Research On Chinese Text Sentiment Classification Based On Neural Networks And Its Application In Public Opinion Analysis

With the rapid development of Internet technology,especially the development of Mobile Internet technology,Netizens could express their opinions on hot events at any time and any places through various and convenient ways.At the same time,the quantity of text data on the Internet has also shown a spurt of growth.Faced with such a large amount of text data,how to dig the sentimental attitudes of netizens and extract valuable information,and then correctly guide the public opinion is an urgent problem to be solved.Classify the sentiments of text data and extract the key information from different sentimental categories could effectively solve this problem.Therefore,text sentiment classification technology and key information extraction technology are widely used in E-commerce,E-government,information management and other fields.At present,the more mature text sentiment classification technology is mainly aimed at English texts,and there are few sentiment classification technology aimed at Chinese texts.Compared with English texts,Chinese texts have large difference in grammatical structures and semantic contents,so the sentiment classification technology for English texts could not be applied to Chinese texts directly.Therefore,we use the neural network method to research on Chinese text sentiment classification.At present,scholars mainly divide Chinese text sentiment into two categories: positive and negative.However,this classification is relatively rough and could not analyze the deeper sentimental tendency of texts.So,we classify the positive and negative sentiment into deeper sub-classifications,build a sentimental classification models to classify positive and negative sentiment and build another sentimental classification model to classify sub-categories of sentiment.On the other hand,scholars' research on sentiment classification of Chinese text mainly focuses on word or sentence granularity,but different combinations of words would convey different sentimental tendencies.If we only analyze sentimental categories of texts through word granularity,we would obtain a less accurate result of classification.Therefore,we refine the texts into word,phrase and sentence granularities,and build a neural network classification model based on these three granularities respectively,and then obtain the results of text sentimental classification based on different granularities.In addition,we also combine the word features,phrase features and sentence features obtained by these three models,and further obtain the multi-level sentimental features of texts.Finally,we realize the multilevel sentiment analysis of text and improve the accuracy of text through the multi-level sentiment classification network.The sentimental classification result of texts could not only understand author's subjective sentimental tendency,texts with different sentiment could convey information of different values.Therefore,result of sentimental classification could also be applied to extract key information of texts.In order to distinguish the contribution of different sentimental tendencies,we introduce the sentimental classification results of texts in traditional word frequency statistics method,and proposed a key information extraction method that combines text sentimental category feature.By crawling the text datasets related to “The Belt and Road” in Sina Weibo during the NPC&CPPCC in 2017,we compare and analyze results of traditional key information extraction method and key information extraction method based on text sentiment feature.The experimental results show that method based on text sentiment feature could extract more accurate key information compared with traditional method,finally we realize the accurate extraction of key information in the complicated text datasets in the Internet.