Product Attribute Mining And Application

E-commerce based on the Internet lead the new trend in the consumption.Making a sudden growth of product reviews data on the internet. Because these comment of consumers contains information that how users evaluate to the product function attributes, component properties and other valuable information. So the product attribute mining technology arises at the historic moment. At present,there are two product attributes mining technology:artificial definition and automatic identification of machine. The method of artificial definition is that domain experts summarizes the properties.The method is accurate, but it can only extract product attributes in the field, without generality,and strong portability. Unsupervised machine automatic identification method, mainly process technology with natural language,But if this technology will increase the redundancy of product attributes and reduce accuracy.In view of the insufficiency of above research, with product reviews information in e-commerce website as the research object, this article put forward and realized a kind of product attributes mining method based on part-of-speech template, and improved the method for complex view.First, Crawl consumers’ product reviews, establish database of the original comment, and get data corpusd digged from product attribute,through the analysis of the rule of e-commerce sites. Second, Pretreatment product reviews through the clauses processing and part-of-speech tagging. Third, analysis deeply the product reviews that has been completed the part of speech sequence annotation, classified characteristics with part of speech template, end up with "candidate product reviews" and "no attribute comments","no opinion","the candidate label" four types of parts of speech sequence features; Fourth, through the window near principle, compare attribute value from the attribute distance vector size, identify the attributes and attribute values of the candidate product reviews and get the correct collocation relations; Fifth, calculate confidence coefficient to the identified properties, the higher the confidence level, the greater of the accuracy of the property right is; Sixth, by setting the stop words, further reduce the candidate product attribute redundancy, improve the accuracy of attribute recognition.In this paper, the study identify the application of the product attributes, First of all, from the perspective of consumers, by identifying the attributes and the proportion, can get hot property of the product that attach most consumers attention, reached the purpose of helping potential consumers make buying decisions. Secondly, from the perspective of enterprise, through the comments from consumers already clinch a deal identify the product attributes, clear advantages and disadvantages of this kind of products, to help enterprises to improve products, improve quality and timely repair for the potential customer relationship. Again, from the perspective of operators, through identify the attributes of the third-party e-commerce sites, improve reliance on e-commerce sites and increase the e-commerce sites of users.The main research contributions of this paper contains:First, by improving the extraction classifier structure, compared with the method based on nouns part-of-speech template, Improve the effective degree of the candidate product. Second, based on data packets COAE2009provided, to improve speech template window threshold setting, to reduce the redundancy of candidate product attribute; Third, add the new confidence calculation module, confidence calculation calculate confidence coefficient according to the attributes in the text context information, such as frequency characteristics, to improve the accuracy of product attribute recognition; Fourth research into the application of identified product attributes improve the value of research of the excavated product attributes.