Collaborative filtering based recommendation methods focus on user-item information for modeling the user interest. However, in social networks the user interest is influenced by other user interests in the local social circle of the active user. In this paper, considering the homophily of relation to similar interests and similar friends, we propose a social item recommendation framework (SocItemRec). Our framework combines both global interest from the user-item information and local interest from social relation information for recommendations. We evaluate our framework on real world data from Sina Weibo, one of the most popular social network sites in China. The experimental results demonstrate that our framework leads to improved performance of top-k item recommendation.