Research On Social Recommendation Algorithm Fusing Time Series And Friendship Relationship

In recent years,social networking services have become increasingly popular,and people often accept friends' recommendations when shopping online.Many scientific researchers have tried to integrate the user's social network information into the recommendation algorithm.Studies have shown that the user's social network information can indeed effectively alleviate the data sparsity and insufficient scalability problems of the collaborative filtering recommendation algorithm,which has certain research significance.However,most of the recommendation algorithms based on social networks only use the binary trust relationship between the user's friends,but ignore the problem of different trust levels between friends due to different interest preferences,and many algorithms do not consider the friends that users trust in different fields will also be different.On this basis,in view of the problems existing in the current algorithm,this article uses the characteristics of the user's friend relationship in the social network and the user's interest behavior to change with time,and proposes an improved algorithm.The specific research content is as follows.(1)In view of the sparsity of user-item data,this paper improves the collaborative filtering recommendation algorithm based on users,and uses the Recommendation users' social information and rating information to propose a recommendation algorithm that combines trust models and user behavior.First,the social trust network is divided according to the project category.Second,the trust model of Recommendation users is established in the social network according to the transitivity of user trust.Considering the differences in the interest preferences of the user and neighbors,the user and friends are calculated using the common scoring items the degree of trust between users,while using the authenticity of user ratings to reduce the impact of false or machine users.Finally,the recommendation weights are formed based on the similarity of the users,and the target items of the Recommendation users are predicted and scored.Comparative experiments on the Epinions dataset show that the proposed algorithm is indeed superior to other collaborative filtering recommendation algorithms in terms of accuracy and recall.(2)Based on the model-based collaborative filtering recommendation algorithm theory,this paper proposes a social recommendation algorithm that combines time characteristics and matrix decomposition based on LFM matrix decomposition.First,based on the scoring time in the scoring information,this paper improves the user similarity measurement method based on the time decay function;secondly,dynamically models the user's global social relationship and local social relationship in the social network;then The user's personal interest preference changes and social relationship model are added to the matrix decomposition,and then the obtained user similarity and predicted similarity are made a difference,and integrated into the loss function to form a training model.Finally,predict the missing score,and then push the generated recommendation list to the user.Simulation experiments were carried out on the Yelp data set.Through experimental comparison and analysis,this algorithm reduced the root mean square error and average absolute error of the Recommendation results.There are 33 figures,10 tables and 83 references in this paper.