Research On Social Recommendation Model Based On User Trust And Tensor Decomposition

In recent years,recommendation system has become an indispensable part of many network applications.Although recommendation algorithms have achieved great success in various network applications,data sparsity is still one of the important bottlenecks affecting the performance of algorithms.In order to solve this problem,the combination of auxiliary data has become an inevitable trend.Therefore,with the development of social network,trust-based social recommendation algorithm has been proved to be an effective solution.These algorithms use social network information to model user preferences and make recommendations.However,most of the current algorithms directly use the absolute trust relationship of social networks to improve the quality of recommendation,but do not consider that the trust of users to each friend is relative and territorial.In the trust recommendation model,considering that users trust different friends in different fields,it can reflect the actual recommendation process more accurately.Therefore,this paper proposes the concept of domain trust,and proposes a social recommendation model based on user trust and tensor decomposition,which achieves higher recommendation quality.Firstly,according to the category,the original scoring data is constructed into the third-order tensor of user-item-domain,which can further dig out the interactive information between different domain categories.Secondly,from the perspective of domain expert trust and domain user similarity,this paper proposes the concept of domain trust,and improve the formula of domain expert trust based on Page Rank algorithm,and proposes a new formula for scoring similarity based on the third-order tensor of user-item-domain.Furthermore,considering the problem of user cold start,social trust is introduced into this model,by setting recommendation weights w_uvkand adaptively balancing the influence of domain trust and social trust on the trust intensity between users,so as to avoid the neighbor dissimilarity problem caused by sparse data and improve the recognition ability of the recommendation system to the neighbor users and then improve the reliability of the recommendation.Finally,user trust is integrated with tensor decomposition recommendation algorithm to obtain the final model of this paper,namely the social recommendation model based on user trust and tensor decomposition.In this paper,a wealth of experiments are carried out on Ciao DVD datasets,and the performance of the algorithm is compared with many advanced algorithms.The results show that the proposed Trust TF model can achieve robust performance for scoring prediction recommendation task and can effectively improve the user cold start problem.