Research On Product Fuzzy Recommendation System Based On Online Review

With the rapid development of web2.0, the technology of online shopping is mature, like Amazon, Taobao and so on. More and more consumers choose online shopping because of its convenience. The manufacturer provides products information which includes products pictures and products parameter in web pages. But this products information is not enough for consumers that a lot of useful information is in online customer reviews. Some products attributes only appear in reviews and these attributes are very important for consumer’s shopping decision. Also, a lot of consumer opinion and sentiment information is in the online reviews, if this information can be deeply mined and analyzed, it can help consumer to make right shopping decision. To solve the problem, we use natural language technology and fuzzy computing to build the fuzzy intelligent recommendation system which base on consumer online reviews.A lot of text processing work is demanded in the recommendation system. The text processing needs nature language processing and text mining technology. In the system, mining products, extracting sentiment words and analyzing the user fuzzy input text the first step is segmentation. Besides, the cluster analysis of reviews sentiment tendencies can help the system to quickly judge the product quality. The system uses Maximum Entropy Model to classify the online reviews (positive or negative). The balance of training corpus can influence the precision of classifier, so the corpus’s scale is very important.The search engine of shopping, like Google shopping, the products ranking is according to the price and sales. But our recommendation system ranks the products according to the consumer opinion which contains in online reviews. After mining the product attributes, the system uses template to extract the sentiment words. We construct the database of sentiment words and the database of fuzzy inference rules. Based on these databases, the sentiment words can be computed by the fuzzy operation. The products are ranked by the result of fuzzy inference. This rank method of system considers consumers’personalization requirements and can be seen as the extension of traditional recommendation method.The system also has the fuzzy recommendation engine which can understand users’ sentiment. The engine can analysis the users’fuzzy requirements and transfers the requirements to the fuzzy function. After computing the fuzzy nearness and fuzzy inference, the system will list some products which are most satisfied the consumers’requirements. The experiment shows that the system has well recommendation result.