This paper proposes a matching degree to study dynamic spatiotemporal characteristics of urban taxi and offers a novel understanding of self-organization taxi dispatch in hotspots on top of the Fermi learning model. The proposed matching degree can not only reflect the overall spatiotemporal characteristics of urban taxi supply and demand but also show that the density of distribution and the distance between the taxis supply and the city center will affect the satisfaction of demand. Besides, it is interesting to note that supply always exceeds demand and they will self-organize into an equilibrium state in hotspots. To understand the phenomenon, we develop the Fermi learning model based on the prospect theory and compared the results with the popular reinforcement learning model. The results demonstrate that both models can account for self-organization behavior under different scenarios. We believe our work is crucial to explore taxis data and our indicator can provide a significant suggestion for urban taxis development.