There has been an explosion of social approaches to leverage recommender systems, mainly to deal with cold-start problems. However, most of the approaches are designed to handle explicit user's ratings. We have envisioned Social PrefRec, a social recommender that applies user preference mining and clustering techniques to incorporate social information on the pairwise preference recommenders. Our approach relies on the hypothesis that user's preference is similar to or influenced by their connected friends. This study reports experiments evaluating the recommendation quality of this method to handle the cold-start problem. Moreover, we investigate the effects of several social metrics on pairwise preference recommendations. We also show the effectiveness of our social preference learning approach in contrast to state-of-the-art social recommenders, expanding our understanding of how contextual social information affects pairwise recommenders.