Reasoning Over Space Improves Location Recommendations
Researchers led by Dongyi Lv on Jan. 8, 2026 present Reasoning Over Space (ROS), a framework that integrates geographic signals into LLM-based generative recommendation for mobility and local services. ROS introduces a Hierarchical Spatial Semantic ID and a three-stage Mobility Chain-of-Thought, aligned via spatial-guided reinforcement learning. Evaluated on three LBSN datasets, ROS delivers over 10% relative hit-rate gains and improved cross-city transfer with a smaller backbone.
Key Points
- 1Introduces Hierarchical Spatial Semantic ID and three-stage Mobility Chain-of-Thought for geography-aware recommendation
- 2Demonstrates over 10% relative hit-rate gains on three LBSN datasets versus LLM baselines
- 3Enables improved cross-city transfer and locality-informed pruning for practitioners building mobility recommenders
Scoring Rationale
Novel, practical LLM-based method with strong empirical gains; limited to LBSN datasets and single preprint source.
Sources
Public references used for this report.
Practice with real Ride-Hailing data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ride-Hailing problems