FIRESPARQL: A LLM-based Framework for SPARQL Query Generation over Scholarly Knowledge Graphs
Published in 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K), 2025
This paper introduces FIRESPARQL, a modular framework for translating natural language questions into SPARQL queries over scholarly knowledge graphs. The framework combines fine-tuned LLMs with optional retrieval-augmented generation and a SPARQL query correction layer. Experiments on the SciQA Benchmark compare zero-shot, one-shot, RAG-enhanced, and fine-tuned configurations, showing that fine-tuning achieves the strongest performance for both query accuracy and query result accuracy.
Recommended citation: Pan, X., de Boer, V., & van Ossenbruggen, J. (2025). FIRESPARQL: A LLM-based Framework for SPARQL Query Generation over Scholarly Knowledge Graphs. arXiv:2508.10467. Accepted at the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K).
Download Paper