Research On Topic Sentiment Classification For Review Tex

With the rapid development of the Internet,more and more users publish their comments with sentiment tendency through social platforms.The analysis of deep-level sentiment tendency of user comments can help the government understand the public’s sentiment changes in a timely manner,and provides useful information and references for the analysis of public opinion on socially significant public events.Topic model aims to extract potential topic information by analyzing documents to help people understand the main content of documents quickly and accurately.Sentiment analysis technology identifies user sentiment information and digs deeper into users’ opinions,views and attitudes about the things they care about.Users’ comments often express different emotional polarities according to different topics,and the existing sentiment models cannot tap into the potential thematic information of texts,leading to the possibility of getting wrong emotional polarities.To address the above problems,this paper proposes a study on topic sentiment classification for review texts.Firstly,the topic model technique on review text is studied,and then on this basis,the text topic features mined by the topic model are applied to sentiment analysis,and a sentiment classification model based on dualchannel feature fusion is proposed.The specific work is as follows:(1)To address the problem that current research on topic models does not fully consider the connections between documents in the corpus,this paper proposes a neural topic model based on graph convolutional networks.Firstly,a large undirected graph is constructed based on the corpus,and secondly,based on the neural variational inference framework,a graph convolutional neural network is used as an encoder to learn the domain feature relationships among document nodes and obtain the topic distribution.The experiments show that the proposed topic model performs better in terms of topic consistency and topic diversity compared with other comparative models.(2)A sentiment classification model based on two-channel feature fusion is proposed for the characteristics of non-standardized,sparse and arbitrary microblog comment text semantics.Firstly,a BERT pre-training model is used to obtain dynamic word vectors,and then a dual-channel feature extraction network is used for feature extraction,using TextCNNattention to extract local features of text on the one hand,and introducing a neural topic model to extract global features of text on the other hand,followed by splicing local and global features to obtain the final text vector,and finally the Softmax function is used to The sentiment classification labels are outputted.The effective improvement of the accuracy of the sentiment classification model by using the BERT pre-training model is demonstrated by comparing with other word vectors;the superiority of the sentiment classification model proposed in this paper is demonstrated by comparing the experiments with other sentiment classification models.