Research On Social Recommendation Algorithm Of Hybrid Tags And Probability Matrix Factorization

Under the background of big data age,users are faced with explosive growth of information.Using recommendation technology in various fields is one of the main methods to solve the problem of information overload.The probability matrix factorization method is proposed and applied in the recommendation field,and good results are obtained.It can reduce the dimension of the user rating matrix,study the properties of high-dimensional space by using low-dimensional space data,alleviate the problem of sparse data and cold start,and improve the computational efficiency.However,the explanation of probability matrix factorization to make recommendation results is poor.The recommendation results only depend on the scoring data,and the recommendation accuracy is limited.The popularity of social networks makes people pay more attention to the application of recommendation algorithm in the field of socialization.Social recommendation algorithm improves the traditional recommendation algorithm to model the user-project binary relationship,and adds social attribute information such as social tags,trust relationships and so on to the probability matrix model to form a social matrix factorization model,which can provide explanation for the recommendation results,while it still haves the above advantages of probability matrix factorization.In social recommendation technology,using social attributes to model users is the key of the algorithm,which can be summarized as two ways: one is using social tags to build user interest model,the other is building trust network between users.By using these two methods,the algorithm can not only rely on scoring information to make recommendations,but also can effectively alleviate the cold start problem.Many researches have been done on usermodeling using social attribute information,but there are still some problems:(1)In the existing algorithms of user interest modeling with social tags,the preference of user tagging is ignored when calculating the weight,which leads to inaccurate user interest modeling and affects the recommendation accuracy.(2)In the trust relationship-based recommendation algorithm,only the number of friend's relationship is considered,and the interest difference between friends is ignored.The cold start problem still exists for new users who lack friends.In order to solve the problem of insufficient modeling of user interest by using tags in social recommendation,considering that the user's interest in tags is not only the number of choices,but also the subjective preference problem,this paper combines the score and tags to calculate the weight,and improves the weight calculation method.In order to solve the problems of insufficient trust relationship mining and deviation of interest among friends in trust-based recommendation algorithm,social tags are incorporated into the calculation of user preference trust,the calculation of trust relationship is improved,and cold start is relieved.To sum up,this paper has carried on the following work:Firstly,this paper introduces the research status of social recommendation algorithm,mainly divided into two parts,namely the status of social tag recommendation and social trust recommendation.Several classical recommendation methods are described,and their advantages and problems are analyzed.Secondly,to solve the problem of imperfect modeling of user interest using tags,a hybrid user tags and ratings probability matrix factorization(TR-PMF)algorithm is proposed.In this paper,the user score is combined with tf-idf method,which is used to calculate the weight in the text field,and the time decay function is fused to mine the user's interest in tags more fully.The similarity of interest is calculated to filter the neighbors and the probability matrix method is fused.A probability matrix factorization recommendation algorithm(TRPMF)is proposed to fuse the tag score.The experimental results show that TR-PMF can better model the user's interest preference and improve the accuracy of score prediction and recommendation.Thirdly,to solve the problem of insufficient trust relationship mining in the recommendation algorithm of social trust,a hybrid user tags and trust relationships probability matrix decomposition algorithm(TT-PMF)is proposedIn this paper,we use social tags to mine trust relationship more fully,and divide user trust into explicit trust and implicit trust.The explicit trust is calculated by combining the tag and the friend relationship,the implicit trust relationship is obtained by considering the social tag and the score comprehensively,and the improved trust relationship is obtained by linearly fusing the social tag and the score,and the probability matrix decomposition algorithm(TT-PMF)is obtained by combining the trust relationship with the So Rec trust model.The experimental results show that the TT-PMF algorithm performs well in accuracy and score prediction results,and can better solve the cold start problem and improve the recommendation accuracy.