Research On Recommendation Algorithm Based On Trust Relationship Of Users

Classic recommendation algorithms,such as collaborative filtering,have been proved to have excellent recommendation ability in real applications,but they also have their own limitations.The traditional collaborative filtering recommendation algorithm only uses the user's historical data for training,which lead to the problems of missing data and sparse input.The model-based collaborative filtering algorithm greatly alleviates these problems.The collaborative filtering algorithm based on matrix factorization is one of them.At the same time,put data like user social information into model to solve the problem of sparse input matrix has been proved to be significantly effective.In this paper,a matrix decomposition collaborative filtering algorithm based on social trust is proposed to solve the user cold start problem in sparse data problem.As a matrix decomposition recommendation algorithm based on trust,TrustSVD algorithm integrates trust information into the algorithm by re-represent rating prediction with user trusts and sharing user latent factor matrix in the processes of matrix decomposition,making the results more accurate than other similar algorithms.But the algorithm only considers the direct trust relationship between users,and does not consider the dynamic propagation of trust.In this paper,we choose to study trust propagation,enrich the sparse trust data set by adding the weak connection trust relationship between users,and mine users more deeply based on the trust measurement.This paper focus on the following two parts:(1)Proposes an improved algorithm based on weak connection trust,W-TrustSVD: Algorithm adds the indirect propagation investigation of trust to the direct trust relationship set of users,constructs the definition of weak connection trust of users to enrich the user trust set,at the same time,the comprehensive measurement method of trust is reasonably constructed,and the user weak connection trust set is selected and constructed in proportion according to the comprehensive trust measurement.The W-TrustSVD algorithm improves the prediction results by adding the weak connection trust set,and experimental results on the open data set confirm the effectiveness of the W-TrustSVD algorithm.(2)Based on the user trust relationship and preference in historical behavior,a matrix decomposition recommendation model based on weighted random walk,WPR-TrustSVD,is proposed to simulate the propagation mode of user trust through random walk: algorithm combines the user-item and user-user networks by linear weighting to integrate trust data and rating data;based on the user's trust preference and scoring preference,the walk weight is defined to weigh the edge of the comprehensive network graph;at last,the random walk result is calculated to filter the user's trust set for personalized recommendation.The model performs well in real data sets.