Research Of Emotional Analysis Methods Based On Machine Learning

With the popularity of the Internet,more and more people published some opinions of their own views through the microblog,post bar and BBS,these data has a huge social and economic benefits and produce the Internet public opinion,so it need the Internet public opinion analysis to monitoring and manage.The key of public opinion analysis is sentiment analysis.In order to achieve the massive Internet text sentiment analysis,the researchers use the methods of statistical machine learning and neural network method to analysis.Now there are less text tendency analyses of short essay,and on the sentence structure Chinese is more complex than English,so it is more difficult to analysis the long text.In this thesis,the author compared the differences between the sentiment analysis based on emotional dictionary and machine learning,then studied emotional dictionary extended algorithm,and invented a kind of based on machine learning long text text sentiment analysis algorithm of CRF.The main content of the thesis is as follows:1.Under a detailed study of text preprocessing,the author built the emotional dictionary and compared the current several short text sentiment analysis scheme,such as emotional dictionary,support vector machine and naive bayes classifier,and analyzed the respective advantages and disadvantages and application scenarios.2.Sentiment analysis algorithm based on emotional dictionary is often used,but all the Chinese emotional dictionaries are poor,so the author invented the emotional dictionary extended algorithm with stanford parsers and semantic orientation pointwise mutual information.3.For the long text,the author invented a kind of based on machine learning long text text sentiment analysis algorithm of CRF,through machine learning conditional random field algorithm to extract the core words,predicted the final emotional tendencies by weight synthesis algorithm.Through a large number of experimental,this algorithm achieved good results.