Collaborative Filtering Recommendation Algorithm Based On Transfer Learning And Context Information

As an effective way to solve the problem of information overload,the recommendation system has been widely used in e-commerce,in the fields of video and music.The model-based collaborative filtering recommendation algorithm has gained wide attention in industry and academia due to its good expansibility and strong academic research.This paper deeply analyzed the recommendation accuracy,data sparsity and scalability in the practical application of recommendation system.This paper proposed two methods of migration learning and context information to solve these key problems.The concrete research contents are as follows:1?In order to improve the accuracy of collaborative filtering recommendation algorithm,this paper proposed a collaborative filtering recommendation algorithm based on transfer learning.The algorithm divided the user set into subspaces and used transfer learning model to improve the accuracy of the recommendation algorithm.The algorithm takes the user behavior data collected by the recommendation system of the actual website with a certain degree of quality difference as the basic assumption,and comprehensively considers the user score information entropy and the user score feature space variance to evaluate the data quality.On the basis of the difference of data quality,the data is divided into sub-groups of different quality,and the recommendation precision of the low-quality group is improved by migrating the item's hidden feature vector of high-quality group.The experimental results show that the difference in data quality does have an important impact on the improvement of recommendation accuracy.2?In order to alleviate the influence of the scalability and data sparsity on the recommendation algorithm,this paper proposed a collaborative filtering recommendation algorithm based on coupled auxiliary information.The algorithm fused user context information and project context information into matrix decomposition model.The algorithm used coupling similarity to calculate the similarity of the classification attribute features,which alleviated the influence of the scalability and improved the recommendation accuracy of the algorithm.Experiments show that this algorithm improves the scalability while ensuring accuracy,and the method of coupling auxiliary information calculation alleviated the impact of data sparsity.3?In order to alleviate the influence of the data sparsity on the recommendation algorithm,this paper proposed a collaborative filtering recommendation algorithm based on fusing trust relationship.The algorithm combined the user trust relationship with the factorization machine model,and comprehensively considered the user's score similarity and social trust relationship as user contribution,the algorithm focused on the asymmetry of user trust relationship.Experiments show that compared with the existing collaborative filtering recommendation model with social relationship,the proposed model obtains higher recommendation accuracy,and the method of integrating user trust relationship alleviated the impact of data sparsity.