Classroom Feedback Sentiment Analysis and Recognition Based on Multimodal Model

This study intends to explore the path of classroom feedback sentiment analysis and recognition based on multimodal models. Based on the integration of audio, video and text data, it realizes the subtle discrimination of students' emotions. The study adopts the technical means of multimodal sentiment analysis. At the beginning, the audio data is transcribed, followed by the extraction of Mel frequency cepstrum coefficients, facial expression recognition is carried out for video data, and emotional feature extraction is carried out at the same time. The word vector model is used to complete the processing of text data. The convolutional neural network and the recurrent neural network are combined to implement the fusion of each modal data, and the sentiment classification analysis is carried out with the help of the deep learning model. The experimental results show that in terms of accuracy, precision and F1-score and other evaluation indicators, compared with the multimodal data fusion model, the single modal model performance fails to meet the standard, especially when the three modalities of audio, video and text are combined with each other. The final accuracy of the model is 88.4 %. Multimodal sentiment analysis can effectively enhance the accuracy of classroom feedback emotion recognition, provide more accurate sentiment analysis methods for related educational matters, and provide data support for teaching optimization and targeted education programs.