Actions carried out in social networks such as Facebook, Twitter, Foursquare, and the like, can greatly benefit the quality of recommendations provided by recommender systems. Usual interactions on these platforms provide valuable information on user preferences and proximity with other users. However, the use of this information by external recommender systems is still incipient and developers have little support to do it efficiently. In this paper, we introduce a model-driven development framework for recommenders systems based on social networks. The core of this framework is an abstract, social network-independent model of recommender systems, which combines domain-independent concepts of collaborative filtering with basic concepts of social networks that can be exploited for recommendation purposes. From this model, developers can specify the structure and algorithms of domain-specific recommender systems at a high abstraction level. An automatic code generation strategy supports the implementation phase. Experiments show promising results in development-time saving.