK USER ENGAGEMENT USING HYBRID PROPHET AND LONG SHORT-TERM MEMORY MODEL Kong Yih Hern1, Lim Khai Yin2 and Chin Wan Yoke1* 1 Faculty of Computing and Information Technology, Tunku Abdul Rahman University College, Kampus Utama, Jalan Genting Kelang, 53300, Wilayah Persekutuan Kuala Lumpur, Malaysia 2 Department of Computing and Information Technology, Tunku Abdul Rahman University College, Penang Branch Campus, Malaysia *Corresponding author: [email protected] ABSTRACT Business forecasting remains a popular topic these days. A reliable business forecast often plays a vital part in an advertising campaign. The amount of attention acquired by posting an advertisement is one of the most essential criteria in determining the efficacy of the advertisement. The number of times that public users engage with a content signifies the amount of attention received, was measured by user engagement. With a good forecast, the advertisement could be promoted to a larger number of people. Facebook, as the most popular social media site, is favoured by majority of the advertisers. Therefore, this study addresses Facebook user engagement by forecasting the optimum date to post an advertisement. Different forecasting models, each with its own strengths and weaknesses, are used to model time series data with various properties. The objective of this study is twofold: to investigate the accuracy of the proposed Hybrid Prophet-LSTM that combines Long Short- Term Memory (LSTM) and FBProphet (Prophet) and to study the holiday impact on user engagement forecasting on Facebook brand pages. Data from 3 popular brand pages in the period of June 2018 to March 2019 was used in the experiments. The results show that the proposed hybrid model outperforms both the standalone LSTM and Prophet across the datasets. Besides, it is found that holiday effect could generally increase forecast accuracy. The optimum date for an advertisement campaign can therefore be determined based on the most forecasted user engagement, which consequently enhances the business income. Keywords: Time Series forecasting, Hybrid forecasting, Business forecasting, Prophet, LSTM, Holiday effect 1.0 INTRODUCTION User engagement refers to the attention, interactivity, perceived user control, and impression from the public users (Brien and Toms, 2008). Businesses are constantly seeking for innovative ways to improve the effectiveness of their advertisements. Adopting ineffective marketing techniques for a marketing campaign would not only squander corporate resources, but will also fail to get the desired results. The absence of user engagement with the advertisement platform was the most common reason for advertising campaigns underperforming (Goldsmith and Lafferty, 2002), (Frolova, 2014). The forecasted user engagement on a given advertisement plays an essential role in maximising the effect of an advertisement. According to the study by Frolova (2014), an effective advertisement can considerably raise volume of sales profits, foster consuming culture, fulfil customer wants for goods, and link advertiser and consumer audience in terms of communication channels. In 176 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 other words, businesses should promote at the best time possible to achieve the most responses or user engagement from the audience. Massive volumes of data from a large number of consumers are being collected through the media, particularly social media. A variety of studies by Schoen et al. (2013); Srinivasan et al. (2013); Breitenecker (2014); Kundi et al. (2014); Yasuko, Etuso and Akira (2014); Li et al. (2015); Di Gangi and Wasko (2016); Lee, Shia and Huh (2016); Debreceny (2019) have utilized social media data for various analysis. Researchers can study human behaviour patterns and predict user engagement using data from social media. In recent years, business forecasting has been a popular topic of study. The approach has been used to forecast time series data such as future stock movement (Sidi ,2020), traffic matrix (Azzouni and Pujolle, 2017), insurgency movement direction (Waeto, Chuarkham and Intarasit, 2017), and user engagement (Srinivasan et al. ,2013). With an accurate forecasting result, user engagement for an advertisement may be easily attained. Selecting the right forecasting model is, therefore, of utmost importance. A variety of forecasting methods are utilised by businesses today. In this study, forecasting experiments are conducted by using Facebook data. This research employs the proposed Hybrid Prophet- LSTM by Kong, Lim and Chin (2021) to forecast user engagement that would in turn assist businesses in making managerial decisions on the commencement of an advertising campaign In Section 2, applications of forecasting techniques in various fields are presented, showing how forecasting models are being used to solve various business problem. In Section 3, the details of dataset and proposed model are explained. The results of the evaluation can be found in Section 4. Section 5 discusses the findings and conclusion for this study 2.0 LITERATURE REVIEW 2.1 Forecasting in Businesses Different models have been employed to analyse and solve various business problems (Polat, 2007). It is important to determine which model to use for solving a business problem. Prediction and forecasting have recently been a popular topic. To overcome the network traffic problem, the study by Azzouni and Pujolle (2017) used forecasting model to predict network traffic matrix. Real-world data from the GEANT organisation network was used to test the feasibility of the forecasting model. The forecasting model is validated that could accurately predict traffic metrics. This forecast result is used to assist network operators in making decisions such as traffic accounting, short-time traffic scheduling, traffic rerouting, network design, long-term capacity planning, and network anomaly detection based on actual network traffic flows. The paper demonstrates how a forecasting model can be used to solve a business prediction problem by providing estimated future values that can be utilised to help making the decisions. Yenidogan et