Research On Comparative Identification And Comparative Opinion Element Extraction

With the rise of web2.0, forums, microblog and paste provide people with a platform to express their views and perspectives. Therefore, a large number of comments emerge in the network. Comparative which is a useful convincing expression, are often used to compare two things or more things, to express their preference for different things in the comments. The study about comparative sentences has practical significance. For example, businesses can compare similar products in the market, so as to enhance the competitiveness of their own products, and provide decision support for ordinary consumers to buy goods. In recent years, utilize Natural Language Processing techniques to identify comparative sentences from the review, to extracte comparative relationship and to determine comparative sentences tendency is increasingly favored by the researchers.This thesis studies comparative sentences identification, comparative relation extraction and the comparative opinion mining. The major work of this thesis include:(1) Chinese Comparative Sentences Identification Method Based on Sequential PatternsIn this thesis, sequential pattern mining algorithm is used for acquiring comparison pattern. In order to enhance the performance of algorithm, sequential pattern mining algorithm is improved. Some small values are set for minimum supports of items (noun) and comparison feature words and part of speech, the minimum support of the other items chooses a larger value between a multiple of item support multiplied and1/N (N is the size of the sequence set). Moreover, comparative sentences are identified by obtained pattern based on sequential pattern mining algorithm in directly match way. The experiment result indicates that the proposed SeqPattMine method is feasible.(2) Comparative Sentences Relation Extraction Based Chinese FrameNet Semantic Role LabelBase on the comparative pattern set, product name set,attribute set and other resources,combining with the phenomenon of comparative sentences in the corpus, label the Chinese FrameNet semantic role, thereby extracte comparative relations. Finally, the experiment results prove that the proposed method is effective.(3) Compare Opinion MiningFour different comparative opinion mining methods are constructed according to four different types of comparative sentences. In the constitution processing, sentiment word set and domain properties collocation set are used. The experimental results are desirable.