Research on the Dynamic and Communication Mode of Public Opinion Emotions Based on Text Mining and Sentiment Analysis

The swift dissemination of public opinions on university-related events is inevitably associated with topic variations, emotions, and complicated interpersonal interaction among the users. To explore this problem, this study proposes a hybrid framework including text mining, sentiment analysis, and opinion evolution modeling based on fuzzy theory. Data about public opinions will be collected on social platforms like Weibo, Douyin, and Xiaohongshu, then will be subjected to data cleaning, Chinese word segmentation, and stop-word elimination. An optimized version of BERTopic is employed to detect topics by means of domain knowledge mapping, keyword pre-exploration, topic labeling via assistance of large language models and fine-tuning of embedding models. Sentiment tendencies will be further divided into types with help of a model that combines RoBERTa-BiLSTM and interactive attention and biaffine feature fusion. Based on these analyses, a fuzzy opinion-evolution model is built to evaluate the impact of peer trust, opinion leader trust, timely and credible responses on emotions and consensus. A case of university cafeteria-related public opinion event will be adopted for empirical analysis. The findings indicate that the proposed framework can identify theme and emotion variations in campus public opinion. Higher opinion leader trust can speed up inter-group consensus while too much reliance on peer interactions may delay it. Late and low credibility responses are likely to increase public opinion risk.