al. (2018) used forecasting to tackle the Bitcoin forecasting difficulty in a recent study. The dataset contains two years' amount of Bitcoin exchange rates against a variety of currencies. The author employed a forecasting model to project future Bitcoin values, which is a critical subject for profit-seeking investors. The Bitcoin values were considered successful for future 90-days forecast with a precision of 94.5%. A credible forecast of future Bitcoin values would be valuable information for investors profiting from their Bitcoin investments. Another study by Li et al. (2015) used forecasting to address a Twitter advertising problem. The click-through rate (CTR) on the Twitter timeline was forecasted using pointwise learning, pairwise learning, and further improvement version based on these two models. In the work, the authors proposed a model that used improvised algorithm based on pairwise and pointwise learning to learn user impressions with the click probability. The 177 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 forecast outcome will alter how Twitter displays advertisements to users, leading to a greater CTR from Twitter users. The outcome from the model is found to be a more successful approach than traditional computational advertising, which are sponsored search and contextual advertising. The author concluded that the proposed method could significantly enhance the users’ CTR on Twitter’s advertisements. The forecasting technique could also be used on social media data for a variety of purposes. Schoen et al. (2013) forecasted future events and developments using social media data. The events include area of politics, finance, entertainment, market demands, health, and other. The same study by Schoen et al. (2013) included influenza incidence, product sales, stock market movement, and electoral results as examples of forecasting applications using social media data. As a result, user engagement is forecasted in order to decide the optimal date to promote. A reliable forecast of user engagement could aid businesses in making strategic decisions about how to execute a successful advertisement campaign that reaches the greatest number of people. 3.0 PROPOSED METHODOLOGY 3.1 Time series data Facebook is the largest and the most favoured social media platform for public users (Bashar, Ahmad and Wasiq, 2012). To determine which variable is important to the research, the Facebook page and post metrics (Insight - Pages, 2021) were examined. Three brands in the categories of food, beverages, and cosmetics were chosen arbitrarily. The purpose of the following sections is to forecast the daily engagement received by a certain Facebook Page in order to reach the largest number of people possible. The target variable is the daily page engagement attribute, which was crawled from Facebook. Page engagement is a daily metric derived from user actions such as clicks, responses, comments, shares, and other forms of interaction with the page. Only one variable, customer page engagement, was examined and forecasted in this study. Three datasets from two distinct sectors were gathered. Two years of daily time series data, starting on June 1, 2018, and ending on March 31, 2021, were gathered as a dataset from the three specified pages. Malaysia Public Holiday has been included to Prophet's holiday component. The purpose of this holiday dataset is to investigate the impact of holiday effects on time series forecast results. To assess the influence of holiday effects, a comparison study is carried out. The number of times of users who engage with a certain page on a daily basis is referred to customer page engagement. As a result, this variable is a daily data variable with a daily count of user engagement. Users' clicks, reactions, shares, comments, and other actions are used to calculate user engagement. Overall, customer page engagement is a measurement of how much public users pay attention to a page. 3.2 Proposed model This study uses the proposed Hybrid Prophet-LSTM by Kong, Lim and Chin (2021) to enhance forecast accuracy. In the suggested hybrid methodology, Prophet is used as the linear model, while LSTM is used to address the residual nonlinear connection in the time series data. Prophet is used to model regular, irregular, and non-regular holiday events. Because time series data contains both linear and nonlinear structure, a nonlinear model, such as LSTM, is used to represent the residual matrix from a linear model. With its remarkable potential of addressing nonlinearity relationships in time series, LSTM is utilised to model nonlinear relationships in the residual matrix to produce better forecast results. To create the 178 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 forecast result, the time series data were first fitted into Prophet. The residual matrix was computed using Prophet's forecasted output, and the residual matrix was then fitted into LSTM. The forecasted residual is used to compute hybrid forecast output. Finally, the output produced by the hybrid model is evaluated using several performance metrics and compared to the results of various models. 4.0 EMPIRICAL RESULTS 4.1 Time series decomposition An analysis is performed to study the individual components in the time series data. The user engagement data was analysed using decomposition feature in Prophet. A decomposition result was generated to understand various characteristic in a time series such as its trend, holiday effects, and multiple forms of seasonality. (a) (b) (c) (d) (e) (f) Figure 2. Example of individual components from Prophet’s decomposition result, a) Data trend, b) Holiday effects, c) Weekly seasonality, d) Yearly seasonality, e) Monthly seasonality, f) Quarterly seasonality The decomposition result displayed in Figure 2 is created using Dataset 1. As shown in Figure 2(a), the engagement does have 2 growths in the historical data but then declined after few months. There is little evidence of continuous potential growth, and the trend remains in the range of 0 to 11,000 engagements. In Figure 2(b), the holiday effects graph demonstrates that only the holiday events occurring