This paper proposes a novel spatial-temporal recurrent neural network (ST-RNN) for predicting the next location of users based on spatial and temporal contexts. Existing methods such as Factorizing Personalized Markov Chain (FPMC) and Tensor Factorization (TF) have limitations in modeling continuous time intervals and geographical distances. ST-RNN addresses these issues by incorporating time-specific and distance-specific transition matrices in each layer of the recurrent network. These matrices capture the dynamic properties of continuous time intervals and geographical distances, respectively. To handle continuous values, the spatial and temporal values are discretized into bins, and linear interpolation is used to calculate the corresponding transition matrices. The proposed ST-RNN is evaluated on two datasets: the Global Terrorism Database (GTD) and the Gowalla dataset. Experimental results show that ST-RNN significantly outperforms existing methods in terms of recall, F1-score, MAP, and AUC. The model is effective in capturing both local temporal and spatial contexts, and its performance is robust to variations in window width and dimensionality. The results demonstrate that ST-RNN can accurately predict future locations by considering both time interval and geographical distance information.This paper proposes a novel spatial-temporal recurrent neural network (ST-RNN) for predicting the next location of users based on spatial and temporal contexts. Existing methods such as Factorizing Personalized Markov Chain (FPMC) and Tensor Factorization (TF) have limitations in modeling continuous time intervals and geographical distances. ST-RNN addresses these issues by incorporating time-specific and distance-specific transition matrices in each layer of the recurrent network. These matrices capture the dynamic properties of continuous time intervals and geographical distances, respectively. To handle continuous values, the spatial and temporal values are discretized into bins, and linear interpolation is used to calculate the corresponding transition matrices. The proposed ST-RNN is evaluated on two datasets: the Global Terrorism Database (GTD) and the Gowalla dataset. Experimental results show that ST-RNN significantly outperforms existing methods in terms of recall, F1-score, MAP, and AUC. The model is effective in capturing both local temporal and spatial contexts, and its performance is robust to variations in window width and dimensionality. The results demonstrate that ST-RNN can accurately predict future locations by considering both time interval and geographical distance information.