Research On Sentiment Analysis Based On The Feature Fusion Of CNN And BLSTM

The rapid development of the Internet facilitates people's communication.Many netizens express their opinions through Weibo,WeChat and other public platforms,resulting in a large number of social network data with subjective emotions.Sentiment analysis technology analyzes and researches a large amount of social network data and mines its potential information,so as to analyze the attention and emotional tendency of netizens on social hot topics,and finally provides support for relevant departments'policy formulation and correctly guides emotional transmission of netizens.The early texts in the field of emotional analysis mainly focused on news,blog and other long text data.With the rapid development of social networks such as Sina Weibo and WeChat,short text sentiment analysis based on product reviews,movie reviews,social hot event reviews has gradually become one of the hottest research topics in the field of sentiment analysis.With the deepening of research,coarse-grained emotional analysis for short texts has been relatively perfect,but fine-grained emotional analysis still has a lot of room for development.However,there is still a lack of fine-grained data sets for sentiment analysis of Chinese reviews.Based on this problem,this paper takes Weibo comments on social hot topics on Sina Weibo platform as the research object,and crawls the commentary data on different topics,including family planning policy,poverty alleviation policy,environmental protection events and haze events.According to certain data processing criteria and annotation criteria,the data set is preprocessed and fine-grained emotional annotation is carried out.Emotional analysis data sets for different topics are obtained.On this basis,an emotional analysis model CNN-BLSTM based on feature fusion of convolutional neural network(CNN)and bi-directional recurrent neural network(BLSTM)is constructed.Based on the separation of convolutional neural network and recurrent neural network,a combination of phrase features extracted by CNN and sequence features extracted by BLSTM is proposed to effectively enhance the ability to extract information of the text.In view of the more important influence of specific emotional words on text sentiment analysis,this paper further proposes CNN-BLSTMATT model,which introduces attention mechanism for introducing the local feature representation extracted by CNN into the emotional feature representation of BLSTM,so as to effectively enhance the ability of BLSTM to capture emotional semantic information,and achieve the enhancement effect of text emotional features.Finally,the experimental results on the constructed data set and the open English data set Stanford Sentiment Treebank(SST)[1]show that the proposed CNN-BLSTM model can achieve better experimental results than the single CNN or BLSTM model.In addition,the proposed CNN-BLSTMATT model has better classification accuracy than the CNN-BLSTM model in the text of explicit emotional expression.