before year 2019 have a negative impact on engagements. Other holiday-related events have a beneficial impact on the time series data. When there is a holiday event, engagement decreases before the year 2019, but increases in various scales in the year 2019 and later. Each of these holiday occurrences had an influence that was discovered and used to change the predicted estimate. Prophet managed the irregular holiday effects during modelling by applying this analysis result. The weekly seasonality shown in Figure 2(c) demonstrates that user engagement to this page drops on Sunday, then gradually increases until Saturday, which is also the week's peak 179 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 engagement. This explained the user engagement behaviour over the course of a week, demonstrating that they are more willing to engage with the page as the weekend approaches. The yearly seasonality is shown in Figure 2(d), and we can see a uniform yearly seasonality in this decomposition result. However, from September 2020 to November 2020, the model captures a lot of noise. These factors will have an impact on the modelling outcome, which will be inaccurate for the specific period in the yearly seasonality. Monthly seasonality is depicted in Figure 2(e) as a wave pattern, with engagement becoming stable at 0% at the beginning of the month, then progressively increasing in total, regaining the lost value until the conclusion of the month, but extremely instable at the same time. Weekly seasonality has an impact on monthly seasonality. As a result, we can refer to Figure 2(c) for the weekly seasonality decomposition explanation. The quarterly seasonality, which is the form of seasonality for one quarter, is shown in Figure 2(f). The quarterly seasonality rises from - 100% at the start of each month to 150% in the 15th days of each month, then returning the value obtained earlier. Monthly and weekly seasonality have a significant impact on quarterly seasonality; see Figure 2(c) and Figure 2(e) for further information. 4.2 Standalone and Hybrid Prophet-LSTM algorithm The linear relationship in time series data was fitted using Prophet and the remaining pattern under the Prophet residual was fitted into LSTM. Five distinct methods are compared for creating a reliable forecast result and examining the impact of the holiday effect on the forecast outcome. Prophet, Prophet without holiday, LSTM, Hybrid Prophet-LSTM, and Hybrid Prophet-LSTM without holiday were among the approaches used. These methods are evaluated using different performance metrics including Weighted Mean Absolute Percentage Error (WMAPE), ������2 score, Root mean square error (RMSE), and Mean Absolute Deviation (MAD). Table 1. Performance metrics for Prophet, LSTM, and Hybrid models Dataset 3 Dataset 2 Dataset 1 WMAPE Prophet Prophet LSTM Hybrid Prophet- Hybrid (No Holiday) LSTM Prophet-LSTM ������2 47.8679% 17.7578% (No Holiday) 99.9339% 46.9547% 95.8725% 16.8768% RMSE 763.306 99.9401% 347.279 99.9946% 15.6264% MAD 504.749 744.791 182.326 243.283 99.9953% 21.1959% 495.586 6.1050% 173.859 WMAPE 99.9930% 21.4081% 94.7999% 5.8045% 227.217 5088.143 99.9928% 2283.617 99.9995% 161.094 ������2 4037.848 5163.085 1160.590 1381.347 5.5680% 55.3595% 4079.288 22.8360% 935.641 99.9996% RMSE 99.5889% 58.1690% 58.5297% 18.8923% MAD 328.844 99.4153% 783.955 99.9799% 1298.541 168.269 337.710 112.955 855.583 WMAPE 170.354 63.780 31.8675% 58.906 99.8619% ������2 283.801 RMSE 117.100 MAD Table 1 compares the results of the three models with different approaches. Models 1 and 2 are the standalone Prophet and LSTM, and Model 3 is the Hybrid Prophet-LSTM. The models were also compared with and without the holiday component. Prophet has a WMAPE of 47.86 %, 21.19 %, and 55.35 %, respectively using the three datasets. By comparing the error rates of the standalone Prophet and LSTM models, the results show that Prophet's error rate is at least double or more than LSTM's error rate. LSTM outperforms Prophet by producing fewer errors and a lower overall error rate. When LSTM is compared to the Hybrid 180 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 model, the hybrid model outperforms LSTM model in every aspect. By having reduced mistakes and error rates, as well as a higher ������2 value, the hybrid model outperforms the LSTM. The hybrid model is an alternative to the traditional model. Prophet model did not demonstrate good modelling in this case since Prophet's holiday component does not significantly improve Prophet's performance. Despite the fact that holiday effects were not adequately visible in Prophet's forecasting results, the holiday component has a significant impact on the hybrid model forecast result for Dataset 3. The holiday component has a minor influence on Datasets 1 and 2, but has a considerable impact on Dataset 3 with an error reduction of 18.89 % to 31.86 % without the holiday component. In overall, the holiday component could improve the Hybrid Prophet-LSTM model in producing a more accurate forecast. Table 2 shows that LSTM outperforms Prophet when it comes to modelling user engagement time series data. As a result, a hybrid model may model both linear and nonlinear time series data with a steady performance because it incorporates the strengths of both linear and nonlinear models. During the study, it was discovered that although the Prophet linear model can detect seasonality in Dataset 1 and 3, but the seasonality captured is unusual and does not show an observable pattern, which can considerably increase forecast error. When there is no observable seasonality pattern in the time series, the forecasting accuracy for these datasets is relatively low. The forecasting models demonstrate their feasibility modelling time series data to forecast user page engagement as a result of the findings. The hybrid model, which was shown to be the best, had a forecast error range of 5.80 % to 18.89 %. Businesses are able to forecast a page engagement by leveraging a good model, which can then be used to determine the optimum day to begin an advertising campaign. Posting an advertisement on a day with higher engagement indicates that the advertisement will reach a larger group of audience. 