Public Opinion on Multisource Data-Driven Identification in Social Networks—Evidence from 323 Food Safety Incidents (2000-2024)

The online public opinion triggered by food safety incidents is related to people's well-being, government trust and social stability. This study proposes a multisource data-driven framework for identifying and analyzing public opinion related to food safety events in social networks. The framework includes four components: data collection through automated web crawling; data preprocessing including noise reduction and time-series structuring; sentiment and thematic analysis using a domain-enhanced emotion dictionary, word frequency clustering, and Latent Dirichlet Allocation (LDA) modeling; and public opinion identification by integrating. A lightweight, interpretable logistic regression model with enhanced features is developed to achieve high classification accuracy. Validation based on three representative cases selected from a dataset of 323 food safety incidents demonstrates that the model achieves an average F1 score of 0.98 thus confirming its effectiveness. This study offers a comprehensive approach to public opinion analysis in the context of food safety and provides transferable methods for early detection of public sentiment in other domains, thereby helping to mitigate potential social impacts.