With the popularity of social network applications, more and more recommender systems utilize trust relationships to improve the performance of traditional recommendation algorithms. Socialnetwork- based recommendation algorithms generally assume that users with trust relations usually share common interests. However, the performance of most of existing social-network-based recommendation algorithms is limited by the coarse-grained and sparse trust relationships. In this paper, we propose a network representation learning enhanced recommendation algorithm. Specifically, we first adopt a network representation technique to embed social network into a low-dimensional space, and then utilize the lowdimensional representations of users to infer fine-grained and dense trust relationships between users. Finally, we integrate the fine-grained and dense trust relationships into the matrix factorization model to learn user and item latent feature vectors. Experimental results on real-world datasets show that our proposed approach outperforms traditional social-network-based recommendation algorithms.