Abstract: Spatio-temporal data mining is an important branch in the field of data mining which has a large number of applications in the real world. Compared with time series prediction, spatio-temporal prediction algorithms need to consider both the temporal and spatial relationships of time series, which has certain complexity. In order to explore the nature of spatio-temporal data and effectively capture the complex spatio-temporal relationships, a spatio-temporal prediction algorithm based on heterogeneous information networks is proposed in the paper, which explicitly models spatio-temporal data as a heterogeneous information network and employs spatio-temporal information propagation paths to represent the rich spatio-temporal interactions. Compared with traditional spatio-temporal models that use different neural networks to capture temporal and spatial dependencies separately, the paper used meta-paths to unify spatio-temporal relationships, providing a new way for spatio-temporal data mining. Extensive experiments were conducted on two real-world open datasets to verify the effectiveness of the model.