Chinese Opinion Target Extraction And Orientation Analysis Based On Syntactic Dependencies

With the development of the Internet, a large number of texts with opinions and comments appear on the Internet. On the one hand, people browse in other people's comments, on the other hand they share their views on some people or things continually. Sentiment analysis can mine the group view from comment texts on the Internet, so it plays an important guidance role for economic development, political decisions and individual behaviors. Sentiment analysis consists of coarse granularity and fine granularity. Current coarse-grained analysis has achieved good results. However, the effect of the fine-grained sentiment analysis is still not ideal.Opinion target extraction and orientation analysis is one of the most important subtasks of the fine-grained sentiment analysis. And opinion target extraction has been a bottleneck restricting the performance of the task. There are four main methods to extract opinion targets, including extraction based on frequent nouns and noun phrases, extraction by exploiting opinion and target relations, extraction using supervised learning, extraction using topic modeling. Many current methods that exploit opinion and target relations can hardly extract opinion targets accurately to which the opinion words are related, especially opinion targets and opinion words are not in the same sentence. Aiming at the problem, based on using syntactic dependencies between words in the Chinese comment text, semantic role labeling, extraction rules and the searching method are used to improve the performance of sentiment analysis in this thesis. The major work of the thesis is as follows:(1) On the basis of the existing dictionaries, sentiment dictionaries are built for sentiment analysis such as Positive Emotion Dictionary, Negative Emotion Dictionary, Positive Opinion Dictionary, Negative Opinion Dictionary, Opinion Statement Dictionary, Subjunctive Mood Dictionary, Adversative Dictionary, Noun Sentiment Dictionary and so on. These dictionaries are mainly used to deal with interference to sentiment analysis from useless components or non-opinion sentences which just express thoughts or wishes, provide dictionaries to semantic rules and orientation analysis.(2) Based on syntactic dependencies, semantic role labeling and some semantic rules are introduced into sentiment analysis. And attribute-head phrases are used to replace common noun phrases to extract candidate opinion targets. It's supposed to improve the precision of opinion targets and opinion words extraction. These rules mainly consider the effect of Chinese semantic knowledge and common sentence patterns to sentiment analysis. The experimental results show that in the NLP&CC 2013 micro-blog evaluation corpus, the method with semantic rules improve the performance significantly.(3) An opinion target searching method is proposed, which is used to improve the precision of practical opinion target search when the system only extract pronouns or opinion target is not found in syntactic dependencies. With combination of lexical meaning and word similarity computation algorithm, the approach reduces the search range of potential opinion targets in the context. The experimental results show that the method can improve the searching precision of opinion targets.