Heterogeneous Region Embedding with Prompt Learning

The prevalence of region-based urban data has opened new possibilities for exploring correlations among regions to improve planning and smart-city solutions. Region embedding, which plays a critical role in this endeavor, faces significant challenges related the varying nature city effectiveness downstream applications. In paper, we propose novel framework, HREP (Heterogeneous Embedding with Prompt learning), addresses both intra-region inter-region through two key modules: Heterogeneous (HRE) prompt learning different tasks. HRE module constructs heterogeneous region graph based on three categories data, capturing contexts such as human mobility geographic neighbors, intraregion POI (Point-of-Interest) information. We use relation-aware embedding learn relation embeddings edge types, introduce selfattention capture global regions. Additionally, develop an attention-based fusion integrate shared information types correlations. To enhance tasks, incorporate learning, specifically prefix-tuning, guides tasks results better prediction performance. Our experiment real-world datasets demonstrate that our proposed model outperforms state-of-the-art methods.