Multimodal Sentiment Analysis for College Public Opinion Early Warning

The exponential growth of social behavior data has reshaped the paradigm of college students' opinion expression and triggered a geometric leap in the rate of public opinion dissemination, making it urgent to strengthen the management of college network public opinion. At present, the early warning mechanism of college network public opinion has the problems of single-modal data analysis and lagging platform monitoring, which is specifically manifested in the reliance on text single-modal monitoring or manual monitoring, and fails to effectively cover the emerging social platforms of college students. This study uses the social platform “FoxFriends Mini Program” frequently used by college students as a data platform to construct a multimodal data processing framework; by obtaining fields such as user ID, posts, comments, and integrating multimodal information such as text, images, and expressions, a fine-tuned Qwen2.5-VL-7B-Instruct large model is used for sentiment analysis. Experiments show that the method proposed in this paper achieves a high macro precision of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0. 9 \%}$</tex>, a high macro recall of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 3 \%}$</tex> and a high macro F1 of 93.5% on a certain college student social behavior dataset, providing a new perspective and idea for the research on early warning of college network public opinion.