The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Analyzing fluctuation of topics and public sentiment through social media data
Author:	Liu, Haoyue
View Online:	njit-etd2022-038 (xiii, 91 pages ~ 2.8 MB pdf)
Department:	Department of Electrical and Computer Engineering
Degree:	Doctor of Philosophy
Program:	Computer Engineering
Document Type:	Dissertation
Advisory Committee:	Zhou, MengChu (Committee chair) Carpinelli, John D. (Committee member) Hou, Edwin (Committee member) Liu, Qing Gary (Committee member) Wei, Zhi (Committee member)
Date:	2022-08
Keywords:	Topic modeling Sentiment analysis Natural language processing Social media
Availability:	Unrestricted
Abstract:	Over the past decade years, Internet users were expending rapidly in the world. They form various online social networks through such Internet platforms as Twitter, Facebook and Instagram. These platforms provide a fast way that helps their users receive and disseminate information and express personal opinions in virtual space. When dealing with massive and chaotic social media data, how to accurately determine what events or concepts users are discussing is an interesting and important problem. This dissertation work mainly consists of two parts. First, this research pays attention to mining the hidden topics and user interest trend by analyzing real-world social media activities. Topic modeling and sentiment analysis methods are proposed to classify the social media posts into different sentiment classes and then discover the trend of sentiment based on different topics over time. The presented case study focuses on COVID-19 pandemic that started in 2019. A large amount of Twitter data is collected and used to discover the vaccine-related topics during the pre- and post-vaccine emergency use period. By using the proposed framework, 11 vaccine-related trend topics are discovered. Ultimately the discovered topics can be used to improve the readability of confusing messages about vaccines on social media and provide effective results to support policymakers in making their policy their informed decisions about public health. Second, using conventional topic models cannot deal with the sparsity problem of short text. A novel topic model, named Topic Noise based-Biterm Topic Model with FastText embeddings (TN-BTMF), is proposed to deal with this problem. Word co-occurrence patterns (i.e. biterms) are dirctly generated in BTM. A scoring method based on word co-occurrence and semantic similarity is proposed to detect noise biterms. In the