Online videos, e.g., YouTube videos, are important topics for social interactions among users of online social networking sites (OSN), e.g., Facebook. This opens up the possibility of exploiting video-related user social interaction information for better video recommendation. Towards this goal, we conduct a case study of recommending YouTube videos to Facebook users based on their social interactions. We first measure social interactions related to YouTube videos among Facebook users. We observe that the attention a video attracts on Facebook is not always well-aligned with its popularity on YouTube. Unpopular videos on YouTube can become popular on Facebook, while popular videos on YouTube often do not attract proportionally high attentions on Facebook. This finding motivates us to develop a simple top-k video recommendation algorithm that exploits user social interaction information to improve the recommendation accuracy for niche videos, that are globally unpopular, but highly relevant to a specific user or user group. Through experiments on the collected Facebook traces, we demonstrate that our recommendation algorithm significantly outperforms the YouTube-popularity based video recommendation algorithm as well as a collaborative filtering algorithm based on user similarities.