Fine-grained Spatio-temporal Event Prediction with Self-adaptive Anchor Graph

TL;DR

This work tackles fine-grained spatio-temporal event prediction by modeling spatial heterogeneity and inter-regional correlations through a novel GSTPP framework. It pairs a global state with region-local states and employs neural ODEs with jump dynamics, all powered by a Self-Adaptive Anchor Graph that locally and adaptively learns anchor positions and inter-anchor edges. The decoder integrates a temporal intensity and a spatial Gaussian mixture conditioned on local states, trained via maximum likelihood. Empirical results on earthquake, COVID-19, and CitiBike datasets show GSTPP consistently outperforms state-of-the-art spatio-temporal models, especially in spatial predictions, demonstrating the practical value of adaptive locality for fine-grained event forecasting.

Abstract

Event prediction tasks often handle spatio-temporal data distributed in a large spatial area. Different regions in the area exhibit different characteristics while having latent correlations. This spatial heterogeneity and correlations greatly affect the spatio-temporal distributions of event occurrences, which has not been addressed by state-of-the-art models. Learning spatial dependencies of events in a continuous space is challenging due to its fine granularity and a lack of prior knowledge. In this work, we propose a novel Graph Spatio-Temporal Point Process (GSTPP) model for fine-grained event prediction. It adopts an encoder-decoder architecture that jointly models the state dynamics of spatially localized regions using neural Ordinary Differential Equations (ODEs). The state evolution is built on the foundation of a novel Self-Adaptive Anchor Graph (SAAG) that captures spatial dependencies. By adaptively localizing the anchor nodes in the space and jointly constructing the correlation edges between them, the SAAG enhances the model's ability of learning complex spatial event patterns. The proposed GSTPP model greatly improves the accuracy of fine-grained event prediction. Extensive experimental results show that our method greatly improves the prediction accuracy over existing spatio-temporal event prediction approaches.

Figures (6)

• Figure 1: The overall framework of the proposed GSTPP model. We adopt an encoder-decoder structure. The encoder simulates the global-local state trajectories, while the decoder generates the spatio-temporal event distribution.
• Figure 2: An example of a spatially localised anchor graph. Each anchor node has a spatial coordinate. The nodes are connected with edges that represent the spatial correlations.
• Figure 3: The statistics of the three datasets.
• Figure 4: The spatial probabilistic and sampling performance of GSTPP using different number of anchor nodes (clusters).
• Figure 5: Overall spatial distribution and anchor positions trained using different numbers of clusters on Earthquakes dataset. The x and y axis represent the longitude and latitude, respectively.