SAGA: Schema-Aware Grounding for Agentic Text-to-SPARQL Generation

arXiv cs.AI · 6 hours ago

Researchers introduced SAGA, a framework that improves how AI agents convert natural language questions into SPARQL queries by using knowledge base schema information to filter property candidates and reduce empty results. The method achieved the highest F1 scores on all nine benchmark settings tested across Wikidata and Freebase databases. This approach reduces the need for large language models to generate correct database queries in a single attempt by incorporating schema constraints during interactive reasoning.