Research And Application Of Collaborative Filtering Algorithm Combining Multi-Source Information

Collaborative filtering algorithm can resolve the problem of "information overload".The collaborative algorithm recommends the interesting information to users by finding groups of similar interests.However,the collaborative filtering algorithm still copes with the challenges of data sparseness and cold start,using single user behavior information cannot solve the above problems.With the development of the Internet,a large number of multi-source information such as item information,social network information have been collected,which has brought opportunities for optimization algorithms.How to use rich multi-source information to solve the problems in collaborative filtering algorithms is a hot research direction.Based on two collaborative filtering algorithms of user-based and matrix factorization,this thesis analyzes the problems existing in the current researches and makes more in-depth research and application using multi-source information.The specific content of this thesis can be summarized as follows:Researched a collaborative filtering algorithm CF-ICTI that integrates item categories and time information.The degree of user's interest in a certain category item is influenced by the number of user's rating and the size of user's rating,and also by the difference in the number of items in each category.This thesis comprehensively considers two situations by setting the threshold,and integrates the item category to optimize the interest similarity calculation.At the same time,time factor is used as weight to optimization rating similarity eases the problem of user interest change.Then the two similarities are linearly weighted to obtain the overall similarity for rating prediction and to recommend for the user.Simulation experiments show that the proposed algorithm improves the accuracy of the recommendation and alleviates the data sparsity problem from the aspect of optimization similarity calculation.Researched a matrix factorization recommendation algorithm MF-DCSN that integrates domain categories and social networks.Different areas have an impact on user social relationships and the user's influence is influenced by user rating and social network structure.First of all,this thesis divides different social networks according to domain categories,so that users trust different users in different fields.Then,in different social networks,the improved Person similarity method and Page Rank algorithm are used to calculate the user's influence under the rating information and social network structure.Finally,social relationship based on user influence is used as the regularization item optimization matrix factorization model.Then solve models for predictive rating and generate recommendations.Simulation experiments show that the proposed algorithm improves the accuracy of the rating prediction,and the new user's cold start problem is alleviated by adding social relationships.The movie recommendation system is implemented based on the CF-ICTI algorithm.Based on the requirements of the movie recommendation system,the architecture analysis,function module division and database design are implemented,the subsystems of front-end movie recommendation and back-end system management are realized.Thus,the movie recommendation system provides interesting movie recommendations for users.And verified the feasibility of CF-ICTI algorithm combined with practical application.