In many practical recommender systems, it is found difficult to obtain explicit feedback from users about the preference for a specific item, such as music, book, movie, etc. Most researches up to this point has focused on tracking various sources of implicit feedback from user behavior including purchase history, browsing patterns, and watching habits, in order to model user preference. In this paper, we investigate a method that uses information exploited from a user's online media sharing activities as a novel source of implicit feedback for recommendation system. We look into elements of media sharing behavior and suggest whether behaviors have the potentiality that could play a role as a predictor of users’ preference. Then in a specific domain, we choose appropriate behaviors by two criteria: abundance and observability. As a representative case, we focus on YouTube, one of the most popular social video sharing sites. By criteria we suggest, we select three behaviors including favorite, upload and view and formulate the simple item-based algorithm based on those behaviors. Through a series of experiments, we evaluate recommendation results obtained from our dataset by comparison with those from other reference algorithms. The results show that favorite and upload have possibility to be used as implicit feedback.