ToPT: Task-Oriented Prompt Tuning for Urban Region Representation Learning

TL;DR

This work tackles urban region representation learning from heterogeneous data by addressing the limitations of task-agnostic two-stage methods. It introduces ToPT, a two-stage framework with Spatial-aware Region Embedding Learning (SREL) that uses a Graphormer-based fusion with distance and centrality priors, and Task-aware Prompting (Prompt4RE) that extracts task-specific prompts from frozen multimodal large language models and aligns them to region embeddings via cross-attention to form enhanced representations. Empirical results on Chicago demonstrate state-of-the-art performance across crime prediction, check-in forecasting, and service call estimation, with improvements reaching up to $64.2\%$. The approach generalizes across different MLLMs and cities, highlighting the importance of explicit spatial priors and task-semantic alignment for robust, transferable urban region representations.

Abstract

Learning effective region embeddings from heterogeneous urban data underpins key urban computing tasks (e.g., crime prediction, resource allocation). However, prevailing two-stage methods yield task-agnostic representations, decoupling them from downstream objectives. Recent prompt-based approaches attempt to fix this but introduce two challenges: they often lack explicit spatial priors, causing spatially incoherent inter-region modeling, and they lack robust mechanisms for explicit task-semantic alignment. We propose ToPT, a two-stage framework that delivers spatially consistent fusion and explicit task alignment. ToPT consists of two modules: spatial-aware region embedding learning (SREL) and task-aware prompting for region embeddings (Prompt4RE). SREL employs a Graphormer-based fusion module that injects spatial priors-distance and regional centrality-as learnable attention biases to capture coherent, interpretable inter-region interactions. Prompt4RE performs task-oriented prompting: a frozen multimodal large language model (MLLM) processes task-specific templates to obtain semantic vectors, which are aligned with region embeddings via multi-head cross-attention for stable task conditioning. Experiments across multiple tasks and cities show state-of-the-art performance, with improvements of up to 64.2\%, validating the necessity and complementarity of spatial priors and prompt-region alignment. The code is available at https://github.com/townSeven/Prompt4RE.git.

Figures (2)

• Figure 1: Our Model Framework
• Figure 2: Impact of MLLMs.