[3O2-IS-3] Agents

11:30 AM - 11:45 AM JST(2:30 AM - 2:45 AM UTC)

[3O2-IS-3-03] Quantifying Social Impact in Innovation-Driven MatchingAn LLM-Based Matching Methodology for Complex Social Systems

〇Arisa Morozumi¹, Hisashi Hayashi¹ (1. Advanced Institute of Industrial Technology)

regular

Keywords:

Large Language Models,Human-in-the-Loop,Social Impact Quantification,Combinatorial Optimization,Regional Revitalization

Matching urban professionals with regional needs presents an ill-structured optimization challenge where success depends on capturing high-context strategic intent. Standard models often fail in this domain because they cannot effectively extract and formalize the implicit requirements of human producers into actionable constraints. This paper proposes a novel matching methodology based on a "Hierarchical Separation" architecture, which decouples the Intention Extraction Layer (utilizing LLMs) from the Optimization Layer. We validate this approach through a real-world case study in Gotsu City, algorithmically reconstructing an expert producer’s "story arc." Experimental results demonstrate that: (1) standalone models fail to satisfy strategic requirements due to their inability to formalize implicit producer intent, and (2) our proposed methodology generates matching scores that significantly correlate with expert intuition. This research contributes a scalable engineering methodology for automating complex social system design by utilizing the LLM as a high-precision encoder to transform qualitative intentions into objective parameters for theory-based optimization.