A Study On Chinese Microblog Opinion Mining Based On Automobile Domain

Opinion mining has been a hot research filed in natural languageprocessing because of its technology and application value. With thedevelopment of Web2.0and mobile internet, microblog has become a newplatform for publication and dissemination for information. Every day,microblog can produce a large amount of information with a great value.There is a lot of data in microblog involving users’ reviews in someproducts. It is really a valuable technology to extract structuredinformation from the unstructured data in microblog both for the users orbusinesses. So how to use natural language technology effectively to doopinion mining form microblog is worthy of deep study.This paper studies opinion mining on microblog product reviews inautomobile domain. And through the analysis on these texts, this paperpresents an analysis approach of the fine-grained views.This paper divide the microblog opinion mining into three phrases:statement delimitation, subject extraction and sentiment analysis. In thephrase of statement delimitation, we introduce the ideology naturallanguage processing based on naturally annotated Web resources. And wedivide a long sentence into several short sentences which are also calledsmall sentence in Chinese, that is, clauses, to help delimitating thestatement. Then we extract subject from the small sentence. Finally, we usestatistical learning methods adding a topic-based feature to analyze thesentiment of the statements.This paper presents a Chinese microblog fine-grained opinion miningapproach focus on the automobile domain. Our research is based on theactual requirement from some project. We improve the process of general opinion mining based on the study of the microblog texts. In order toenhance the performance of the mining results, we introduce the concept ofsmall sentence and topic related features when analyzing the sentiment ofthe statements. Using decision tree and finding adding emotionalcharacteristics we can achieve78.5%accuracy rate when dividing thecomma into two categories. In sentiment analysis, we use SVM to train thedata and figure out that using the topic related features can obtain betterresult up to81.9%accuracy rate. The experimental results have shown thatour approach is reasonable and effective.