Comprehensive Analysis and Application of Geological Disaster Information Leveraging Topic Modeling and Sentiment Mining

GANG CHENG^1,2Yaxi Wu¹Desheng Cao^1*Keshun Wei^3*Ye Wang¹Yongfei Wu¹

• ¹North China Institute of Science and Technology, Langfang, China
• ²Nanjing University, Nanjing, China
• ³University of International Business and Economics, Beijing, China

In recent years, with the rapid advancement of urbanization in China and the successive implementation of major national strategies such as the Belt and Road Initiative, the Sichuan–Xizang Railway, and the South-to-North Water Diversion Project, the potential risks and losses from geological disasters have continued to rise. Secondary disasters—including landslides, mudslides, and barrier lakes triggered by earthquakes—have significantly intensified the overall impact, posing severe challenges to disaster monitoring, early warning, emergency response, recovery, and reconstruction efforts. In this context, how to leverage new information technologies to achieve in-depth mining and application of geological disaster data has become a critical issue in disaster risk reduction and sustainable crisis management. This study focuses on topic modeling and sentiment analysis of disaster-related data, using geological disasters in China as a background. First, it reviews the recent advances in topic modeling and sentiment analysis techniques. Then, based on data characteristics and applicability, two major social media platforms—Weibo (Sina Weibo) and Rednote (Xiaohongshu)— are selected as primary data sources. The advantages of the LDA topic model (e.g., its unlabeled and multi-topic capabilities) and the lightweight processing efficiency of the SnowNLP sentiment analysis algorithm are discussed. As a case study, the "1·07" earthquake in Xigaze, Tibet, in 2025 is analyzed. The LDA model is used to conduct multi-topic classification and clustering visualization of Weibo disaster topic data. Combined with the SnowNLP sentiment analysis algorithm, the phased sentiment evolution judgment application is carried out using the 6-month Rednote comment data. The results demonstrate that the LDA model effectively extracts geological disaster-related themes—such as emergency response and post-disaster recovery—and that sentiment analysis technology can reveal phase-based patterns in public emotions. These findings provide scientific support for geological disaster emergency management and public opinion guidance. The research also expands the application potential of topic modeling and sentiment analysis in the field of geological disasters and offers a direction for future integration and optimization of multimodal social media data.