The Research Of Sentiment Polarity Discrimination Approach Towards Chinese Microblog Text

Microblog is a social media of sharing real-time short messages,which can provide a platform for netizens to communicate ideas and exchange views.Users can not only browse topic information which they're interested in as an audience,but also provide information to others as a publisher.This new type of social media has been widely accepted by people and achieved explosive growth of user number and information content in the recent ten years.The microblog texts involve in a wide range of topics,offering rich corpus to the research of sentiment analysis.Sentiment polarity discrimination is a specific task of sentiment analysis,it can classify the orientation of the emotional text into positive and negative type though processing and analyzing.Sentiment polarity discrimination of Microblog aims to determine the subjective viewpoint of users' comments towards hot topics,news and products,so that it can be used to monitor public opinion,marketing and so on.The microblogs contains civil aviation speech key words are used to the research of sentiment polarity discrimination.Two algorithms are respectively proposed in this paper,one is a method based on sentiment words and semantic rules,the other is a method based on Adaboost and classifier weighted voting.The first method combines the sentiment lexicon and semantic similarity algorithm to extract sentiment words,so that the words out of the lexicon can be find.Then the semantic rules between sentence and inner-sentence can be used to calculate the sentiment score of Microblog,the weighted summation of emoticon and text turn into the final score.The second method is based on machine learning.Adaboost as an ensemble method is used to promote the weak classifier,and then combine three different classifiers by weighted voting to obtain the final classifier to classify the test set.The corpus of this paper is obtained by microblog crawler,then they are manually annotated to do the experiment.The experiments show that both of the methods get good result.The first method doesn't rely on domain knowledge and has a general applicability to Microblog.The second method promotes the performance of weak classifier and overcomes the defect of single classifier by the fusion of them.