5.0 CONCLUSIONS Hybrid Prophet-LSTM was utilised in this work to combine linear and nonlinear models to produce improved forecast results. The proposed approach incorporates features such as a customizable calendar of events or holidays. In addition to irregular holiday occurrences, this study models seasonality and trend components. Experiments were carried out to verify the effectiveness of the proposed model. We can see that without the holiday components in the hybrid model, the models perform better in overall. The suggested model has the smallest forecasting errors and performs well across a wide range of datasets and scales of variance. As a result, it can be inferred that while attempting to produce an accurate forecast, there are two critical factors to consider. The linear model's compatibility would be the first concern. The performance of the linear model was found to have a significant impact on the hybrid forecast result. In the hybrid model, a well-performed output in the linear model would offer an exceptional outcome. The selection of features is the second factor. Only one variable was chosen for forecasting in this study, resulting in a univariate analysis. These studies by Hummel and Sligo (1971); Saccenti et al. (2014) implied that both multivariate and univariate approaches should be used because the results from these two analyses are complementary. Analysing the relationship between the dependent variable and other independent variables would be the future work for this study. It can be stated that the results of this study will be possible to forecast dates with the most user engagement. However, any managerial judgment should not be made solely on the basis of this variable. This study serves the purpose of exploring the expected advertisement effect. Businesses should instead validate the result with numerous approaches, such as referring to the expert knowledge and experiences in the domain, conducting experiment using other datasets, and compare the 181 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 result with other single-variable data analysis on advertising, as advertising investment is a complicated practice in the actual world (Dawes et al., 2018). REFERENCES Azzouni, A. and Pujolle, G. (2017) ‘A Long Short-Term Memory Recurrent Neural Network Framework for Network Traffic Matrix Prediction’. Available at: http://arxiv.org/abs/1705.05690. Bashar, A., Ahmad, I. and Wasiq, M. (2012) ‘Effectiveness of Social Media As a Marketing Tool: an Empirical Study’, International Journal of Marketing, Financial Services & Management Research, 1(11), pp. 88–99. Available at: www.indianresearchjournals.com. Breitenecker, R. J. (2014) ‘Customer engagement behaviour in online social networks – the Facebook perspective Sofie Bitter * and Sonja Grabner-Kräuter’, 14. Brien, H. L. O. and Toms, E. G. (2008) ‘What is User Engagement? A Conceptual Framework for Defining User Engagement with Technology’, Journal of the American Society for Information Science and Technology 59(6):938-955, (April). doi: 10.1002/asi.20801. Dawes, J. et al. (2018) ‘Forecasting advertising and media effects on sales: Econometrics and alternatives’, International Journal of Market Research, 60(6), pp. 611–620. doi: 10.1177/1470785318782871. Debreceny, R. (2019) ‘Research in Social Media : Data Sources and Methodologies Research in Social Media : Data Sources and Methodologies’, (December 2017). doi: 10.2308/isys-51984. Frolova, S. (2014) ‘Svetlana Frolova THE ROLE OF ADVERTISING IN PROMOTING A PRODUCT Thesis CENTRIA UNIVERSITY OF APPLIED SCIENCES Degree Programme in Industrial Management’, (May). Di Gangi, P. M. and Wasko, M. (2016) ‘Social media engagement theory: Exploring the infuence of user engagement on social media usage’, Journal of Organizational and End User Computing, 28(2), pp. 53–73. doi: 10.4018/JOEUC.2016040104. Goldsmith, R. E. and Lafferty, B. A. (2002) ‘Consumer response to Web sites and their influence on advertising effectiveness’, Internet Research, 12(4), pp. 318–328. doi: 10.1108/10662240210438407. Hummel, T. J. and Sligo, J. R. (1971) ‘Empirical comparison of univariate and multivariate analysis of variance procedures’, Psychological Bulletin, 76(1), pp. 49–57. doi: 10.1037/h0031323. Insight - Pages (2021). Available at: https://developers.facebook.com/docs/platforminsights/page (Accessed: 29 March 2020). Kong, Y. H., Lim, K. Y. and Chin, W. Y. (2021) ‘Time Series Forecasting using a hybrid Prophet and LSTM model’, in. Kundi, F. M. et al. (2014) ‘Detection and scoring of Internet Slangs for sentiment analysis using SentiWordNet’, Life Science Journal, 11(9), pp. 66–72. doi: 10.6084/M9.FIGSHARE.1609621. Lee, T.-S., Shia, B.-C. and Huh, C.-L. (2016) ‘Social Media Sentimental Analysis in Exhibition’s Visitor Engagement Prediction’, American Journal of Industrial and Business Management, 06(03), pp. 392–400. doi: 10.4236/ajibm.2016.63035. Li, C. et al. (2015) ‘Click-through prediction for advertising in twitter timeline’, Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data 182 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Mining, 2015-Augus, pp. 1959–1968. doi: 10.1145/2783258.2788582. Polat, C. (Niğde Ü. (2007) ‘The Role of Forecasting and Its Potential for Functional Management: A Review from the Value-Chain Perspective’, Dokuz Eylül Üniversitesi Sosyal Bilimler Enstitüsü Dergisi, 9(1), pp. 373–398. Saccenti, E. et al. (2014) ‘Reflections on univariate and multivariate analysis of metabolomics data’, Metabolomics, 10(3), pp. 361–374. doi: 10.1007/s11306-013-0598-6. Schoen, H. et al. (2013) ‘The power of prediction with social media’, Internet Research, 23(5), pp. 528–543. doi: 10.1108/IntR-06-2013-0115. Sidi, L. (2020) ‘Improving S&P stock prediction with time series stock similarity’. Available at: http://arxiv.org/abs/2002.05784. Srinivasan, B. V. et al. (2013) ‘Will your facebook post be engaging?’, International Conference on Information and Knowledge Management, Proceedings, pp. 25–28. doi: 10.1145/2512875.2512881. Waeto, S., Chuarkham, K. and Intarasit, A. (2017) ‘Forecasting time series movement direction with hybrid methodology’, Journal of Probability and Statistics, 2017. doi: 10.1155/2017/3174305. Yasuko, K., Etuso, G. and Akira, I. (2014) ‘P REDICT F ACEBOOK I MPRESSIONS ADOPTING A MATHEMATICAL MODEL OF THE H IT’, 2(1), pp. 63–68. Yenidogan, I. et al. (2018) ‘Bitcoin Forecasting Using ARIMA and PROPHET’, UBMK 2018 - 3rd International Conference on Computer Science and Engineering, (February 2019), pp. 621–624. doi: 10.1109/UBMK.2018.8566476. 