Research On Recommendation Algorithm Integrating Community Division And Collaborative Filtering

Collaborative filtering technology is one of the most widely used technologies in personalized recommendation systems.User based collaborative filtering algorithms calculate user similarity based on a user rating matrix,predict target user ratings based on similar user ratings,and implement recommendations for them.The idea is simple and recommendations are effective.However,the continuous growth in the number of users and projects has led to excessive similarity calculation,time efficiency is challenged,and scalability needs to be improved;the scarcity of items that users rated results in a very sparse user rating matrix,and the accuracy of recommendations needs to be improved;new users do not have historical scoring records,resulting in the inability to calculate similarity between users,and the system cannot make effective recommendations.This cold start issue can affect the user experience.Aiming at the problems existing in traditional collaborative filtering algorithms,this thesis has done the following research work:(1)A recommendation algorithm ICDCF that combines implicit social network community division and collaborative filtering is designed.The algorithm treats the common interests of users in projects as social relationships.Firstly,an improved Jaccard similarity coefficient that takes into account user implicit relationships is used to measure the strength of social relationships between users;construct an undirected weighted implicit social network with users as the vertex and user similarity as the weight of the connecting edges;then,users of implicit social networks are divided into communities based on the spectral clustering idea;then implement user based collaborative filtering recommendations within the community.This algorithm can effectively avoid the inaccurate similarity calculation problem caused by sparse data and few common scoring items in the collaborative filtering recommendation stage,and can reduce the computational burden of target users searching for neighbors,improving time efficiency.Experimental results on the classic dataset Movie Lens 100 K and Film Trust demonstrate the performance advantages of ICDCF.(2)In order to solve the user cold start problem and further improve the accuracy of recommendation,a recommendation algorithm CICDCF that combines user context information and ICDCF algorithm is designed.Based on the CDCF algorithm,user context information and the comprehensive similarity simultaneously which considers user historical rating data and user context information are introduced in;users are clustered according user context information to determine the central user.The target new user is classified into the community that the nearest central user belongs to.Both new and old users are conducted collaborative filtering recommendations within the community based on comprehensive similarity.The results of performance comparison experiments with ICDCF algorithm on the Movie Lens 100 K dataset and Movie Lens 1M dataset show that the CICDCF algorithm can solve the user cold start problem to a certain extent while further improving recommendation accuracy.(3)In order to verify the practicality of the CICDCF algorithm,a simple personalized book recommendation prototype system was developed.The system applies the CICDCF algorithm to generate a list of books that may be of interest to new and old users.The application results show that the CICDCF algorithm can be effectively applied to personalized book recommendation systems for book recommendation.