Research And Application Of Recommender System Based On Social Network And Attribute Information

In the era of information overload,users cost so much time to find useful information from the massive information.Aiming at solving this problem,recommender systems learn user preference from user information and then mine information which user interests from the massive information to recommend to user.However,in practice,the recommender systems suffer from data sparsity,cold start and other problems,these problems reduce the recommendation quality of recommender system.In response to these questions,the relevant researchers use many supplementary information to alleviate these problems.The social network relationship information and attribute information is studied in this paper,and some shortcomings of the existing research are found,two algorithms are proposed and experimental analysis are conduct to prove the effectiveness of the algorithms.The main work and achievements of this paper are as follows:1.In the aspect of social network,social fusion recommend method based on individual differences is proposed in this paper.Aiming at the shortage of traditional methods to evaluate the trust value of social users,this model uses PCC and JACCARD to calculate the user ratings similarity,and then combines the social ratings similarity with the user's friends' number information in the social network to calculate the trust value between them.This approach takes into account the individual differences in users' rating process,uses variable ways to integrate user interests and social user interests to calculate the user's ratings on the items,and use corresponding regular norm to avoid over-fitting.The experimental results show that this method can improve the prediction accuracy in a certain extent.2.In the aspect of attribute information,a recommend method based on attribute preference self-learning is proposed in this paper.Cause social relationship forms slowly and difficult to acquire,this paper uses attribute information to alleviate data sparsity and cold start problems.User or item collaborative filtering algorithms is interpretable,but the training process is slow;matrix factorization methods are fast but lack concrete explanation,this method uses user'spreference for the attribute value and the attribute ratings to predict user ratings.The experimental results show that the model is running fast,and better than the traditional collaborative filtering method and the matrix factorization model when ratings data is sparse.3.At the end of this paper,this paper designs a general recommender system.The system uses ratings data,social network information,attribute information,according to the sparsity of user information to use different models to alleviate the problem of cold start and data sparsity in the recommender system.