Opinion Target Recognition And Orientation Judgment In Opinion Mining

The advent of web2.0makes online reviews grow exponentially. How to get usefulreviews quickly from the mass information is a real problem thar people face. In thiscontext, opinion mining develops rapidly, has become a hot topic and draws more andmore people’s attention. Opinion mining has many tasks and can be dividen intoopinion extraction, opinion classification and opinion retrieval and summarizationaccording to the task level. The extraction of opinion targets belongs to opinionextraction, the orientation judgement belongs to opinion classification and they are twoimportant tasks of opinion mining. Extracting opinion targets and judging the sentimentorientation of reviews by using opinion phrases is this paper’s work.Most researchers use template-based method to extract opinion targets. Theextraction of opinion targets and opinion phrases is an important task in opinion mining.As template-based methods have several disadvantages: too much manual intervention,the lack of template coverage, and having difficulty in identifying opinion target andopinion phrases with long distance, this paper presents a method that could extracttemplates automatically and identify the opinion combination by using the probability.In detail, this paper describes the structure of opinion tagets and opinion phrases byPOS sequences, makes conformity rules according to opinion combinations’ POSstructure characters and gets templates of opnion combinations. This paper makesextraction rules and designs extraction algorithm on basis of the problems that mayencounter during the extraction of opinion combinations. In the algorithm, the distancebetween opinion targets and opinion phrases can be changed. The algorithm uses theco-currence probability of opinion targets and opinion phrases to select opinioncombinations.In order to judge the sentiment orientation, this paper bases on the HowNet lexicon,mines the information of thesauruses and qualifiers and consider the impacts ofsentiment words and qualifiers. In allusion to the problem that the basic method oforientation judgement may misjudge the orientation of sentiment words, this paperproposes to correct the sentiment strength by using thesauruses. After analyzing thestructure of opinion phrases, this paper uses stop word list to eliminate unsentimentwords and judges the opinion orientation by considering the impacts of sentiment wordsand qualifiers.The proposed work has been evaluated on the corpus of COAE2011, comparedwith two baseline methods, and obtains a good result.