Sentiment Anlaysis And Its Application On Civil Aviation Related Comments

With the rapid development of microblog,the information flow across the Web spreads traditional media and becomes more influential,so social hot events spreads faster and affects wider scope.In the process of civil aviation,its rapid development unavoidably exposed all sorts of problems,these problems also spread quickly through the means of information,which can easily cause strong repercussions in the community.How to effectively extract the information we wanted for data analysis,and then classify the emotional tendency of social hot events and recognize the opinion target of these hot events in the comments,has become an important research field of the transmission of Internet public's opinion.In this paper,the main research object is weibo comment text.First,we extract the civil aviation related information from massive comment text data,after that we do manual annotation for the emotional tendency of these comments and study emotion classification method based on support vector machine.For the classifier,the features extracted are important for training result.In this paper,we considered two kinds of feature extraction plan,word frequency and words' emotional attribute characteristics,both got a better classification result.Both of the methods' accuracy are above ninety percent.In view of the negative reviews,finding opinion targets has important research significance.This paper studies the classification of opinion target.As the traditional information gain only considers the impact on the characteristics of the category,large number of features in short microblog are potentially to be ignored,thus affecting the classification effect.This paper proposes a method based on the information gain that considers the length of text influence on feature.The paper also studied the Labeled LDA model for words used in the different opinion target.Meanwhile we have studied the relationship of commentary language based on Gibbs sampling results,constructed network for different opinion targets which vividly shows the correlation between polarity words.