Proactive caching is a promising technology in 5G wireless networks. Small-cell base stations (SBS) can cache popular contents to assist the macro base station, and proactive caching are considered to cope with the weak backhaul links of SBSs. However, obtaining popular contents and making the optimal caching strategy may be challenging. In this paper, a novel learning-based approach is proposed, in which regularized singular value decomposition (RSVD)-based collaborative filtering (CF) is used to estimate the content popularity and transfer learning (TL) is adopted to improve the estimation accuracy. Then considering the interaction between users and SBSs, a distributed iterative algorithm is designed to make a caching strategy with the goal to maximize the number of users who can be served by neighboring SBSs. Experiments have been conducted to evaluate the performance of the proposed algorithms and simulation results demonstrate the effectiveness of our learning-based approach for proactive caching.