Sentiment Analysis of online reviews based on LDA and AP-Bert model

Abstract

The purpose of this paper is to construct a more accurate behavior matrix by using fine-grained aspect level emotion analysis method. Firstly, LDA topic extraction model is used to extract the topic of online review text, and the concerned attributes are extracted. According to the characteristics of online comments, a BERT emotion analysis model with enhanced pooling was proposed. Activation function layer and max-average pooling layer were designed to solve the over-fitting problem of BERT model in the process of emotion analysis. Finally, by combining LDA extraction results and AP-Bert sentiment analysis results, the proportion matrix is obtained. Experimental results show that the accuracy, recall rate and AUC value of AP-Bert model are better than those of the same type of model and original BERT model.