In the face of massive multimodal information, it has become one of the current research hotspots to categorize it according to its sentiment so as to guide users to find valuable information from a large amount of data. Based on the application of fuzzy logic in sentiment analysis, this paper designs a method to analyze sentiment tendencies in a multimodal Chinese corpus. Firstly, text, audio, and video features of the multimodal Chinese corpus are extracted, and a fuzzy sentiment dictionary is constructed. Then, the double hesitant fuzzy set is used to reduce the intensity of the sentiment, and the fuzzy sentiment value is calculated. Then, the fusion of sentiment lexicon, intuitionistic fuzzy inference, and fuzzy comprehensive evaluation model is used to obtain the final sentiment tendency analysis results. The models constructed based on different lexicons all converge after 4 epochs, indicating that the model has strong feature learning ability. After combining the sentiment lexicon, the accuracy of the model’s sentiment classification improves by 2.27%. Compared with other common sentiment classification models, the precision rate, recall rate and F1 value of this paper’s model are improved by 2.41%-6.57%, 2.36%-4.91% and 2.38%-5.58%, respectively. The result of inclination to positive in the sentiment analysis of this paper’s model is 82.3%, with a difference of only 1% from the average value of 83.3% of user evaluation, and it is better than the analysis model of plain text (80.8%), which proves that the multimodal sentiment analysis model in this paper can correctly complete the analysis of sentiment inclination of the review data. This paper provides a new feasible approach for the propensity analysis of multimodal sentiment.