With the development of location based social network, location recommendation, a typical recommender system, plays a more and more significant role in addressing data overloading, enhancing user engagement and improving platforms’ profit. Most existing researches on location recommendation are based on matrix factorization, which cannot capture the complicated relation between users and locations. In addition, in location based social network, social relation data is important for building user demographics, and therefore it becomes a major concern that how to combine social relation data to help improving recommendation quality. In this paper, a location recommendation approach based on deep learning is studied. By designing two novel designs, a social-aware sampler and a social-enhanced regularizer, the social information is integrated. Extensive experiments on two real-world datasets demonstrate that the proposed methods can significantly improve the recommendation accuracy compared with existing models.