Research On Collaborative Filtering Recommendation Algorithms Based On Social Relations And Community

Due to the explosive growth of various information on the Internet,users are unable to obtain the information they need timely and accurately when faced with so large amounts of data.Under such a circumstance,recommend system are proposed to tackle this problem.The core task of recommendation system is to analyze user behavior,mine user preferences and push information for users.Collaborative filtering recommendation technology is widely studied because it only needs user-item rating information to recommend and it is not difficult to implement.However,collaborative filtering algorithms face problems such as sparse data,cold start,and poor scalability.In recent years,some studies have begun to integrate the social relationship and community structure information among users into collaborative filtering recommendation.The results show that social relations and community structure information can alleviate data sparsity and enhance scalability.Therefore,recommendation based on social relations and community structure has important research significance.On this basis,in view of the problems still existing in the current algorithm,this paper conducts a study on the recommendation algorithms in depth combining social relations and community structure.The specific research contents are as follows:1.Current recommendation algorithms based on social relationships often only consider trust relationships among users,but ignore the influence of distrust relationships and interesting correlation among users.To solve this problem,this thesis studies a regularized matrix factorization recommendation model that integrates social relations and interests.Firstly,the global and local topological characteristics of the network are used to structure the trust and distrust relationship matrices among users.Then,an improved method of interest preference similarity calculation among users is proposed based on project similarity.Finally,the trust matrix,distrust matrix and interest correlation are combined to make recommendations for users in the process of matrix decomposition.The simulation results show that the proposed algorithm has a lower recommendation error than the traditional social recommendation algorithms and can alleviate the cold start problem.2.Although current recommendation algorithms based on community structure have improved the processing ability of large data sets,it often only considers one type of community,which leads to the error of recommendation is considerable.In response to this problem,this thesis studies a recommendation model that combines user community structure and scoring joint community.Firstly,using trust relationship and similarity among users,k-means algorithm that improved initial cluster centers is used to discover user community.Then,the user-item scoring matrix is used to mine the joint community,and the scoring sub-matrix is constructed.Finally,a matrix score is made for integrating user community structure in sub-matrix.The simulation results show that the proposed algorithm can reduce the error of rating prediction while ensuring high recommendation efficiency.