Research And Impementation On Chinese News Sentiment Classification System

The Internet news information is timely and the content is rich, and more and more people used to read the news information from the Internet. Howev-er, there are many pieces of unhealthy information can affect people's mind a-nd misunderstand the young, so social media monitoring from news informatio-n has drawn much attention in the NLP filed, and opinion mining has become one of the key technologies for handling and analyzing the Internet news inf-ormation. One of the most widely-studied sub-problems of opinion mining is s-entiment classification.In the thesis, the sentiment classification objects generally contain words, s-entences, documents and so on. Since Chinese language is complex, the senti-ment classification research based on word level, sentence level, document lev-e1 faces a great many of difficulties. So the thesis studies the text sentiment c-lassification in Chinese news, and does the following innovative research respe-ctively both the sentence level and the document level.First, construct the corpus and build the sentiment dictionary based on Chi-nese news.Second, do the sentiment classification research based on sentence level.The thesis presents three sentiment classification models in sentence level. The Syntactic Path model:first, collect the syntactic path patterns containing t-opic words and sentiment words matching, and put the patterns into a database. Second, input the sentences, and construct the parse tree to determine whether a relationship between the topic word and the sentiment word exists. The Ve-ctor Space model:based on the topic word, calculate the distance between the topic word and the sentiment word forward and backward respectively, then c-alculate the sentiment score based the vector distance. The Force model:sear-ch the topic words, the sentiment words, the noise words to determine the sen-timent orientation. Third, do the sentiment classification research in document level.The document sentiment classification research is based on the sentence le-vel, the thesis presents three sentiment classification models in document level. The Semantic Orientation model:the approach has four steps:(1) the news d-ocuments are preprocessed; (2) the sentiment words and the negative words are integrated processed; (3) the topic words and the sentiment words are integra-ted processed; (4) the weight is calculated based on a sentiment word dictiona-ry and the context information. The Machine Learning model:use SVM classi-fier, first do feature selection to train the documents, then input test documents, and use SVM classifier to get the document sentiment orientation. The Force model:search topic words and the sentiment word in the document to determi-ne the sentiment orientation.Use the above models to test the constructed corpus based on Chinese ne-ws, the experimental results indicate that the Syntactic Path model performs th-e best in the sentence level and the Semantic Orientation model exhibits the b-est performance in the document level.