Research On Opinion Mining And Sentiment Analysis For Chinese Micro-blog

With the rapid development of the internet, especially the growing popularity of Web 2.0 technology, the majority of internet users have become content producers. Since Chinese micro-blog was borned in 2009, users’ habits have been changed, which brings lots of new problems, in which opinion mining and sentiment analysis of Chinese micro-blog gradually becomes a hot research topic and has gained more and more attention of many experts and scholars. Meanwhile, with the continuous development of natural language processing technology, machine learning technology and big data technology, opinion mining and sentiment analysis of Chinese micro-blog is pushed to a new climax.Chinese micro-blog brings a lot of new application requirements, such as opinion target extraction and opinion analysis, sentiment analysis of micro-blog, and opinion information retrieval of micro-blog. However, because of many characteristics of Chinese micro-blog, such as short texts, colloquial content, incorrect methods of writing, disorderly sentence structure and having retweeting, the traditional opinion mining and sentiment analysis technology cann’t satisfy the requirements of micro-blog data processing. In order to satisfy these new requirements, we mainly research emotional intensity quantitative calculation, emotional elements collocation extraction, opinion sentiment analysis and opinion retrieval, and explore new method to improve the accuracy and practicability of opinion mining and sentiment analysis of Chinese micro-blog. The major research work and its contributions are as follows:(1) Through the analysis of emotional intensity of sentiment words in Chinese micro-blog, we research emotional intensity quantitative calculation technology and put forward emotional intensity quantitative calculation based on normal distribution for basic sentiment words while classification-based is for composite sentiment words. The experimental results show that the performance of emotional intensity quantitative calculation is greatly improved and the calculation results are closer to people’s subjective judgement.(2) Through analyzing the features of opinion target and opinion-bearing word in Chinese micro-blog, we extract emotional elements collocation. Using the lexical, syntactic, semantic and positional features, we propose a co-extracting model for emotional elements collocation based on multi-features. Furthermore, implicit opinion target is identified via innovative use of forwarding relation between micro-blogs, and finally implicit emotional elements collocation is extracted. The experimental results show that emotional elements collocation extraction gets better result.(3) Many characteristics of Chinese micro-blog make opinion orientation identification exist many difficulties. According to micro-blogs’ opinion expression, we design emotional elements collocation rules and use them to achieve micro-blogs’ opinion orientation. What’s more, we mine more characteristics of Chinese micro-blog, and further explore opinion orientation identification based on the combination of support vector machine and emotional elements collocation rules. Finally we give optimization method based on forwarding relation and relaxation labeling technology. The experimental results show that the performance of opinion orientation identification is greatly improved.(4) We design and achieve opinion retrieval system for Chinese micro-blog. Emotional intensity quantitative calculation, emotional elements collocation extraction and opinion orientation judgment are then integrated into opinion retrieval to provide services. The research results are applied to pre-research project, and make the research better combined with practical application.