Research On Node User Alignment Model And Application Of E-commerce System On Cross Bookstore Platform

With the rapid development of Internet e-commerce,more and more users are beginning to use multiple e-commerce platforms for shopping at the same time,which leads to the research of only a single e-commerce platform to recommend to users is no longer the best solution.Through heterogeneous network alignment to find the same users distributed on different e-commerce platforms,and to integrate user attribute information and historical behavior information,it can alleviate the characteristics of unbalanced data distribution,large noise,and uneven quality of single e-commerce users.Therefore,thesis studies cross-e-commerce user behavior characteristics and user interest preferences from two levels of technology and application of heterogeneous e-commerce network alignment.The contribution of thesis can be summarized as follows:1.At the technical level of heterogeneous e-commerce network alignment,in view of the characteristics of e-commerce user behavior data sparse and complex data feature space,while taking into account the advantages of the confrontation generation network in learning data distribution and enhancing data samples,a method based on data Enhanced and data representation cross-e-commerce user alignment model.In view of the sparseness of effective behavior data of consumers on e-commerce platforms,the samples are enhanced with homomorphic data through an unsupervised confrontation generation network model to obtain more effective experimental data.In view of the complexity of the data feature space of the e-commerce platform "user-behaviorcommodity",considering the advantages of the heterogeneous network representation model JUST in capturing the structure and semantic information of multiple types of nodes,it is proposed to incorporate user product interest to change the randomness UBC2vec,a new method of wandering strategy,embeds the data feature space of heterogeneous e-commerce platforms.In view of the high complexity of user alignment calculations,a "user-commodity" two-part graph is constructed to divide users' roles,match heterogeneous e-commerce users with the same role,and reduce the number of matching users with large differences in interest groups.Thereby reducing the computational complexity of the alignment algorithm for heterogeneous e-commerce users.2.At the application level of heterogeneous e-commerce network alignment,thesis applies heterogeneous network alignment to the personalized recommendation of ecommerce bookstore users.In view of the sparseness of user rating data on a single ecommerce platform,the user alignment model is used to find common users in different e-commerce platforms to realize the completion of user attribute information and user behavior information,thereby alleviating the sparseness of user rating data on a single ecommerce platform.At the same time,considering the advantages of gray system model for dealing with less data and poor information,a personalized recommendation model based on user alignment and gray system theory is proposed.First,the user item ratings are processed at equal time intervals to obtain user rating sequences with equal time intervals.Then,accumulate and generate transformations based on the scoring sequence,and use the least square method to estimate the parameters.Finally,predict the user's rating for unrated items and make TopN recommendations.Finally,the validity and reliability of the model proposed in thesis are verified through real e-commerce data.Experiments show that the model proposed in thesis can effectively align user identities on heterogeneous e-commerce platforms,and fusion of aligned users can improve the accuracy of personalized recommendations.