Text Semantic Orientation Classification Algorithms For Network Public Opinion

Public opinion is to point to the attitudes on the social management and social and political held by our people around all kinds of social events and the occurrences, developments and changes. In recent years, the web has become one of the main carrier of social public opinion, and due to the growth of the Internet web pages is in a exponentially way, network public opinion has gradually become a powerful voice focused by government and enterprises. Seize the guidance of network public opinion can timely understanding the attitude and orientation of the people for a particular policy or a event, then producers can make improvements according to the users’ requirements in network public opinion. So gathering network public opinion orientation is now a hot problem in the natural language processing. Network public opinion orientation analysis needed a computer natural language processing technology efficiently and accurately to identify the information covered in the websites, analysis and classification the public opinion rapidly and effectively.Common emotional tendency analysis algorithm is needed to rely on the network users who choose such as agree or oppose in simple voting forms to reflect the attitude and orientation of events or products. However today, for the web users, however, this is far from enough, people’s comments in the form of language, such as microblog. Its subtle differences in semantics and contextual information content analysis become a challenging problem. This system is established for network public opinion text semantic orientation analysis model, in order to improve the accuracy of the classification of semantic orientation. The main research contents include:(1) It builds an emotional dictionary based on HowNet and ontology emotional words in Chinese. This dictionary uses positive and negative emotional vocabulary of HowNet emotional dictionary, and use the polarity annotations of the ontology emotional words in Chinese for reference, and artificial add network commonly used words, formed a small emotional dictionary for text preprocessing.(2) It proposed a keyword weight calculation method for network public opinion. The method adds position weight, emotional weight and balance variables on the traditional TF-IDF weight algorithm. It makes the algorithm more suitable for the emotional tendency analysis. In this experiment, the improved calculation method achieved better classification effect.(3) It proposed a public opinion orientation analysis algorithm based on Hidden Markov Model. Through the text of key sequence as the observed sequence of the HMM model, use the decoding algorithm get the optimal sequence corresponding to the observed sequence, calculate the final emotional tendencies. This method is compared with the classical Naive Byes and SVM classification algorithm, and it has certain increase in performance.