Recommending Commodities Based On Text Mining

E-commerce sites provide the evaluation services of the purchase of goods to online shopping users. These comments about shopping can provide some reference when users decide to buy something.In this paper, we analyzed and dealt with the comment text provided by users, and got the useful information from unstructured comment text by using the technologies of text mining. That useful information are used to sort the commodities that the users have evaluated, and recommend the commodities as purchasing according to the priority sequence, provide valuable reference information to buyers and avoid users waste time to look up and compare.In this paper, the proposed method dealt with the user comment text about the electronic commerce website with some kind of pre-processing, such as arrangement, word segmentation and removing stop words. The Character-Words are extracted by the way of doing word frequency statistics of comment text, and frequency inverse document frequency are applied to improve word feature category distinguishing ability. After a large number of electronic commerce comment text of website are collected, through text cleaning up, word segmentation, word frequency count and a series of pretreatment, characteristic word corpus has been constructed.In order to calculate the similarity of the comment text with characteristic word corpus, key word vectors of the comment text are extracted by pre-processing and word frequency count. For word similarity calculation of semantic resources and calculation methods of detailed analysis, using of synonymy thesaurus based word similarity calculation method for keyword vector and characteristics of thesaurus similarity calculation. According to the results of the similarity calculation of the user comments with characteristic word, rank of the purchase products and recommendation are carried out.Finally, according to the general analysis of the e-commerce website, the database of the commodity recommendation system is designed and the experiment program is completed. The collected user comments are used to complete the recommendation experiment, and the experimental results were investigated and analyzed.