Research On Sentiment Orientation Analysis Of Online Public Opinion Topic

With the rapid development of Internet, tremendous freedom is provided for people to express personal opinions. By means of Internet, people can publish their attitudes and opinions for the public affairs concerned about. With the quick increasing amount of subjective reviews focused on the hot or focal events on the Internet, the research about online public opinion analysis is springing up for the purpose of analyzing state-and-trend and evaluating threat level of online public opinion, and becomes a hot research topic in the area of intelligent information processing. As one of kernel technologies about online public opinion analysis, sentiment orientation analysis deals with the reviews of online public opinion topic, and gets whole sentiment orientation of the topic by the broad commonalty. So it can provide the services for related departments to understand people's opinions and make scientific decisions. This dissertation makes deep research on the technology of sentiment orientation analysis of online public opinion topic, including: the frame design of sentiment orientation analysis, semantic orientation identification of Chinese words, opinion topic identification, and sentiment orientation identification of online public opinion topic. Some major contributions are listed as follows:(1) A frame of sentiment orientation analysis is designed according to the characteristics of online public opinion topic. Firstly, the analysis of traditional sentiment classification is made through experiments when chosen to process the reviews of online public opinion topic. Then the advantage and disadvantage of sentiment classification is analyzed, and a conclusion is reached, that is, the method of sentiment classification is not appropriate for sentiment orientation analysis of online public opinion topic. Finally, a frame of sentiment orientation analysis appropriate for online public opinion topic is designed based on analysis above.(2) A method of Chinese semantic orientation identification based on semantic orientation similarity is proposed. Firstly, the ability of each word that expresses sentiment is evaluated using the phrase modes while candidate words are picked out, so the influence of objective words to the identification precision is avoided, and the computation is reduced effectively. Then the concept of semantic orientation similarity is introduced, and its value is computed from four connection relations between words. Finally, a sentiment dictionary is constructed after semantic orientation of all words is identified. Experimental results show that this method outperforms SO-PMI method and HowNet-based method, which are also prone to extend sentiment dictionary.(3) A method of opinion topic identification based on association analysis is presented based on the concept expansion of opinion topic. According to the mode of opinion topic generation in network reviews, the new method adopts four steps to finish opinion topic identification, i.e. inner topic word identification, inner topic constitution, outer topic identification and topic organization. The difficulty to analyze opinion topic of single review is avoided, and opinion topic can be identified with integrated semantic information. Experimental results show that opinion topic of the reviews about online public opinion topic can be identified effectively through this method.(4) A method of sentiment orientation identification of online public opinion topic is introduced according to the commonness in the forms of the reviews about online public opinion topic. On the basis of sentiment dictionary construction and opinion topic identification, the new method identifies the whole sentiment orientation of online public opinion topic. Firstly, syntax analysis is made to compute the orientation values of sentiment words in the context. Then the process of semantic mode matching to the sentences in the reviews is realized, and sentiment orientation of the sentences, which have simpler structure and can reflect the opinions of reviewers, is identified. Finally, the orientation values can be modified through the way of clustering method to identify the reviews with similar intepreted languages, and then sentiment orientation of the topic is estimated. Experimental results show that sentiment orientation about online public opinion topic can be identified effectively through this method.