[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

BifrostRAG: Bridging Dual Knowledge Graphs for Multi-Hop Question Answering in Construction Safety

Created by
  • Haebom

Author

Yuxin Zhang (Department of Construction Science, College of Architecture, Texas A&M University, College Station, USA), Xi Wang (Department of Construction Science, College of Architecture, Texas A&M University, College Station, USA), Mo Hu (Department of Construction Science, College of Architecture, Texas A&M University, College Station, USA), Zhenyu Zhang (Department of Construction Science, College of Architecture, Texas A&M University, College Station, USA)

Outline

This paper addresses the application of information retrieval and question answering from safety regulations to automated construction compliance verification. To address the existing challenges posed by the linguistic and structural complexity of regulatory texts, we propose BifrostRAG, a dual-graph-based RAG system that explicitly models linguistic relationships (Entity Network Graph) and document structure (Document Navigator Graph). BifrostRAG enables large-scale language models to infer both the meaning and structure of texts through a hybrid retrieval mechanism that combines graph traversal and vector-based semantic retrieval. Evaluation results on the multi-step question dataset demonstrate that BifrostRAG achieves 92.8% precision, 85.5% recall, and 87.3% F1-score, significantly outperforming existing best-performing vector-only and graph-only RAG baseline systems. Error analysis further highlights the comparative advantage of the hybrid method over single-mode RAGs. These results establish BifrostRAG as a powerful knowledge engine for LLM-based compliance checking, and its dual-graph hybrid search mechanism provides a transferable blueprint for navigating complex technical documents in knowledge-intensive engineering domains.

Takeaways, Limitations

•
Takeaways:
◦
Proposal of a new RAG system, BifrostRAG, that significantly improves multi-step question-answering performance in complex technical documents (e.g. safety regulations).
◦
Demonstrate the effectiveness of a hybrid retrieval mechanism based on a dual graph (linguistic relations and document structure).
◦
Presents important advances in the development of LLM-based automated compliance verification systems.
◦
Provides a transferable architecture applicable to a wide range of knowledge-intensive engineering fields.
•
Limitations:
◦
Only the evaluation results for a specific domain (construction safety regulations) are presented, so further research is needed on generalizability.
◦
Lack of detailed description of the size and diversity of the dataset used.
◦
Further research is needed on the applicability and performance for other types of complex documents.
👍
Made with Slashpage