Research And Application Of Product Recommendation Algorithm Based On Implicit Feedback

Nowadays,physical stores have entered the cold winter,but online shopping is developing rapidly.Shopping websites provide a variety of goods,which greatly meet the diverse needs of people.However,with the increasing number of products in the website,people increasingly feel that online shopping is time-consuming and laborious.This is due to the "information overload".With the recommendation system,this problem can be effectively solved.The purpose of recommender system is to recommend items for users by analyzing their interests and hobbies.In common collaborative filtering recommendations,users need to give "explicit" rating and evaluation of projects to reflect the degree of preference.Because it is difficult to get users' scores in the field of commodity recommendation,it is difficult to apply.However,through the analysis of implicit feedback data,that is,the behavior data generated in the process of continuous interaction between users and shopping system,users' preferences for goods can also be collected.Therefore,this paper will use the implicit feedback data generated by users' click,collection,shopping cart and payment in the process of online shopping to realize the function of product recommendation for users.The main research work of this' paper is as follows:(1)In view of the problem that the implicit feedback data of users can not accurately measure the degree of users' interest and preference,this paper studies how to transform the implicit feedback data into the scoring data of commodities through the calculation formula to form the user commodity scoring matrix,so as to apply the collaborative filtering recommendation algorithm based on users to realize recommendation.(2)In order to solve the problem of large user space and time-consuming and sparse matrix when calculating the similarity of the obtained user commodity scoring matrix,a similarity calculation method based on the similarity factor of user behavior is proposed by improving the traditional similarity calculation method.Firstly,K-means clustering is used to cluster the user's identity attributes,and the cluster set of the target user is regarded as the user space range of similarity calculation,which effectively reduces the number of times that need to be compared in similarity calculation;secondly,after the commodity candidate set is obtained by setting the threshold value,the user behavior similarity factor is introduced to calculate the reference score,which is used to fill in the user evaluation It solves the problem that the number of goods in the intersection of scoring goods is less in the sparse matrix in the traditional method,and improves the similarity calculation method.Experiments show that the performance of the proposed algorithm is better.(3)A commodity recommendation system is designed and implemented with Java language,and the algorithm proposed in this paper is put into practice.