LLM-Assisted Knowledge Graph for Steel Industry VOC Governance
Chat-ISV is an innovative Q&A system that uses multiple agents and a knowledge graph to process unstructured scientific information about volatile organic compounds (VOCs) in the steel industry. Developed by researchers, it works with a carefully selected dataset to create a Neo4j knowledge graph that includes 27,180 nodes and 81,779 semantic connections. The system employs various techniques like prompt-constrained extraction and multi-agent routing, along with interactive subgraph visualization. In benchmark tests involving 400 blind evaluations by experts, the optimization of topology significantly reduced isolated nodes from 57% to just 4.08%, while Chat-ISV maintained impressive factual accuracy, lowering the chances of inaccuracies in rare industrial questions.
Key facts
- Chat-ISV is a knowledge graph enhanced multi-agent Q&A system for steel-industry VOCs governance.
- The knowledge graph uses Neo4j with 27,180 nodes and 81,779 semantic edges.
- Topology optimization reduced isolated nodes from 57% to 4.08%.
- 400 expert blind evaluations were conducted.
- The system parses a curated steel-industry VOCs literature corpus.
- It combines prompt-constrained extraction, chunk-centered topology optimization, multi-agent routing, source-backtracking retrieval, local literature retrieval, open-domain knowledge access, and interactive subgraph visualization.
- The system addresses hallucination risks in general large language models for low-frequency industrial questions.
- The research is published on arXiv with ID 2605.27071.
Entities
Institutions
- arXiv
- Neo4j