With the rapid development of social media, massive user behavior data and sentiment expression have become important research resources. Existing sentiment analysis methods still have limitations in dealing with complex text and user behavior features. In this study, we constructed a support vector machine-based model for analyzing user behavior and sentiment tendency of social platforms, processed unstructured text data through vector space model, and built a high-dimensional mixed-feature sentiment classifier by combining mutual information value feature extraction and latent semantic analysis. In terms of methodology, firstly, text data is preprocessed by vector space model, and data annotation is performed by using lexicon and weakly labeled information; secondly, feature extraction is performed by using mutual information value computation, and SVM algorithm is used to construct sentiment classifiers; finally, empirical analysis is performed to analyze the user behaviors and emotional tendencies of social platforms. The results of the study show that: short video playing, liking, commenting and sharing behaviors conform to the power law distribution with long-tail effect; the actual conversion rate of short video is only 2.83%, which shows that the user participation is low; in terms of sentiment analysis, the sentiment density of user J in the 38-day observation period reaches 0.8957227, which is significantly higher than that of other users; and the characteristic value of the sentiment transmissibility of user X is 2.83, which is significantly higher than that of other users. The conclusion of the study shows that the constructed highdimensional mixed-feature SVM model can effectively reflect users’ behavioral characteristics and emotional tendencies, providing a technical method for social platform user behavior prediction, emotion monitoring and crisis warning.
