Statistics And Study On The Method Of Combining Chinese Opinion Extraction Rules

With the rapid development and popularization of the Internet technology,especially the rise of Web2.0, online shopping, microblogging, blog and BBS forumsand other emerging Internet applications is becoming the hot spot of interest, commentsonline shoppingblog, microblogging, and views on the BBS is also increasing.Thesecomments and opinion information contains a wealth of value. Government departmentsunderstand the views on one of the policies and regulations or events through thenetwork reviews, and timely scientific decision-making.Commercial organizationsmonitor customer feedback and market trends through the investigation of networkproduct reviews, and thus improve their services and products or to take more effectiveand targeted marketing strategy.Consumers can also browse the comments of others ona product network to assess their purchasing decisions. Therefore, how effectiveprocessing and the analysis of subjectivity is one of the network information processingproblems to be solved. Opinion Mining research is meet this application to develop andhas become one of the focuses of current research in the field of natural languageprocessing (NLP). Opinion mining consists of three contents, namely subjectivityidentification, opinion polarity classification and opinion extraction. This paper studiesopinion extraction relevant aspects of the problem.In this paper, a method of combining statistical and rule studies evaluation object inthe Chinese subject characteristic and evaluation object-sentiment word problem ofrelationship extraction. specifically, a study from the following aspects.First, the paper first introduces the basic concepts of opinion mining and itssub-tasks, the level and focus of the research, Then introducing the Chinese opinionmining research status, as well as opinion mining research status at home and broad.Secondly, the evaluation object extraction in opinion sentence is one of the keyissues of fine grit word-level sentiment analysis to the study. In order to improve theperformance of the evaluation object extraction, we utilize conditional random fieldmodel, the maximum entropy model, support vector machine three model to synthesizeterm information, frequency information, dependencies and other information, and emphasis to study the different features and different models for the influence of theevaluation object extraction.The experimental results show that after comparison of the three models in theintroduction of various aspects of feature information conditional random is moresuitable for the identification of the evaluation object.Finally, combining statistical and rule-based approach to research evaluation object,sentiment word relationship. First apply the t-test and rules to extract the evaluationobject, the emotional words and then use the dictionary, and finally use the mutualinformation method to extend the evaluation object, the emotional words. And we useChapter III’s evaluation object identification’s best results and the nearest matchingmethods and standards for evaluation object data using the same method to compare.The experimental results show that: the proposed method on the evaluation objectemotional words relationship between the extraction performance is effective.