The Research Of Personalized Recommendation Methods Based On Category Similarity And Classification

With the development of internet and E-commerce, the recommendationsystem has gradually become an important research field of E-commercetechnology, and attracts many researchers' attention. Collaborative filteringtechnology and category-based recommendation technology are the mainresearch issues in the field of personalized recommendation, and in this paperwe focus our research on these two issues.In the field of collaborative filtering recommendation technology, theproblems that exist in today's collaborative filtering technology have beenanalyzed and points out that with the development of E-commerce, themagnitudes and commodities grow rapidly, which resulted in the extremesparsity of user ration data. For not being taken the category of different itemsinto account, the traditional similarity measure methods work inaccurately inthis situation. Aiming at this problem, a new method has been brought forwardto deal with users rating data in order to improve the veracity of items' similarity.When computing the initial forecast rating of items,a new method is introducedto improve the data density of users rating matrix. And at the final forecastrating of the items, a new method is presented to improve the veracity of usersforecast rating. In order to improve the character of real time, an incrementalcollaborative filtering recommendation algorithm based on category similarity ispresented. The experimental results show that the new methods can efficientlyresolve the existed problem of traditional methods in the condition of theextreme sparsity of user rating data, and the quality of the recommended resultcan also be efficiently improved.In the field of category-based recommendation technology, a newalgorithm is introduced which has a faster speed to receive the sequence accesstransaction set, and then, the user preparatory recommendation model andcurrent input sequence with user character are used to get the user'srecommendation aggregate corresponds to the user recommendation model.Meanwhile, it gives the definition of page interest and calculation method andby the preparatory recommendation aggregate, the final recommendationaggregate to the current user correspond to the users' recommendation modelcan be gotten with better accuracy. At last, the correctness and validity of thesemethods in the paper are validated through the experimentation.