Exploring Event-based Dynamic Topic Modeling*

The diversity of public users in X’s (formerly known as Twitter) platform and the instantaneous speed by which information is delivered and presented in the data stream makes it an informal and extremely rich data source for important events. Although X provides a means to filter certain posts (formerly known as tweets), a real time summary of information pertaining to a specific event is not readily available. While previous studies made use of and developed automated and dynamic event detection on social media platforms, there are minimal attempts to perform dynamic topic modeling using this data. This study explores the use of dynamic topic modeling as an approach to event-based topic modeling with data from the X platform. Specifically, the study explores the performance of dynamic topic modeling on posts that mention certain locations in Metro Manila, Philippines using an optimized LDA topic model versus a base LDA model and a BERTopic model. Results from experiments showed that the optimized model consistently outperformed its base model counterpart in terms of topic coherence score, but is slightly on par with BERTopic model.