Sentiment Analysis Based On DNN And Vector Space Model For Chinese Micro-Blog

With the rapid development of Internet,Chinese micro-blog which is a new social networking platform,has gradually penetrated into all aspects of social life.Chinese Micro-blog text messages usually contain strong subjective attitudes and emotional tendencies of publisher and in a very short period of time,a large number of Chinese micro-blog text can be aggregated.Effective sentiment analysis for Chinese micro-blog text has great commercial value and social value.In this thesis,analyzing deeply in Chinese language style of the micro-blog,we study Chinese micro-blog sentiment analysis based on DNN and Vector Space Model.our main researches is as follows:Study in-depth automatic extraction technology for micro-blog abstract emotional features.In this thesis,based on the traditional text representation model,we use DNN algorithm to extract automatically abstract emotion features.Combined with short concise text of micro-blog,DNN is constructed by SAE in this thesis.In order to more fully and effectively represent emotional information of micro-blog text during construction of the vector space,this thesis introduces the emotional factor and the structure factor to improve the feature selection methods base on information gain and introduces the positional information of feature words to improve the weight calculation method based on TF-IDF.Study in-depth emotional classification for micro-blog text.In the process of sentiment classification for Chinese micro-blog text,Estimated emotion class,which micro-blog text belongs to,will be effectively improved by the relationship between micro-blog text and specific emotion category.This thesis uses conceptual models for micro-blog to express emotional categories.It proposed to the Expand Space Algorithm(ESA)based on conceptual model.ESA effectively solutes the problem which the traditional text representation model can't be effectively stored the relationship between micro-blog text and specific emotional categories.For the selection problem of conceptual model,this thesis presents a conceptual model selection method based on the spatial density and exclusive power of emotion.Experimental results show that the relevant algorithms,which are used to automatically extract micro-blog abstract emotional features based on DNN,are effective.Using abstract emotion feature vector to repress micro-blog text can effectively improve estimated effect of emotional category for the micro-blog text.Accuracy,recall and F-score have been improved to a certain extent.Experimental results show that in the recognition of emotional category for micro-blog text,the ESA based on conceptual model is valid.