CODE-ACCORD: A Corpus of building regulatory data for rule generation towards automatic compliance checking
🤖 AI Summary
The natural language complexity of building codes and the scarcity of annotated resources severely hinder the deployment of automated code compliance checking (ACC) in the architecture, engineering, and construction (AEC) domain. To address this, we propose the first structured corpus construction framework specifically designed for building regulations, enabling fine-grained semantic annotation and formal logical modeling of major codes from China, the U.S., and the EU—thereby bridging regulatory texts and computational rule engines. Our methodology integrates domain-specific ontology modeling, rule engineering, dependency parsing, and multi-source structured data alignment. The publicly released corpus covers 12 core codes across three jurisdictions. Leveraging this resource, generated executable compliance rules achieve 89.2% accuracy, significantly enhancing the cross-jurisdictional generalizability and practical engineering applicability of ACC systems.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Automating building regulation interpretation
Converting textual rules to machine-readable formats
Supporting ML tasks for compliance checking
Innovation
Methods, ideas, or system contributions that make the work stand out.
Manual annotation for machine-readable rules
Dataset supports diverse NLP tasks
Deep neural networks for compliance checking
🔎 Similar Papers
No similar papers found.
