A Tensor-based Recommendation Algorithm Using Heterogeneous Implicit Feedback

In modern e-commerce websites,personalized recommendation services play a vital role.Personalized recommendation can not only help users to explore new shopping needs,but also help e-commerce websites to improve user stickiness.However,in the e-commerce website,users often do not have the impulse to evaluate the products,and the user's operation behavior becomes a more important research factor.The existing research in this area mainly uses the user purchase data for recommendation,but the times of purchase does not completely represent the user's preference.Moreover,most users' operating data on goods is very sparse,which makes the recommendation results inaccurate.However,different types of user actions in ecommerce can provide additional potential and valuable information to the recommendation system.The paper first divides implicite feedback into auxiliary feedback and target feedback.Auxiliary feedback includes clicking,favoriting and wanted,target feedback refers to purchase.Then we analyze the data in the user's operational behavior,and analyzes the auxiliary feedback and social information in detail.Furthermore,a recommendation algorithm based on tensor decomposition technique is proposed,which uses the CP decomposition(CANDECOMP/PARAFAC Decomposition)to model the user's implicite feedback,and adds the social regularization in the tensor decomposition model to model the social information between users.This scheme exposes the hidden dependency among users,items,and actions and breaks the limitation of the user–item matrix.Moreover,it also considers the social information as regularization terms to obtain a trust relationship between users and their friends.The experimental results obtained from a real-world dataset show that the proposed algorithm outperforms other compared methods,thus effectively improving the performance of the recommender system.