Research Of Collaborative Filtering Recommendation Algorithm Based On Relationship Among Entities

As the key branch of personalized information service and decision support research field,recommender system has been the research highlight in academia and industry in recent years.Recommendation algorithm is the kernel of recommender system.Among all these algorithms,Collaborative filtering recommendation algorithm is the one that has good recommendation effectiveness and widespread application.After the research of traditional collaborative filtering recommendation algorithms and social recommendation algorithms,we proposed a new collaborative filtering recommendation algorithm that based on relationship among entities.Overlapping community detection algorithm was firstly used to obtain the community structure in users' social network and get the user clusters based on community structure.Then,we designed the calculation method of vectors of users' interest with the exploitation of items' category information and users' rating behaviors.With these vectors as inputs,fuzzy clustering algorithm was used to get user clusters of common interest.We also quantified users' interest level on these two different kinds of user clusters.After that,we put forward a comprehensive approach of similarity computation based on users' rating behaviors and items' category information.Finally,we put the association information of different entities as influential factors into matrix factorization model to obtain the collaborative filtering recommendation algorithm that based on relationship among entities.Experiments has been conducted on real dataset to verify the effectiveness of the algorithm.The result of experiments demonstrated that,compared with traditional collaborative filtering recommendation algorithms and social recommendation algorithms,our algorithm got better recommendation accuracy and also effectively alleviated the decline of accuracy caused by cold users.