Research On Personalized Recommendation System Based On Relevance Characteristics

Faced with the geometric growth of product data in the Internet era,how to free consumers from the information ocean has gradually become the focus of the various walks of life.Personalized recommendation technology develops rapidly under the impetus of this demand.Association rules,as an important part of personalized recommendation technology,can better cope with the problem of "data explosion".In the process of researching the related knowledge of association rules,this thesis finds that there are some drawbacks in the practical application of association rules,such as low efficiency and difficult selection of rule results.In view of the above shortcomings,the main research contents of this paper can be summarized as follows:(1)In view of the low efficiency of association rules mining,This paper considers that the fuzzy clustering algorithm can aggregate the products with associated characteristics to the greatest extent.Thus the data amount of association rule mining only needs to be carried out in the target user's belonging clusters based on the historical behavior information of the target user.In this paper,a fuzzy clustering algorithm is introduced to preprocess the input data of association rule algorithm.The pretreatment process is improved by combining subtractive clustering and Mahalanobis distance,which reduces the time-consuming of traditional association rule algorithm,and solves the problem of random selection of initial clustering center and effective selection of distance measure in traditional fuzzy clustering algorithm.This paper uses the relevant datasets in UCI to carry out experimental verifications from different angles.Through the final experimental results,It is clear that the improved fuzzy clustering algorithm has good clustering effect.(2)Aiming at the difficulty of optimum selection of association rules,a score prediction method of rule results combining user's scoring is proposed.The method uses the usage of similar neighbors in collaborative filtering recommendation algorithm and takes user's scoring information as the main reference of selection tendency.This method can complete the second effective mining of recommendation results.The validation process of the method is based on the data in MovieLens.Through the analysis of the experimental results,The method do make full use of the rule information,and achieves the purpose of improving user satisfaction at the same time.(3)The Thesis Designs and implements the personalized recommendation system platform for e-commerce.The platform uses B/S structure and the logical strategy of separating the front-end function from the back-end function structure while linking the content,On the basis of simulating whole shopping process of Taobao Mall,the platform embeds the algorithm studied in this paper into the system model to complete the personalized recommendation function for target users.