An exploratory analysis using topic modeling: tracking evolution and loyalty from stack overflow users interests

Abstract

The web presents many platforms for sharing knowledge among users, such as newsletters, blogs, social networks, and communities. Among them, Stack Overflow is a popular question and answer (Q&A) community allowing users to share and acquire knowledge on computer programming topics. If explored, Stack Overflow posts present information that can provide many insights into the shared knowledge. Some works have explored these posts and generated relevant information, but the databases they analyzed are outdated now. Besides, many of them did not consider posts’ authorship and publish dates in their analyses, which may provide useful temporal and user-centric insights. Therefore, this work made several experiments to infer Stack Overflow topics and analyzed them employing proposed metrics to measure the relative popularity of the topics and their drift. Then, these metrics were applied to analyze topics’ popularity across temporal and authorship information, tracking the popularity evolution and the drift of these topics globally and for each individual user. The methodology employed topic modeling, natural language processing, statistical techniques, and validating experiments to accomplish these promising results, which were made web-accessible for free. Finally, the addressed methodology was shown to be effective on performed analyses, which successfully inferred Stack Overflow topics and tracked their general and user-centric popularity evolution