Research On Personalized Recommendation Technology Of Electronic Commerce

With the rapid development of the Internet,the scale of the e-commerce market is continually expanding and all kinds of the goods are too numerous to enumerate.When users face a huge amount of goods,they can't quickly find a satisfactory product.It reduces the satisfaction of the users with the e-commerce sites.Many development of e-commerce recommendation system have been made to fix the series of problems.The e-commerce recommendation system has the characteristic of personalization and intelligent,it uses statistics and knowledge discovery technology to recommend the goods that interests the users when the system is interacting with users.After years of development,the electronic commerce recommendation system has been widely used by many large e-commerce website,they have introduced recommendation system to a different level and have achieved good results.However,the number of users and commodities increased dramatically,the accuracy and real-time performance of the recommendation can't meet the increasing demand of the consumers'.Therefore,this paper analyzes the existing problems in the recommendation system and proposes corresponding solutions,On this basis,this paper designs and implements an e-commerce recommendation model.The research work mainly includes:(1)This paper proposes a slope one recommendation algorithm based on the similarity of users and commodities.The traditional slope one algorithm and some improved slope one algorithm treat all the users and commodities equally and can not distinguish the impact that the different users and commodities have on the prediction score.To solve this problem,this paper proposes a slope one algorithm based on the similarity of users and commodities,and the experiments show that the proposed algorithm can effectively improve the accuracy of the slope one algorithm.(2)This paper proposes a collaborative filtering algorithm based on score prediction and matrix decomposition technology.As the number of users and goods in e-commerce sites increased,the online computation cost increases rapidly,making recommendation system unable to offer high quality recommendation within a short period of time.In addition,users usually grade a small number of goods,causing the user ratings matrix sparse.Making score prediction using a small amount of data is not reliable.To solve the problem,this paper proposes a collaborative filtering algorithm based on score prediction and matrix decomposition.The algorithm firstly utilize the result of slope one to fill the user-item rating matrix and reduce sparseness of the matrix.And then it uses SVD technique to reduce the dimension of the matrix and utilize the k-means clustering algorithm to cluster the users.Finally it predicts score for the target user.In addition,the similarity calculation of users is improved.On the basis of the similarity of user ratings,the algorithm adds the attribute of users to the original algorithm.The experiments show that the algorithm can not only reduce the online computation,but also has ideal recommendation quality.(3)This paper designs and implements an e-commerce recommendation model.This paper complements the overall design of electronic commerce recommendation model,and carries on the detailed design for specific modules.Finally,it expounds the recommended model of concrete realization process and completes the corresponding functional test.