An Optimized Influencer Rating Model using a Joint Event and Theme based Approach. (c2020)

Abstract

The continuous development of social media platforms and exponential level of users' engagement are playing a key role in turning social media platforms into a data source that is indispensable for understanding the behavior of people. Yet this comes along with all types of challenges that needs efficient solutions and big data analysis techniques to allow capturing the different dimensions evolving over social media. Moreover, the importance of detecting influencers over social media platforms has become one of the most challenging research topics given that influencers are at the core of decision-making strategies and leading events' directions on Social Media. Generally, determining influencers can increase revenue and utility; however, measuring the influence of users is essential for determining influencers. Influencers should be credible, reliable, trustworthy, knowledgeable in the domain being discussed and have a high impact that derives people's opinions and lead them towards the proper decisions. However, influencers might play different roles when speaking about misinformation and conspiracy during sensitive and trending event. While different techniques were developed to select influencers over social networks, identifying influencers remains an evolving topic due to the dynamic nature of the social media users and use in addition to their extremely increasing growth which attracts high research interests across multiple disciplines including data science, psychology, sociology, and different human sciences all of which have been studying the topic from multiple angles. In this thesis, we further aim at identifying influencers through proposing influence rates calculation mechanism to find real and highly influential users at a certain event over Twitter using a mixed theme and event base approach with an emphasis of maximizing accuracy calculation by integrating historical influence rates in addition to content and profiles reputation. We further apply our approach on a global pandemic, the novel Coronavirus, and then we provide results and performance analysis. Finally, we conclude our work by summarizing our major findings and discussing future work.