Recently, urban traffic management has encountered a paradoxical situation which is the empty carrying phenomenon for taxi drivers and the difficulty of taking a taxi for passengers. In this paper, through analyzing the quantitative relationship between passengers' getting on and off taxis, we propose a time-location-relationship (TLR) combined taxi service recommendation model to improve taxi drivers' profits, uncover the knowledge of human mobility patterns, and enhance passengers' travel experience. Moreover, the TLR model uses Gaussian process regression and statistical approaches to acquire passenger volume, mean trip distance, and average trip time in functional regions during every period on weekdays and weekends, and allows drivers to pick up more passengers within a short time frame. Finally, we compare our proposed model with the autoregressive integrated moving average model, the back-propagation neural network model, the support vector machine model, and the gradient boost decision tree model by using the real taxi GPS data in Beijing. The experimental results show that our optimizing taxi service recommendation can predict more accurately than others by considering the 3-D properties.