183 of 225 ICDXA/2021/19 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 E-ENTERTAINMENT: FACTORS TO ONLINE GAME ADDICTION AMONG TAR UC STUDENTS IN KL Foong Zeng Yaw1, Kah Ming Cheok1, Kai Zhun Ng1, Jian Xiang Teo1, Tin Ting Ting2 and Siew Mooi Lim1* 1Faculty of Computing and Information Technology, Tunku Abdul Rahman University College, Kampus Utama, Jalan Genting Kelang, 53300, Wilayah Persekutuan Kuala Lumpur, Malaysia 2Faculty of Information Technology, INTI International University, Negeri Sembilan, Malaysia *Corresponding author: [email protected] ABSTRACT Online gaming is now a well-known electronic entertainment for people all around the world, especially university students. However, the increasing popularity of online gaming may lead to addiction, a problem that has received significant attention. This study aims to identify the factors relating to online gaming addiction. Based on 118 responses collected from online questionnaires, the bivariate correlation test was employed to examine the relationship between depression, loneliness, motivation for escapism and motivation for achievement with online game addiction. Cohen’s effect size f2 for each path is calculated. The findings show that depression, loneliness, motivation for achievement and motivation for escapism have a high positive correlation with online game addiction with large effect size. Keywords: Online game addiction, depression, loneliness, motivation for achievement, motivation for escapism 1.0 INTRODUCTION Video games now have a widespread presence in electronic entertainment, supported by over 2 billion active gamers worldwide. The video games market is expected to be worth over USD90 billion by 2020, with online games making up a large portion of growth and revenues (McDonald, 2017). A good example of the popular online game Fortnite, which has an estimated 125 million players (Fortnite, 2018). It is estimated that the online gaming industry will continue to grow exponentially because of the COVID-19 lockdowns. Online gaming’s popularity, especially among the youth and young adults can largely be caused by online games’ ability to provide a form of escapism from the reality of the world and make people feel belonged in a community through the social aspects of the games (André et al., 2018). However, despite the positive aspects of playing games, the increasing popularity of online games has caused detrimental effects as the amount of online gaming addiction (OGA) cases increased. Several consequences follow, ignoring responsibilities such as jobs and family, health problems and in the most extreme cases, deaths. The worries of OGA are enforced with the World Health Organisation (WHO) acknowledging “gaming” falls under the category “Disorders due to addictive behaviours”, effectively treating gaming the same as alcohol, drugs and gambling (WHO, 2021). Rising cases of OGA leads to a question: “What makes people addicted to online gaming?”. Despite the consequences of online gaming 184 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 addiction being well researched and known, there is a lack of research on why people get addicted to online gaming in the first place. It is theorized that the factors that lead people interested in online gaming in the first place is what causes them to be addicted. Yee (2006), in his research, has identified several factors that cause people to be interested in a type of online game known as Massively Multiplayer Online Roleplaying Game. Among these factors are motivation for achievement and motivation for escapism. In another study, loneliness and depression are discovered to have a mutual enforcing loop with online gaming addiction (Kim et al., 2009; Burleigh, 2018). An individual that is unsatisfied or disappointed in their lives may develop gaming motivation for escapism and achievement, which drive an individual to play online games. Gaming motivation strengthened by their psychological state eventually causes online games addiction. We have chosen depression and loneliness as our psychological factors since they are commonly faced by university students. Motivation for escapism and achievement are chosen for our gaming motive factor since they are logical derivations from an unsatisfied life. In short, this paper aims to prove psychological factors (depression and loneliness) and gaming motivations (escapism and achievement) are positively associated and are predictors of online gaming addiction. 2.0 LITERATURE REVIEW 2.1 Online Gaming Addiction (OGA) Research on OGA has been done across various fields of knowledge, education, information systems, including social psychology, and psychiatry (Hsu et al., 2009; Xu et al., 2012; Kim et al., 2008). In a large-scale longitudinal study of 3034 elementary students in Singapore, pathological gaming, defined as gaming that resulted in a dysfunctional family, friends, and school relationships. It was also related to depression, social phobias, anxiety, and lower grades in 9% of the study participants (Gentile, 2011). Based on these estimates, the recent, significant increases in online gaming is alarming. Based on the 9% estimate, over 11 million Fortnite players may exhibit a harmful gaming pathology (Fortnite, 2018). Given the prevalence of online gaming with an estimated 2.2 billion active gamers worldwide, the problem may be even more substantial (McDonald, 2017). 