Social Recommendation Algorithm Based On Poisson Tensor Decomposition

With the technology developments of science and Internet,especially the popularization of mobile Internet,more and more information is generated in the network.Through the analysis and utilization of these information,many applications have been derived,which greatly improve the living standard of the people.But at the same time,because of the rapid growth of data,information overload problem.Recommendation system plays an important role as a tool to solve information overload.Collaborative filtering algorithm is the most widely used recommendation algorithm,and its main is to use the user rating information of goods calculate the similarity between users,using similar friends interested in the user is recommended.When lack of user rating information of goods,however,this method is powerless,and more widespread in the real environment is the user of the implicit feedback behavior of commodities,aimed at the situation in this paper,consider using implicit feedback,it is using the user access to the history of the commodity information to predict the user may be interested in.In addition,the collaborative filtering algorithm has the problems of data-sparsing and cold-start,which greatly reduces the recommendation performance.In order to help solve these problems,some scholars suggest combining the social information of users.But the current scholars focus on exploring the influence of the users and how to use the trust relationship between users,it did not give full play to the user's social network of information such as interest similarity.For this,this article on the user's favorite preference on the basis of fully considering the social characteristics of users,with users belonging to the same interest of the community friends relationship,make full use of the user's social network information not only can ease the cold-start problem and enhance performance.At the same time,we take into account the user's time behavior.In normal,the user's behavior is influenced by the time context factors and there are frequent and time correlation,however,most of the traditional collaborative filtering algorithm was not pay attention to this,but it's very important to recommend the right program for users at the right time.We propose a social recommendation method based on Poisson tensor decomposition.This is a kind of probability model,it can not only effectively using the user's social information,and provides a new way of combining the socialization information and time characteristics of user preferences,it can use implicit feedback of user's behavior data and user social data to predict and recommend commodity the in time that user may be of interest in.The SPTD algorithm is based on the probabilistic model of Poisson decomposition model.In the process of solving the model,the model only needs to use the implicit feedback data of the positive item of the data to iterate,and has a good performance about the computational efficiency.Compared with the traditional decomposition model,the Poisson decomposition model can grasp the sparse characteristics of data,and has better recommendation effect on sparse data,and it can be easily extended to massive data sets.Therefore,the work of this paper is as follows:(1)Using the characteristics of user information and user access to the data of the history of goods to calculate the interest similarity between users,according to the interest in similarity calculation for the user's interest community,selection the friends who belong to the same interest of the community of user help recommend effect.(2)In view of the fact that users have less feedback on the display of products in the real situation,this paper uses the user's implicit feedback data to recommend the products.(3)Considering the user's time behavior,this paper proposes a model combining user's time behavior,which can show the user's interest preference at different time.(4)This paper proposes how to combine the user's social information with the context information,so as to provide a new idea for how to make full use of the information provided by both.Finally,through experiments in two real data sets,this paper proves that the SPTD algorithm has better recommendation ability and has excellent computational efficiency.