Recommender systems play an important role in modern e-commerce systems. Rating prediction is an important task for recommender systems. Recent studies in social recommendation enhance the performance of rating predictors by taking advantage of user relationship. However, these approaches mostly rely on user personal information to make a prediction. Due to privacy concerns, we should avoid using user personal information. In this paper, we present a rating prediction approach relying on deep learning. The approach is easy to implement and does not reveal any personal information. Experiments on real-world data sets showed that the approach outperforms state-of-the-art in both warm-start and cold-start problems.