Research On Social Recommendation Algorithm Based On Probability Matrix Factorization

The recommendation algorithm can quickly and accurately recommend information that meets preferences and needs for users,and solves the problem of information overload.The application of recommendation algorithms not only enhances the user experience,but also brings higher economic benefits to the enterprise.In recent years,recommendation algorithms have achieved great success in both research and application fields,but they still face serious data sparsity and cold start problems,which greatly reduces the effectiveness of recommendation.The emergence and rapid development of social media have provided new ideas to alleviate these two problems,so social recommendation algorithms have become a hot research topic in recent years.The social recommendation algorithm combining two kinds of data,trust relationship and scoring information,alleviates the problem of data sparsity and cold start to a certain extent.However,the existing social recommendation algorithms still have two shortcomings.First,the problem of low utilization of trust relationship data,and secondly,when the user preference characteristics are modeled,the trust relationship data and the scoring data are not considered to belong to different feature spaces.problem.Both problems have led to a decrease in the prediction accuracy of the social recommendation algorithm and the quality of the recommendation list.This thesis is based on the social recommendation algorithm based on probability matrix factorization,and studies the above two issues in the field of social recommendation.The innovations and main tasks of our thesis are as follows:1)The existing social recommendation only uses the user node adjacency information in the binary data,and ignores the hidden network topology information,which leads to a decline in the utilization of trust relationship data,which in turn reduces the quality of the recommendation results.This thesis uses the idea of trust propagation to propose a user trust quantification algorithm TQ?RWR based on Random Walk with Restart,which measures the trust value between users by mining the user credibility implicit in the trust network topology.Based on TQ?RWR,a social recommendation algorithm RS?RWR based on quantitative trust is proposed.Experiments on three public data sets show that the trust metric algorithm proposed in this thesis effectively improves the utilization of trust relationship data and improves the accuracy of recommendation results.2)At present,the social recommendation algorithms based on probability matrix decomposition are based on the assumption that the rating data and trust data share a user preference feature space.However,the analysis found that this is inconsistent with people's actual thinking mode,which leads to inaccurate modeling of user preference characteristics and reduces the recommendation effect.This thesis proposes the concept of basic feature space,and considers that user rating preferences and trust preferences are mappings of user basic features.Using TQ?RWR quantized trust data and scoring data,a probability matrix factorization social recommendation algorithm PTRust PMF based on the basic feature space is proposed,and an optimization method is designed for the algorithm.The comparative experiments on three data sets with other five existing recommendation algorithms show that the PTrustPMF algorithm proposed in this subject improves the recommendation effect,alleviates the sparsity problem and the cold start problem.