Research On Sentiment Analysis Of Chinese Text

With the rapid development of Internet technology, the network is not only an important source of information, but also becomes the best platform for people to express their personal opinion and share their own experience. "Reading network" gradually turned into "interactive network". The network is filled with a large number of the subjective text which are released by the users. These subjective text may be the comments of a product or a service, the public opinions on a news event or the national policy and so on. If we can make full use of these data, it can make contributions to many application fields, such as to help the government to monitor network public opinions or help enterprises to do market analysis. However, these data increase at a speed of exponential every day, it is obviously not feasible to analyze them in an artificial way, which requires a lot of manpower and time. How to get the information what we need through effective means from the huge number of the network resource? Sentiment analysis technology arises at the historic moment.Sentiment analysis is one of the hot research topics in computer area, it aims to automatically analyze the attitude tendency from the text through the computer technology, and make the emotional classification of the text. After investigating the research status of sentiment analysis, this paper do some research on sentiment analysis in view of the insufficiency of it. The main research contents and innovation points include:(1) In view of the popular use of the network-based language, this paper put forward a method based on word embedding to judge the emotion of words. By means of the language model, word embedding were used to compute the semantic similarity between words, and the emotional polarity of the target word was gained. This method can solve the problem that the traditional emotional dictionary is not perfect and hard to extend.(2) A method of sentiment analysis of Chinese sentences which is based on syntax tree structure and semantic rules is proposed, and it combines the method of sentiment analysis of words based on word embedding. On the one hand, it fully takes into account the syntax and grammar information of the sentence. On the other hand, for the phenomenon of the sentence emotional polarity transfer, some sentiment semantic rules were defined which combined with some features such as negative words, degree adverbs and transitional words. This method effectively improve the performance of the sentence emotional tendency analysis.(3) In consideration of the emotional diversification of Micro-Blog, this paper adopt the method based on Recursive neural network to analyze the Micro-Blog topic comments. Through the Softmax classifier, sentences are classified to different emotions, including happiness, anger, sadness, fear, evil and amaze. Then, according to the degree adverbs of the sentences, different sentences will have different weights to compute the tendency of the topic and predict its development trend.