Research On Analysis Of Emotional Tendency Based On Tieba Text

With the rapid development of global future network development in the 5G field and the next generation of innovative networks NovoNet cloud,a large number of data containing subjective ideas are required.To interact with each other in real time,people are prefer to write some personal opinion briefly on social media.At present,Microblog,Wechat,Baidu Post Bar are the widely used for communicating with each other.In particular,Baidu Post Bar is a common form of paste bar accepted by numerous Chinese netizens.Through the Baidu post bar,some information are used to analysis the emotional state,and achieve public opinion monitoring,consider multi-faceted relationships.With the prosperity of social culture in the society,the text of emotional analysis technology has a broad application prospects.In this thesis,the NLP & CC2014 corpus is widely used in the field of Chinese emotional analysis,and adopted to revise and expand the post-bar dictionary.We carry out the text-based emotion judgment and emotion classification experiments based on the dependent syntax to verify the validity of the proposed dictionary.The specific work of this thesis is summarized as follows:1.A new emotional dictionary which contains a total of 34,144 emotional words is constructed:1)The strength values of 27466 words in the empirical language lexical lexicon of Dalian Polytechnic are revised at some degree,which solves the problem that when the degree words and emotional words appear at the same time.Meanwhile,the emotional intensity of the phrase cannot be correctly reflected in the problem.2)A total of 397 emotional words are selected from the 4000 microblogs of NLP &CC2014 microblogging emotional reference corpus,and the emotional category and emotional intensity are marked with reference to the ontology library.3)We use point mutual information technology(PMI)method to label the emotional category and intensity of the Taiwan University Simplified Chinese emotional dictionary(NTUSD)and HowNet dictionary.Accordingly,4890 emotional words are added.4)Combined with the features selection algorithm and the sequential floating forward algorithm,we propose a new text feature selection algorithm termed SFFS-MI.Experiments results show that our algorithm performance is effective for the text analysis.2.In the building of Baidu Post Bar database,Support Vector Machine(SVM)technique is used for emotion classification experiment,the results show that: 1)The experiments are conducted on education,sports,culture,entertainment and so on.Meanwhile,we also conduct experiments through Decision Tree and K nearest neighbor algorithm.Compared with the Decision Tree and K nearest neighbor algorithm,SVM obtain the best experimental results.3.In addition,to improve the recognition efficiency,we conduct experiments on the improved feature selection algorithm based on mutual information.And we obtain better recognition performance among the different classifiers for emotion classification.