ADS-POI: Agentic Spatiotemporal State Decomposition for Next Point-of-Interest Recommendation

ADS-POI addresses a fundamental limitation in next POI recommendation: most models compress a user's entire history into a single latent vector, entangling routine patterns with short-term intent and temporal regularities. This reduces flexibility and accuracy. The new framework represents a user with multiple parallel evolving latent sub-states, each governed by its own transition dynamics. A context-conditioned mechanism selectively aggregates these sub-states to form the decision state for prediction, allowing different behavioral components to evolve at different rates while staying coordinated.

Extensive experiments on three real-world benchmark datasets from Foursquare and Gowalla show that ADS-POI consistently outperforms strong state-of-the-art baselines under a full-ranking evaluation protocol. The results demonstrate that decomposing user behavior into multiple spatiotemporally aware states leads to more effective and robust next POI recommendation. The code is publicly available on GitHub, enabling further research and application in location-based services.