Abstract: With the rapid growth of Web2.0,social recommendation has become one of the hot research topics in the last few years.It is the key point to improve recommender systems using social contextual information in a more efficient way.The existing social recommendation approaches mainly take advantage of user’s direct connection(explicit relation).This paper detailed social relation as explicit relation and implicit relation and obtained the user’s reputation by using his/her historic records.Then we proposed a recommendation framework capturing user’s global social relation(reputation)and local social relation(explicit relation and implicit relation).Using two real datasets,Douban and Epi-nions,we conducted a experimental study to investigate the performance of the proposed model GDLRec.We compared our approach with existing representative approaches.The results show that GDLRec outperforms other methods in terms of prediction accuracy.

Key words: Social recommendation,Matrix factorization,Reputation,Implicit relation