As Location-Based Social Network (LBSN) services become increasingly popular, next Point-Of-Interests (POI) recommendation emerges as one of many important applications of LBSNs. With the growing ability of collecting information, more and more temporal, spatial, social contextual and semantic tags information is collected in systems, which makes the location prediction problem becomes feasible. Some works, like Factorizing Personalized Markov Chain (FPMC), Tensor Factorization (TF), Recurrent Neural Networks (RNN), etc., have been proposed to address this problem, but they all have their limitations. In this study, we extend Long-Short memory recurrent neural networks (LSTM) and propose a novel method called POI-LSTM. POI-LSTM can model social contextual and semantic tags information in each layer, and employ temporal and spatial contexts in more efficient way. Experimental results show that the proposed POI-LSTM model yields significant improvements over the competitive compared methods on two typical datasets, i.e., Yelp and Foursquare dataset.