In order to explore the emotional evolution law of online public opinion in universities, the traditional sentiment analysis method was enhanced by integrating the psychological distance factor to improve the effect of emotional evolution analysis. The latent Dirichlet allocation (LDA) topic model was used for topic recognition, and a dual-channel feature sentiment analysis model incorporating attention mechanism (DCMAM) based on the pre-trained model ALBERT was proposed to classify the sentiment of comment texts. By combining with topic recognition and sentiment classification, three evaluation indexes of temporal relevance, spatial relevance and semantic relevance were constructed, and the evolution of online public opinion sentiment in universities was analyzed from three dimensions: temporal distance, spatial distance and cognitive distance. The results show that the accuracy, recall and F1 value of the DCMAM model reach 91.32%, 91.55% and 91.60%, respectively, in the comparative experiments of multiple groups of sentiment analysis models, which can better identify the sentiment tendency of the comment text. The psychological distance evaluation index shows that there are differences in the public opinion attention and emotional evolution of network users under different psychological distances.