Research On Sentiment Orientation Clustering For Chinese Text Comment

With the development of the Internet and E-Commerce, especially the rapid expansion of Mobile Internet, the amount of information in the Internet is increasing quickly. Consumers will be confronted with great difficulties when they want to find the information of the product which they need. Therefore, there are very important theoretical significance and practical values for consumers and producer of research on using computer technology to analyze, conclude and handle subjective texts with sentiment.Clustering analysis is a technique for statistical data analysis, and applied in many domains. In text clustering analysis, it’s difficult to choose document representation model.This paper focuses on studying sentiment orientation clustering problem based on Chinese text comments written by auto consumers. The main content is as follows.1. Comment match, which is composed of comment target and evaluation word, is very important in clustering analysis for text. Now methods for comment target extraction extract only noun and noun phrase, and they won’t extract verb and verb phrase. This thesis proposes "modified algorithm of comment target extraction based on syntactic analysis". This algorithm selects noun, noun phrase, verb and verb phrase as candidates, and filtrate noises. Experimental results show that precision ratio and recall ratio of the modified algorithm move up slightly. The result of algorithm will be applied to the matching process of comment targets in the bipartite graph-based document representation model proposed by this thesis.2. Vector space model causes a loss to semantic in sentiment orientation clustering. This thesis proposes the bipartite graph-based document representation model. Vertexes can be partite two disjoint kinds, which represent comment targets and comment words. Edge of bipartite graph represent comment matches. In the matching process of comment targets, the "modified algorithm of comment target extraction based on syntactic analysis" is used. Experimental results show that the bipartite graph-based document representation model do better than in sentiment orientation clustering.