Social Media Network Rumor Theme Mining and Evolutionary Analysis based on DTM Models

Abstract

 Mining social media network rumors and analyzing topic popularity and evolution is crucial for accurately understanding and guiding public opinion development and predicting rumor trends. Using Weibo rumor text data as the analysis sample, this study employs the dynamic topic models (DTMs) to extract topics from rumors across different periods. The optimal number of topics for the DTM model is determined using consistency indicators, while topic popularity is assessed using topic intensity. The analysis covers four aspects: hot topics, discourse characteristics, topic popularity, and evolution patterns. The study identifies five major categories of hot topics: international events, local tourism, school education, social security, and epidemic prevention and control. In terms of discourse characteristics, rumors exhibit four main features: colloquialism, emotionalism, vagueness, and multimodality. Regarding topic popularity, epidemic prevention and control topics show a high level of popularity, while local tourism and school education topics display a gradually increasing trend. In topic evolution, topics of international events remain relatively independent, whereas the other four categories are closely related. This study provides valuable insights for government efforts to manage public opinion and curb the spread of rumors.