Research On Analyze Product Feature Form Product Reviews Based On Syntactical Parser

Driven by web 2.0 technology, booming e-commerce makes online shopping recently became a kind of fashion. Various forms of e-commerce websites like Network Mall and Group Purchase etc. springing up, a lot people buy goods and published their product reviews on the net. As feedback information of user experience, user product reviews gains more attention and how to analyze and comprehend user comment become the important task of web mining.Product reviews mining is application of Opinion Mining in product reviews, comparing the superiority of differences between products by analyzing the sentiment of the product reviews. Product reviews mining involved in Natural Language Processing, Data Mining and Machine Learning technology. Although some scholars have researched product reviews mining and resolved the problems of polarity classification, them haven't studied the product feature information in further either analyzed the sentence by semantic level, so there are some error in negative sentences and comparative sentences. In accordance with these problems, this paper proposes a new system of product reviews mining based on syntactic analysis.Firstly, this paper add the polar word of Informal Language polar word to polarity dictionary according to the dynamic and exotic characteristics of the language of product review, improve the adaptability of method to Informal Language. Secondly, using the approach of TFIDF normalize the weight of polar word which make of the method of analyzing polar word more scientific and reasonable. Thirdly, proposing the tree structure model of sentences analysis based on the four factors opinion review and judging the polarity of the comparative sentences by analyzing the position of the subject of comparative sentences and the polarity of comparative word. Finally, proposing the basis of bottom-up based on syntactic analysis of product reviews SP-PMI algorithm base on SO-PMI and designing the product feature mining system. Experimental results show that this algorithm improve the precision of the original method of SO-PMI, especially for analyzing the polarity of the comparative sentence, it enhance the recall ratio and the precision nearly to 13%. In addition, the product feature mining system also obtain the comparison of the characteristics of the feature of same class and difference class products, has great enterprise practical value.