ADS-POI: Agentic Spatiotemporal State Decomposition for Next POI Recommendation
A novel approach known as ADS-POI (Agentic Spatiotemporal State Decomposition for Next Point-of-Interest Recommendation) has been introduced to enhance next POI recommendations. This technique tackles the shortcomings of current models that condense user history into a singular latent representation, which mixes various signals such as habitual mobility trends, immediate intentions, and time-related patterns. Instead, ADS-POI characterizes a user through several parallel evolving latent sub-states, each influenced by distinct spatiotemporal transition dynamics. These sub-states are combined selectively via a context-conditioned mechanism to create the decision state for predictions. This innovative structure allows for adaptable state evolution and improved responsiveness to various decision-making contexts. The research can be found on arXiv with the identifier 2604.20846.
Key facts
- ADS-POI is a spatiotemporal state decomposition framework for next POI recommendation.
- It models user mobility as a spatiotemporal sequence with multiple behavioral factors.
- Existing methods entangle heterogeneous signals in a single latent representation.
- ADS-POI uses multiple parallel evolving latent sub-states with individual transition dynamics.
- Sub-states are selectively aggregated via a context-conditioned mechanism.
- The framework aims to improve flexibility and adaptability in decision contexts.
- The paper is published on arXiv with ID 2604.20846.
- The announcement type is cross.
Entities
Institutions
- arXiv