Research And Analysis On Opinion Mining In A Specific Domain

With the rapid development of Internet technology, the web has become animportant place for customs to air their’s opinions and other information. Although alarge amount of online information can bring us great convenience, it also makes us ata loss.Therefore how to collect and summarize these useful opinion informationeffectively from these vast information becomes an important research subject.Opinion Mining, also called sentiment orientation analysis, is aim to abstract thesubjective information and analysis sentiment orientation automatically from textswhich contain emotions. Opinion Mining is a new research area relating to naturallanguage processing、information extraction、 information filtering、automatic textcategorization、 text mining and other fields. The application prospect of opinionmining is very considerable. It can be applied to many areas of national economy andpeople’s livelihood, such as e-commerce recommendation system, informationsecurity filtration systems.In this paper, we build an opinion mining system in a specific domain andanalyze the key tasks of opinion mining. Our research consists of the construction ofvarious dictionaries、abstracting the subjective information、topic extraction、sentiment analysis and so on.The construction of dictionaries is a fundamental job of opinion mining. Webuild our basic Chinese polarity lexicon base on some existing Chinese and Englishsentiment lexicons. In the same while,we also build domain lexicon、domain polaritylexicon、dynamic polarity lexicon、network polarity lexicon、privative lexicon、degreeadverb lexicon. In basic polarity, we not only assign the sentiment orientation but alsolabel sentiment intensity for each sentiment level. The construction of dictionaries haslaid the foundation for the following research tasks.The extraction of the Chinese subjective information is also an importantfundamental job. In this paper, we separately design the method of subjective cluesand the method of the subjective patterns, and finally combine the both to extract thesubjective sentences. We compare the three methods on the same test data. Theexperiment results show that the method of the subjective clues and patterns is bothreasonable and effective.Topic extraction is a critical task in opinion mining.We design two methods tostudy it, topic extraction base on syntax analysis and base on opinion words, and thenanalyze the advantages and disadvantages of two methods. Finally, we do a researchto identify the implicit topics. It can be proved by the experience that the accuracy and recall rate in the whole topic extraction module will improve after adding the implicittopic identification.Sentiment analysis is also a critical task. We study the context polarity intensityof sentiment words which modify the topic after analyzing the topic sentimentorientation. We also design some phrase modes to calculate the context polarityintensity of sentiment words. This method is reasonable via the experience.After all, we realize an opinion mining prototypical system in a specific domainand point out the research ways in the future.