Chinese Review Text Sentiment Analysis Based On Deep Learning

With the rapid development and popularity of the Internet,various comment texts on the Internet have proliferated,such as product comments of e-commerce platforms,social comments of social networking platforms,and service comments of life service platforms.Most of these comment texts contain the emotional tendency information of the reviewers on the review objects.Mining the emotional tendency information is of great significance to individuals,enterprises and society.The text sentiment analysis technique in natural language processing can extract emotional tendency information from comment texts with subjective emotions.Traditional sentiment analysis methods based on dictionary and machine learning have been unable to meet the needs of sentiment analysis of massive comment texts.Therefore,this thesis studies the sentiment analysis method based on deep learning,and the object of study is commodity reviews and film reviews.First of all,in response to the relatively lack of Chinese high-quality sentiment analysis commentary corpus,this thesis used crawler technology to crawl a large number of product reviews and movie reviews from the network,and carries out pre-processing work such as emotional annotation,text cleaning,Chinese word segmentation,to get the review text corpus of this article.Secondly,the word2 vec model was used to text vectorize the preprocessed comment text,and sentiment analysis experiments were performed on basic deep learning models convolution neural network(CNN),long short-term memory(LSTM),and gated recurrent unit(GRU).Since the word vectors trained with word2 vec only contain the semantic information of the word but lack the emotional information and weight information that is beneficial to sentiment analysis,this thesis proposed a word vector representation method combining word2 vec,dictionary and TF-IDF,and carried out word vector comparison experiments on the comment text corpus.Finally,in order to combine the semantic feature extraction ability of bidirectional gated recurrent unit(BiGRU)with the deep feature extraction ability of CNN,this thesis proposed a BiGRU-CNN model combining BiGRU and CNN,and compared the model with the other excellent models on the comment text corpus.The experimental results have shown that compared with the basic word vector representation,the proposed fusion word vector representation could improve the accuracy of sentiment analysis of comment text.On the improvement of the model,the BiGRU-CNN model proposed achieved the accuracy of 93.36%,78.65%,91.73% and 78.52% respectively in two categories of commodity emotion,three categories of commodity emotion,two categories of movie emotion and three categories of movie emotion.Compared with the other excellent models,the effect is better,which verifies the validity of the model proposed in this thesis.