Research On Text Sentiment Analysis Based On Bert And Multi-granularity Convolutional Capsule Network

With the development of society,a large number of user comment texts in different fields have been generated from the Internet.These texts contain rich sentimental tendencies,from which Internet users or consumers' attitudes towards the matter can be discovered.This information is of great significance to individual users,businesses or national regulatory authorities.In the era of big data,with the continuous upgrading of technology and hardware facilities,the amount of comment text data on various platforms is increasing explosively.Only relying on statistical methods can no longer meet the growing demand for sentiment analysis of comment text,so how to use artificial intelligence Accurately and efficiently digging out emotional tendencies in review texts has become a major research issue in industry and academia.Sentiment analysis is also known as opinion mining.Emotional tendency mining through artificial intelligence technology helps to understand online public opinion in time,and its research is of great significance for obtaining comment sentiment trends.The main research contents of this paper are as follows:(1)In order to more realistically simulate the needs of sentiment analysis of comment texts in a complex Chinese language environment,a large number of user comment texts were obtained from two websites,Jingdong and Douban,through data collection technology,and the text was preprocessed and divided into five thousand according to the amount of data.,20,000 and 50,000 data sets of different sizes to verify the analysis effects of different sentiment analysis models on different types and different sizes of data sets.Aiming at the problem that the traditional Word2 vec and Glo Ve text representation models cannot accurately express the semantics between words,this paper proposes to use the BERT pre-training model to represent the text of the corpus.After experimenting with the results of Word2 vec and Glo Ve text representation,the word vector generated by BERT is verified.Can achieve better classification results.Then five basic deep learning models were constructed based on CNN and RNN respectively,and the data set built in this article was used to verify the effect of sentiment analysis of the model.(2)Based on five basic deep learning models,a sentiment analysis model that integrates bidirectional gated recurrent network,multi-head self-attention mechanism and multi-granularity convolutional capsule network(Bi GRU-AMCaps Net)is proposed.First,a bidirectional gated recurrent network is used.The text sequence features are initially proposed,and then the multi-head self-attention mechanism is used to weight the key features.Finally,the feature fusion is performed through the multi-granularity convolutional capsule network to output the classification results.The experimental results show that based on the BERT word vector,the model proposed in this paper has an accuracy of 87.29%,89.62%,and91.88% in the three data sets of movie review text,and the accuracy of the three data sets of e-commerce review text.82.19%,85.63%,and 87.46% are better than other models in this field,which proves the application value of the model proposed in this paper.