CausalPOI: Spatio-Temporal Graph-Based Causal Modeling for Cold-Start POI Check-in Forecasting

Researchers introduce CausalPOI, a spatio-temporal graph-based machine learning framework designed to predict check-in patterns for newly opened Points of Interest by modeling causal relationships between locations. The approach outperforms existing methods by capturing functional dependencies between POIs rather than relying solely on proximity, offering improved forecasting accuracy for urban planning applications.

CausalPOI addresses a gap in urban analytics by tackling the cold-start problem in POI forecasting—predicting behavior for newly introduced locations where historical data is unavailable. Traditional spatio-temporal models depend on proximity-based assumptions and correlation patterns, missing the deeper causal mechanisms that drive foot traffic and commercial success in urban environments. This research bridges that divide by introducing causal representation learning to urban computing.

The framework's innovation lies in its use of structurally aligned treatment and control graphs to simulate both factual and counterfactual scenarios. Rather than treating POIs as isolated nodes, CausalPOI models semantic and spatial relationships through a Spatio-Temporal Functional Interaction Graph, revealing how urban interventions affect neighboring locations. The use of real-world SafeGraph datasets provides grounding in actionable, commercially relevant data.

For urban planners, commercial developers, and municipal governments, this represents a significant capability—predicting how a new retail location, transit hub, or venue will perform before launch. Investors and real estate professionals could leverage such forecasting to assess commercial viability more accurately. The causal modeling aspect particularly matters because it enables scenario analysis: understanding not just where foot traffic will concentrate, but why, and how policy interventions or infrastructure changes ripple through urban ecosystems.

The open-source release signals academic commitment to reproducibility and potential adoption. As cities increasingly rely on data-driven decision-making, tools that combine spatio-temporal accuracy with causal reasoning become essential infrastructure for urban innovation.

• →CausalPOI predicts check-in patterns for newly opened POIs by modeling causal relationships rather than proximity alone.
• →The framework uses structurally aligned treatment and control graphs to simulate factual and counterfactual urban scenarios.
• →Spatio-Temporal Functional Interaction Graph captures semantic and spatial dependencies between multiple points of interest.
• →Real-world SafeGraph validation demonstrates significant performance improvements over existing spatio-temporal baselines.
• →Causal modeling enables actionable insights for urban planning, commercial site selection, and intervention analysis.