BEST-KAG: Multimodal Knowledge Graph and LLM Framework for Building Engineering Standards QA
A novel framework named BEST-KAG (Knowledge-Augmented Generation for Building Engineering Standards) has been introduced to improve the process of answering questions related to building engineering standards. This framework, outlined in an arXiv paper (2608.11244), tackles shortcomings in current workflows that depend on keyword-based document searches and manual interpretation of clauses, which do not facilitate multi-clause reasoning, multimodal knowledge use, or traceable evidence at the clause level. BEST-KAG features a multimodal knowledge graph (MKG) for a cohesive representation of document structures and diverse standard knowledge, a rule-LLM hybrid for scalable knowledge extraction, and a graph-retrieval system. It has generated an extensive knowledge graph encompassing 251 building engineering standards, with 171,652 nodes and 310,914 edges, marking a significant advancement for the architecture and engineering industries by enhancing standard compliance and safety assessments through large language models and knowledge graphs.
Key facts
- BEST-KAG stands for Knowledge-Augmented Generation for Building Engineering Standards.
- The framework is presented in a paper on arXiv with identifier 2608.11244.
- It addresses limitations in existing standard application workflows, including keyword-based retrieval and manual cross-clause interpretation.
- BEST-KAG introduces a multimodal knowledge graph (MKG) for unified representation of document hierarchy and heterogeneous standard knowledge.
- It uses a rule-LLM hybrid knowledge construction pipeline for scalable multimodal knowledge extraction.
- The framework created a large knowledge graph with 251 building engineering standards, 171,652 nodes, and 310,914 edges.
- The framework supports multi-clause reasoning, multimodal knowledge utilization, and traceable clause-level evidence linkage.
- The development is aimed at enhancing building safety and sustainability through better standard compliance.
Entities
—