Event Detection and Evolution in Online Social Networks

This thesis delves into the analysis of online social networks, specifically focusing on event detection and evolution. These networks, which have attracted billions of users, are a rich source of user-generated content and a platform for social interaction and information sharing. Event detection and evolution on these networks provide valuable insights into public trends, crisis management, market intelligence, and product feedback. However, this task faces several challenges, including information noise, disinformation, linguistic and semantic diversity, event boundary ambiguity, and the need for large-scale data processing.

To address these challenges, this thesis proposes innovative methods and models to process complex factors, which will enhance the accuracy and efficiency of event detection and evolution. This thesis aims to advance the field of event detection and evolution analysis within online social media environments. Firstly, a self-identification Event Detection model based on the topic modelling is presented to filter high-quality tweets and detect potential events effectively. Based on the previous model, we further develop the Twitter-Robot model, which is distilled from a large language model to capture the rich semantic and contextual information within tweets accurately. Finally, considering the dynamic and evolving nature of events, a novel event evolution approach based on event representation learning is proposed by utilising both semantic and syntactic information to enrich event feature representation. The accuracy and efficiency of these proposed models and methods have been demonstrated through extensive experiments.