Chinese Restaurant Industry Review Sentiment Analysis Method Based on Integration Model

User evaluation analysis is essential for improving user satisfaction and merchant service quality in the catering industry. Sentiment analysis, a key task in natural language processing, can help understand users’ emotions and attitudes. However, traditional sentiment analysis methods often fail to capture global text information accurately, resulting in suboptimal and incomplete analysis results. To address this issue, we propose an integrated model for sentiment analysis that combines BERT (Bidirectional Encoder Representation from Transformers), TextRNN (Text Recurrent Neural Network) and TextCNN (Text Convolutional Neural Network) feature processing prediction model and weight assignment model. The model exploits BERT’s powerful semantic representation capabilities, as well as the long-term and short-term memory and local feature extraction capabilities of TextRNN and TextCNN, and considers different levels of information in the text. Through the weight assignment model, the model can dynamically adjust the output weights of the three basic models, to obtain the optimal integrated result. We evaluate the integrated model on a user evaluation data set of the catering industry, and the results show that the integrated model surpasses the three basic models in terms of accuracy, precision, recall and F1 score, demonstrating its effectiveness and superiority.