2.2 Depression and OGA Research theorizes that increased levels of depression are related to different forms of addiction (Griffiths et al., 2016). The possibility of depression leading to OGA may be explained based on addictions often functioning as maladaptive emotional regulation strategies (Stavropoulos et al., 2016). Based on the research, loneliness and depression were proven related to symptoms of pathological gaming in a mutually upholding cycle (Krossbakken et al., 2018). The researcher Taechoyotin (2020) has discussed that the person might feel stressed, depressed, or anxious by the problems in the real world and may choose to use the game world (where they feel safe and secure) to escape these feelings. Burleigh (2018) has identified that depressed adolescents were significantly more likely to be addicted to online games when they experienced stronger Game Avatar Relationships. 2.3 Loneliness and OGA Loneliness can be defined as the unhappy and disturbing sentiment due to the absence of companionship. In general, it is assumed that loneliness relates to social isolation. However, 185 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 people can be lonely even when they are surrounded by other people. Accordingly, qualitative aspects of social relationships may be more closely connected to loneliness than quantitative ones. Based on the past research, a cross-sectional study conducted by Kim et al. (2009) indicated that a reciprocal relation between pathological gaming and loneliness among adolescents cognitive-behavioural model of PIU. The study showed that individuals who were lonely or did not have good social ability may develop strong compulsive Internet use behaviours. The researchers Jeong et al. (2015) have discussed that loneliness is positively related to game addiction. When people are socially excluded or feel lonely, they used to search for channels to gratify their needs or to relieve their stress. Access to online games is a relatively easy way to solve desire or release stress because online games are a channel close at hand. Thus, they can frequently contact online games rather than interact with others. Furthermore, the researcher Chen and Leung (2016) has discussed that loneliness was significantly linked to mobile game addiction. 2.4 Motivation for Escapism and OGA Yee (2006) debated that motivation for escapism is one of the four components of game immersion. Escapism refers to a person's attempt to avoid thinking about or to run away from real-life problems by engaging in an online experience (Yee, 2006). The research published by Bányai et al. (2019) constructed a questionnaire that collected 4284 results from e-sport and recreational gamers. The results stated that the escapism motive appeared to be the common predictor of problematic gaming among both e-sport and recreational gamers. Another research analysed 27 studies, with only 7 studies with negative outcomes, 9 studies with positive outcomes and 11 studies having an escapism relationship with both mixed outcomes in a given independent study (Hussain et al., 2021). It was found that in western countries, escapism via video games held a stronger association with negative outcomes while in non-Western countries, escapism via video games is more likely to lead to positive outcomes. Another study, published by Šporčić and Glavak-Tkalić (2018) had gathered 509 young adults via questionnaire with the hierarchical regression analyses suggested that escapism is a significant predictor of problematic online gaming. 2.5 Motivation for Achievement and OGA In a self-determination theory, competence, the feeling of the capability to perform tasks, is one of three basic inborn needs of human beings. In accordance with this, Yee (2006) suggests that in gaming, the sense of achievement originates from three components: advancement, mechanics, and competition. Advancement is the players' interest in gaining power and accumulating in-game wealth. Mechanics refers to players' interests in analysing the underlying rules, levelling up characters, and optimizing character performance. Competition refers to the desire to challenge and compete with other players. Studies conducted in the past have suggested that specific psychological characteristics drive OGA. Yee (2006) collected online survey data from 30,000 users of Massively Multi- User Online Role-Playing Games (MMORPGs) over a period of three years to explore users’ demographics, motivations, and derived experiences. His study reveals that male players were significantly more likely to be motivated by the achievement and manipulation factors (Yee, 2006). Following this framework, Chang et al. (2018) examined the mediational effects of multiple gaming motives, from online game involvement to problematic Internet use. They discovered that advancement motives have a positive relationship with online game 186 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 involvement. T’ng and Pau (2021) assessed 1175 Malaysia MOBA gamers to study the avatar in the relationship between motivations of gaming and OGA. The findings revealed that motivation of achievement, motivation of immersion, and identification of avatar positively predict OGA. Besides that, the results of Khan and Muqtadir’s (2016) research indicated that problematic gamers had stronger motivation for socialization, achievement, and immersion compared to non-problematic gamers. 3.0 RESEARCH METHODOLOGY Four factors that are deemed the most possible causes for gaming addiction in TAR UC students which are depression, loneliness, motivation for escapism, and motivation for achievement are selected for this study. The justifications are as follows. Past studies suggest that MMO players create an avatar in which they often imbue part of their identity and idealized identity (Bessiere et al., 2007). This may prompt them to project their idealized selves onto their avatars as a way to regulate related depressive emotions (Bessiere et al., 2007). Therefore, we hypothesize that depression is related to online gaming addiction. Psychopathologies, including Attention-Deficit/Hyperactivity Disorder (ADHD) and depression, were the most significant factors of online gaming addiction in individuals. The idea is that people who suffer from psychological problems (e.g. loneliness) may use online or video games as a way to satisfy their needs, which cannot be fulfilled in real life so