Research On Fine-grained Sentiment Analysis Base On Chinese Micro-blog

Micro-blog as a new kind of social networking platform, which can publish short but informative text whenever and wherever and this information on behalf of the user views or preferences, as well as for monitoring public opinion mining, marketing, rumor control are of great significance. At present, the text sentiment analysis focuses on binary, that is, positive and negative. Meanwhile, microblogging text have spoken of serious issues such as data sparse for deep mining and analysis is not enough. So emotional fine-grained discrimination is very necessary. This article includes the following aspects:First,We designed a model of the fine-grained discrim ination method based on the emotional elements for a significant emotional words microblogging.According dependency parsing, analyzing the dependencies between words in a sentence, find the word level emotional impact factor: adverbs, negative words and other elements of the model and the establishment of affective and emotional elements of the model to calculate the score. For each particle size were s ummed, the maximum value corresponding to the fine-grained as microblogging emotional strategy.Second, This paper designed a multi-feature SVM microblogging emotion fine-grained discrimination method for no obvious emotion word for the Micro-blog, In the feature selection stage improvement of traditional CHI feature selection instability, considering only the number of micro-Bo without considering word frequency proposed CHI- TFIDF feature extraction method, word frequency, word frequency inverse document introduced into the CHI algorithm. The experiment proved that the improved algorithm CHI CHI stability and effectiveness than the traditional strong. The microblogging sentence structure and semantic features such as the introduction of SVM algorithm is trained sentiment classification.The methods this paper provided herein comparing machine learning algorithms SVM test com paring the methods provided herein is substantially higher than the indicators of SVM. We attend the third natural language by the China Computer Society and sponsored by the Chinese Society for processing calculations released Chines e microblogging emotion recognition and classification evaluation mission, get good results in the second set of results seven five teams submitted.