Research On Microblog Sentiment Analysis Technology Based The LDA Model

With the popularity of internet, internet has become a very important way for the public to get information, publish information and transfer information. Meanwhile, the network has appeared so much subjective text, such as forum posts, microblog and so on. Through those subject words, people can be free to express their feelings and thoughts. Microblog is one of the most widely popular network applications in the world. People have gradually accustomed to obtain or exchange information and express emotion in social network by microblog. The huge amount of microblog users and the microblog data have great potential commercial value and social value. Accordingly, research on microblog have drawn more and more attention, microblog sentiment analysis is just one of them. Microblog sentiment analysis aims to judge sentiment tendency of the text by analysis and mining the subjective information.Application of microblog sentiment analysis technology is very extensive. Sentiment analysis can be used for public opinion monitor, product recommendations, information prediction and so on. Microblog sentiment analysis has been becoming a hot research direction and it also faces the difficulty due to the short text’s conciseness. So it is worthy of further study.Microblog sentiment analysis technology can be regarded as a classification problem, focusing on the discrimination on two emotional tendencies: positive, negative. The classification method of microblog sentiment analysis technology based on emotional word and machine learning. People have made many achievements about microblog sentiment analysis, but there still exists some shortcomings. Such as the accuracy is not high enough and strong dependence on the labeled corpus. In this paper, we want to find an efficient generic method. Firstly, we summarize and analyse the concept and basic methods about sentiment analysis, putting forward the method of semi-supervised training in view of the current situation of shortage of tagging corpus. Secondly, we make full use of the overall information of microblog through the method of LDA(Latent Dirichlet Allocation) topic model. Finally, we proposed multiple approaches to improve sentiment analysis by building a LDA based sentiment mixture model in a semi- supervised training framework. The proposed microblog sentiment analysis model outperforms the universal model.