that they escape from negative moods. Consequently, emotionally susceptible individuals may be deeply immersed in virtual life. Thus, we hypothesize that loneliness is related to online game addiction. A handful of research projects have suggested that escapism motivation increases the extent of online game playing (Yoo, Sanders and Cerveny, 2018). We suggest that higher levels of engagement and more time spent on the game can afford players the opportunity to be more familiar with the game world and to acquire a sense of belonging and closeness, which, in turn, can lead to online game addiction. A study by Chang, Grace M.Y.Hsieh & Sunny S.J.Lin (2018) shows that the desire for advancement encourages players to stay in the game. Sepandar Sepehr & Milena Head (2018) also suggest that the perception of video game competitiveness is a strong predictor of gameplay satisfaction. Within a gaming environment, a player with the need for increased competence is likely to seek more power, higher- performing characters, and rare items to outperform others, which help generate feelings of capability. Therefore, we propose that the motivation for achievement keeps gamers engaged in the gaming environment, which, in turn, facilitates online game addiction. With this in view, the followings are hypotheses of this study: H1: Depression is positively related to TAR UC students online gaming addiction. H2: Loneliness is positively related to TAR UC students online gaming addiction. H3: Motivation for escapism is positively related to TAR UC students online gaming addiction. H4: Motivation for achievement is positively related to TAR UC students online gaming addiction. To test our hypotheses, an online questionnaire is used. The questionnaire was distributed to TAR UC students in Google Forms from the 1st of August, 2021 until the 19th of September, 2021 (Figure 1). The questions are split into different sections with respect to which variable they are intended to gather data for. For determining the OGA of the individual, we use the Lemmens et al. (2009) OGA scale, which is a widely used game addiction scales. We adapted Yee’s gaming motivation items to measure achievement 187 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 motivation and escapism motivation. For loneliness, the questions were crafted in regard to the UCLA loneliness scale (Version 3). As for depression, we referenced the Center for Epidemiologic Studies Depression Scale (CES-D) for the questions. The questions used to acquire data about the variables are as follow: Game Addiction 1. I think about playing games all day long. 2. I often find I have to increase my playing time to get the desired enjoyment. 3. Me or others unsuccessfully tried to reduce my game use. 4. I feel anxious when I am unable to play games frequently. 5. I often have arguments with others (e.g. family, friends) over the time spent on playing games. 6. I often neglect other important activities (e.g., school, work, sports) to play games. Depression 1. I lack the motivation to do simple things such as cleaning and showering. 2. I do not have hope for the future. 3. I have no goals, or have given up on them. 4. I feel worthless, and guilty when people care for me. 5. I often feel lost and confused. Loneliness 1. I often feel that I am not close to anyone. 2. I often feel my interests and ideas are not shared by those around me. 3. I often feel I am isolated from others. 4. I often feel left out. Escapism 1. I enjoy being immersed in a game world. 2. I often play so I can avoid thinking about some of my real-life problems or worries. 3. I often play to relax from the day's work. 4. It is important for me that the game allows me to escape from the real world. Motivation for Achievement 1. It is important for me to level up my character as fast as possible. 2. It is important for me to acquire rare items that most players will never have. 3. It is important for me to become powerful in games I play. 4. It is important for me to accumulate resources, items or money. Figure 1. Questionnaire Items. 5.0 RESULTS AND DISCUSSION Analysis of the data shows that Cronbach’s alpha is well above 0.94, which is indicative of a strong internal consistency of the questions. Table 1 shows the demographic information regarding the responses. 188 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Table 1: Questionnaire respondents’ demographic statistic. Frequency Percent Gender 35 29.7% Female 83 70.3% Male 2 1.7% Programme 1 0.8% Accounting 1 0.8% Advertising 1 0.8% Computer Science 6 5.1% Corporate Administration 1 0.8% Data Science 10 8.5% Engineering 1 0.8% Enterprise Information Systems 1 0.8% FCCI 9 7.6% Graphic Design 6 5.1% Information Security 15 12.7% Interactive Software Technology 1 0.8% Internet Technology 2 1.7% Logistics and Supply Chain Management 1 0.8% Marketing 1 0.8% Mass Communication 1 0.8% Mechatronic Engineering 3 2.5% Multimedia Design 33 28.0% Software Engineering 18 15.3% Software Systems Engineering 1 0.8% Software Systems Development 1 0.8% Architecture 1 0.8% Finance and Investment 1 0.8% International Business Mechatronic Engineering 118 100% Pearson Correlation and Effect Size (Cohen’s f2) are shown in Table 2. Based on Table 2, there is a significant positive relationship between Depression and Online Game Addiction (r = 0.531, sig = 0.000) with large effect (f2 = 0.39). Thus, H1 is accepted. Online games allow players to create an avatar that they may imbue their personality into. It is a possible driver to escape depressive feelings which leads to OGA. The result is consistent with the findings of Burleigh (2018) which demonstrated that depression is statistically significant related to OGA. Table 2. Pearson Correlation and Effect Size. Pearson Sig Mean Std Dev Cohen’s Correlation 2.42 1.16 f2 Online Game Addiction 0.39 Depression 0.531 .000 189 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 Loneliness 0.531 .000 2.79 1.13 0.39 Motivation for Escapism 0.571 .000 3.60 0.97 0.48 Motivation for Achievement 0.662 .000 3.22 1.11 0.78 Dependant Variable: Online Game Addiction (OGA) f2 ≥ .02 = small effect; f2 ≥ .15 = medium effect; f2 ≥ .35 = large effect (Cohen, 1988) Pearson Correlation for Loneliness and OGA is positive, r = 0.531 with effect size of f2 = 0.39 (large effect). This concludes there is a high and positive correlation between Loneliness and OGA and practically significant in the real world (effects size Cohen’s f2 is large (Pritha, 2021)). Therefore, H2 is accepted in which there is a correlation between Loneliness and OGA. The social aspect of online games provides an easy way for social interaction for people who feel lonely. Other than that, online games provide anonymity and are less socially demanding. This could explain loneliness leading to addiction. The result is consistent with the findings of Jeong et al. (2015) which revealed that loneliness, aggression and depression predict OGA. When people are socially excluded or feel lonely, they used to search for channels to gratify their needs or to relieve their stress. Access to online games is a relatively easy way to solve desire or release stress because online games are a channel close at hand. Thus, they will rather choose to socialize in games, distancing themselves from others, which eventually leads to addiction. Pearson Correlation for Motivation for Escapism and OGA is also positive, r = 0.571 with large effect size, f2 = 0.48. Thus, H3 is accepted in which motivation for escapism is related to online gaming addiction and this is significant in the practical real world. The relationship of Motivation for Escapism is higher compared to Depression and Loneliness. This could be attributed to trending online games now that feature immersive worldbuilding. The result is consistent with the findings of (Hussain et al., 2021) which states that motivation for escapism is correlated with Online Gaming Addiction. Based on Table 2, the Pearson Correlation for Motivation for Achievement and OGA is r = 0.662 which is a high correlation between Motivation for Achievement and OGA. Thus, H4 is accepted with the largest effect size, f2 = 0.78. The acceptance of H4 suggests that our initial assumptions were correct, as people who view achievement as important may be attracted to online gaming. The result is consistent with the findings of T’ng et al. (2021) which revealed that motivation of achievement positively predicts OGA. Since it is easier to gain achievement within a virtual environment, motivation of achievement may drive players to play online games to satisfy their desire for achievement. This leads to increasing playtime to hunt for higher level of achievement, driving them to become addicted. Table 3. Pearson Correlation between factors. DP LON ME MA 0.443 DP Pearson Correlation - 0.642 0.362 0.000 0.321 Sig. (2-tailed) 0.000 0.000 0.000 0.516 LON Pearson Correlation 0.642 - 0.443 0.000 Sig. (2-tailed) 0.000 0.000 - ME Pearson Correlation 0.362 0.443 - Sig. (2-tailed) 0.000 0.000 MA Pearson Correlation 0.443 0.321 0.516 Sig. (2-tailed) 0.000 0.000 0.000 DP – Depression; LON – Loneliness; ME – Motivation for Escapism; 190 of 225 ICDXA/2021/20 @ICDXA2021

International Conference on Digital Transformation and Applications (ICDXA), 25-26 October 2021 MA – Motivation for Achievement Additional statistical analysis is conducted to examine the correlation between factors and the result is shown in Table 3. It is found that there are significant positive relationship between the four factors especially between Depression and Loneliness (r=0.642, sig=0.00). It is interesting to further find out the correlation between factors that could possibly affect students addiction in the online gaming in the future research. For example, students that are lonely would probably have depression and this could cause the students to immerse in the virtual world of gaming. 6.0 CONCLUSIONS In short, we have collected data from 118 respondents to answer our research questions and have proven that depression, loneliness, motivation for achievement and escapism have a positive relation with OGA and the effect sizes are large. Our results show that motivations for achievement and escapism are closely related to OGA. Although not as strong as the previous two factors, loneliness and depression are still related to OGA. This shows that all four factors are predictors of the tendency of OGA in a person. As these four factors are closely related, this seems to suggest that all four factors and OGA form a mutually reinforcing loop in which an individual unsatisfied with real-life gains online game addiction, which causes the individual to be more unsatisfied with real-life and becomes even more addicted. Further research can be done to include the population from more universities or schools. It is also recommended to further identify factors that cause online game addiction with the consequences of OGA methods of OGA prevention in order to protect young generation from OGA. REFERENCES André JP, Jan KC, Dominika BO, Florian E and Leane A (2018) Online gamers, lived experiences, and sense of belonging: students at the University of the Free State, Bloemfontein. Qualitative Sociology Review 14(4): 122–137, https://doi.org/10.18778/ 1733-8077.14.4.08. Bányai F, Griffiths MD, Demetrovics Z and Király O (2019) The mediating effect of motivations between psychiatric distress and gaming disorder among Esport gamers and recreational gamers. Comprehensive Psychiatry 94, https://doi.org/10.1016/j.comppsych. 2019.152117. Bessière K, Seay AF and Kiesler S (2007) The ideal elf: Identity exploration in world of warcraft. Cyberpsychology and Behavior 10(4), https://doi.org/10.1089/cpb.2007.9994. Burleigh TL, Stavropoulos V, Liew LWL, Adams BLM and Griffiths MD (2018) Depression, Internet gaming disorder, and the moderating effect of the gamer-avatar relationship: an exploratory longitudinal study. International Journal of Mental Health and Addiction 16(1): 102–124, https://doi.org/10.1007/s11469-017-9806-3. Chang SM, Hsieh GMY and Lin SSJ (2018) The mediation effects of gaming motives between game involvement and problematic Internet use: Escapism, advancement and socializing. Computers and Education 122, https://doi.org/10.1016/j.compedu.2018.03.007. 191 of 225 ICDXA/2021/20 @ICDXA2021