Scholarlink

<p>The proliferation of urban Point of Interest (POI) data has created unprecedented opportunities for intelligent resource allocation and regional economic strategies. Strategic group analysis—identifying entities with similar market positioning, target demographics, and competitive strategies—has proven critical for urban planning and economic development. However, traditional approaches operate at the brand level, lacking the spatial granularity required for site-specific decisions such as optimal business selection for vacant land. We introduce POI-level strategic group discovery, which transitions from brand-level aggregates to location-specific precision by analyzing semantic attributes and spatial contexts at precise geographic coordinates. This task faces three critical challenges: (1) the absence of objective ground truth labels and prohibitive annotation costs create severe data scarcity; (2) modeling multi-dimensional strategic relationships demands frameworks capturing higher-order group-level interactions beyond pairwise competition; (3) conventional supervised learning fails in few-shot scenarios with limited labeled examples per group. To address these challenges, we propose a comprehensive framework with three key components: an LLM-based annotation pipeline that automatically generates high-quality labels through semantic reasoning; a heterogeneous urban hypergraph formulation where four types of hyperedges capture multi-dimensional relationships; and the Hypergraph Conditional Neural Process (HCNP), a meta-learning framework enabling robust few-shot discovery through bi-level encoding and auxiliary tasks. We construct three city-wide benchmark datasets integrating multi-source urban data with LLM-annotated labels. Extensive experiments demonstrate significant performance improvements over existing methods, offering valuable insights for urban planning and sustainable economic development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Few-shot urban strategic group discovery with LLM annotation and hypergraph learning

  • 期刊论文

  • 发表日期：

    2026年1月1日

查看全文
  • Hongru Lu,
  • Jiaqi Jiang,
  • Xiao Jing,
  • Xin Lin,
  • Haoxiang Zhang,
  • Zhiang Wu

摘要

The proliferation of urban Point of Interest (POI) data has created unprecedented opportunities for intelligent resource allocation and regional economic strategies. Strategic group analysis—identifying entities with similar market positioning, target demographics, and competitive strategies—has proven critical for urban planning and economic development. However, traditional approaches operate at the brand level, lacking the spatial granularity required for site-specific decisions such as optimal business selection for vacant land. We introduce POI-level strategic group discovery, which transitions from brand-level aggregates to location-specific precision by analyzing semantic attributes and spatial contexts at precise geographic coordinates. This task faces three critical challenges: (1) the absence of objective ground truth labels and prohibitive annotation costs create severe data scarcity; (2) modeling multi-dimensional strategic relationships demands frameworks capturing higher-order group-level interactions beyond pairwise competition; (3) conventional supervised learning fails in few-shot scenarios with limited labeled examples per group. To address these challenges, we propose a comprehensive framework with three key components: an LLM-based annotation pipeline that automatically generates high-quality labels through semantic reasoning; a heterogeneous urban hypergraph formulation where four types of hyperedges capture multi-dimensional relationships; and the Hypergraph Conditional Neural Process (HCNP), a meta-learning framework enabling robust few-shot discovery through bi-level encoding and auxiliary tasks. We construct three city-wide benchmark datasets integrating multi-source urban data with LLM-annotated labels. Extensive experiments demonstrate significant performance improvements over existing methods, offering valuable insights for urban planning and sustainable economic development.