Research On Sentiment Analysis Methods Of Social Network Text

With the explosive proliferation of online social network platform, such as domestic Weibo and Wechat, foreign Twitter and Facebook, more and more people are increasingly inclined to share their opinions and emotions on the social network platform. At the same time, social network platform generates a massive amount of sentiment-carrying social network text. Mining and analyzing the sentiment of these social network text, can gain valuable insight into the opinions of individuals, which are very important significance for optimizing personalized recommendation, and better monitoring public opinion.However, social network text with colloquial, time-sensitive and networking features, brings tremendous challenges to the traditional opinion mining, sentiment analysis and so on. In addition to social context, retweet behavior relationship data is also marked feature of social network text. So, this paper combined with graph ranking model and made use of behavior relationship data to analyze sentiment of microblog. But, analyzing the sentiment of social network text, there are also some other problems, in which most significant problem is distribution imbalance of corpus. So, this paper made further research, and then proposed a cross-domain sentiment analysis method based on extraction of sentiment key sentence.On the one hand, according to those typical features of social network text, this paper presented a sentiment analysis method based on graph ranking for microblog. First, this method utilized rich emoticons in microblog to unsupervised annotate a part of corpus, which could make up distribution imbalance of corpus; Then, for the sparse features of social network text such as microblog, this paper proposed a feature extraction method based restricted Boltzmann machine; At last, taking advantage of context relationship which are retweet and different microblog published by the same user, construct microblog relationship graph and combine graph ranking model to analyze microblog sentiment. This model made full use of the text and non-text behavior relationship data, could analyze microblog sentiment well.On the other hand, considering the distribution imbalance of social network text, this paper proposed a cross-domain sentiment analysis method based on extraction of sentiment key sentence. At first, based on the observation that not every part of the text is equally informative for inferring the sentiment orientation of the whole text, the concept of sentiment key sentence was defined. Then construct heuristic rules and combine with machine learning to extract sentiment key sentence of web review. Naturally the data is divided into key and detail views. Finally, utilize ensemble learning to integration these two views effectively, and then to some extent, solve the problem distribution imbalance of corpus bought by the changes of social network text in period of rapid development of social network.