Pew Research Center estimates that as of 2014, 74% of the Internet Users used social media, i.e., more than 2.4 billion users. With the growing popularity of social media where Internet users exchange their opinions on many things including their daily life encounters, it is not surprising that many organizations are interested in learning what users say about their products and services. To be able to play a proactive role in steering what user’s say, many organizations have engaged in efforts aiming at identifying efficient ways of marketing certain products and services, and making sure user reviews are somewhat favorable. Favorable reviews are typically achieved through identifying users on social networks who have a strong influence power over a large number of other users, i.e. influential users.
Twitter has emerged as one of the prominent social network services with 320 million monthly active users worldwide. Based on the literature, influential Twitter users have been typically analyzed using the following three models: topic-based model, topology-based model, and user characteristics-based model. The topology-based model is criticized for being static, i.e., it does not adapt to the social network changes such as user’s new posts, or new relationships. The user characteristics-based model was presented as an alternative approach; however, it was criticized for discounting the impact of interactions between users, and users’ interests. Lastly, the topic-based model, while sensitive to users’ interests, typically suffers from ignoring the inclusion of inter-user interactions.
This thesis research introduces a dynamic, comprehensive and topic-sensitive approach for identifying social network influencers leveraging the strengths of the aforementioned models. Three separate experiments were conducted to evaluate the new approach using the information diffusion measure. In these experiments, software was developed to capture users’ tweets pertinent to a topic over a period of time, and store the tweet’s metadata in a relational database. A graph representing users was extracted from the database. The new approach was applied to the users’ graph to compute an influence score for each user.
Results show that the new composite influence score is more accurate in comprehensively identifying true influential users, when compared to scores calculated using the characteristics-based, topic-based, and topology-based models. Also, this research shows that the new approach could leverage a variety of machine learning algorithms to accurately identify influencers.
Last, while the focus of this research was on Twitter, our approach may be applicable to other social networks and micro-blogging services.
| Identifer | oai:union.ndltd.org:unf.edu/oai:digitalcommons.unf.edu:etd-1789 |
| Date | 01 January 2017 |
| Creators | Gamal, Doaa |
| Publisher | UNF Digital Commons |
| Source Sets | University of North Florida |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | UNF Theses and Dissertations |
Page generated in 0.0031 seconds