UniQGen: LLM Agents for Graph Query Generation Across Languages
UniQGen is a constraint-based framework using LLM agents to dynamically extract and refine graph query clauses into executable, intent-aligned queries across query languages. It extends the Chase & Backchase algorithm family with dynamic reasoning over query constraints and LLM interaction for quality estimation. The framework addresses the underexplored area of Cypher and property graphs in KGQA, targeting unified query generation for industry settings. Evaluated on Freebase deployed on Amazon Neptune with benchmarks GraphQ, GrailQA, and WebQSP, UniQGen outperforms state-of-the-art methods.
Key facts
- UniQGen is a constraint-based framework for graph query generation.
- It employs LLM agents to extract and refine query clauses.
- The method extends Chase & Backchase algorithms.
- It supports Cypher and property graphs.
- Evaluated on Freebase with Amazon Neptune.
- Benchmarks used: GraphQ, GrailQA, WebQSP.
- UniQGen outperforms state-of-the-art approaches.
- Targets unified KGQA in industry settings.
Entities
Institutions
- Amazon Neptune