Research On Cross-Domain Information Transfer Learning Algorithms In Recommender Systems

With the rapid development of Internet,we enter the information era.The Internet is providing us with more information and services.However,when we enjoy the advantages of network,we also have to face a lot of spam.This creates "information trek" and "information overload" problems,which lead to the birth of personalized recommendation systems.About the personalized recommendation systems,people give a large number of recommendation algorithms,among which collaborative filtering algorithm is one of the most successful technologies at present.It has been widely used in many fields,such as e-commerce,social network,review site and so on.However,with the increase of recommendation system scale,collaborative filtering technology faces great challenges,especially the sparsity problem has become the main bottleneck of many collaborative filtering algorithms.In recent years,the idea of cross-domain information transfer learning has provided a new way to alleviate the common sparsity problem in recommendation systems.That is,we can transfer the information in relatively dense auxiliary domain to the target domain to improve the performance of the recommendation systems.In view of this problem,this paper mainly discusses from the following aspects:First,we give Codebook matrix construction method and Codebook matrix transfer learning method in traditional cross-domain information transfer learning algorithm,and point out that the critical point of cross-domain information transfer learning lies in the construction of the Codebook matrix.On this basis,we introduce a novel Codebook matrix construction method which is based on general matrix factorization method,and further give the cross-domain information transfer learning algorithm which is based on general matrix factorization.Then,we consider adding user social information into the process of matrix factorization to improve the performance of user/item latent feature matrix generated from matrix factorization.Based on this,we introduce social matrix factorization into the process of construction of Codebook matrix and further present the cross-domain information transfer learning algorithm which is based on social matrix factorization.Finally,we consider selecting a novel clustering algorithm,that is Kind AP one-class clustering algorithm,to cluster user/item latent feature matrix which is obtained from social matrix factorization and generate user/item partition information.Based on this,we can achieve the construction of Codebook matrix and establish the improved cross-domain information transfer learning algorithm which is based on social matrix factorization.Above all,this paper mainly researches three cross-domain information transfer learning algorithms which are based on matrix factorization.On this basis,we compare these three algorithms with traditional algorithms and verify the effectiveness of the algorithms proposed in this paper by a series of experiments.