Abstract: With the rapid development of Internet and information technology,information overload becomes more and more seriously.Recommender system can provide personalized recommendations to both individual users and businesses (such as e-commerce and retail enterprises).The data sparsity and prediction quality are recognized as the key challenges in the existing recommender systems.Most of the existing recommender systems depend on collaborating filtering (CF) method,which mainly uses the user-item rating matrix to represent the relationship between users and items.Se-veral researches consider utilizing extra information to improve the accuracy.However,most of the existing methods usually fail to provide accurate information for predicting recommendations,as there is an assumption that the relationship between attributes of items is independent and identically distributed,while,there are often several kinds of coupling relationships or connections existing among items or users in real applications.This paper incorporated the coupling relationship analysis to capture under-discovered relationships of items and aimed to make the ratings more reasonable.This paper proposed a coupled attribute-based matrix factorization model,which can capture the coupling correlations between items effectively.The experimental evaluations demonstrate the proposed algorithms outperform the state-of-the-art algorithms in the warm start and cold start settings.

Key words: Recommender systems,Similarity,Matrix factorization,Cold-start,Predicting