Reasoning Over Space: Enabling Geographic Reasoning for LLM-Based Generative Next POI Recommendation

2026-01-08

arXiv.org

AI Summary

This work addresses the limited ability of existing large language model (LLM)-based recommender systems to effectively model geographic signals, which hinders their performance in mobile and local service scenarios. To overcome this, we propose a geographically aware generative framework for next point-of-interest recommendation that encodes spatial semantics through hierarchical Spatial Semantic IDs (SIDs), explicitly models user mobility logic via a three-stage Mobility Chain-of-Thought (Mobility CoT), and refines the generation process using spatially guided reinforcement learning. Our approach is the first to deeply integrate explicit geographic reasoning into the LLM-based recommendation pipeline. Experiments on three LBSN datasets demonstrate over a 10% improvement in hit rate compared to the strongest LLM baselines, with superior cross-city generalization capability even at smaller model scales.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Generative recommendation with large language models (LLMs) reframes prediction as sequence generation, yet existing LLM-based recommenders remain limited in leveraging geographic signals that are crucial in mobility and local-services scenarios. Here, we present Reasoning Over Space (ROS), a framework that utilizes geography as a vital decision variable within the reasoning process. ROS introduces a Hierarchical Spatial Semantic ID (SID) that discretizes coarse-to-fine locality and POI semantics into compositional tokens, and endows LLM with a three-stage Mobility Chain-of-Thought (CoT) paradigm that models user personality, constructs an intent-aligned candidate space, and performs locality informed pruning. We further align the model with real world geography via spatial-guided Reinforcement Learning (RL). Experiments on three widely used location-based social network (LBSN) datasets show that ROS achieves over 10% relative gains in hit rate over strongest LLM-based baselines and improves cross-city transfer, despite using a smaller backbone model.

Problem

• geographic reasoning
• LLM-based recommendation
• next POI recommendation
• spatial signals
• location-based social networks

Innovation

• Geographic Reasoning
• Hierarchical Spatial Semantic ID
• Mobility Chain-of-Thought
• Spatial-guided Reinforcement Learning
• Generative POI Recommendation