GRISP: Guided Recurrent IRI Selection over SPARQL Skeletons

TL;DR

GRISP is a novel method that uses a small language model to generate and refine SPARQL queries from natural language questions, improving question-answering over knowledge graphs.

Contribution

It introduces a fine-tuned language model approach for generating and iteratively refining SPARQL query skeletons for knowledge graph question-answering.

Findings

Outperforms state-of-the-art methods on Wikidata and Freebase benchmarks.

Uses a joint training approach on skeleton generation and re-ranking.

Achieves better accuracy with a small language model.

Abstract

We present GRISP (Guided Recurrent IRI Selection over SPARQL Skeletons), a novel SPARQL-based question-answering method over knowledge graphs based on fine-tuning a small language model (SLM). Given a natural-language question, the method first uses the SLM to generate a natural-language SPARQL query skeleton, and then to re-rank and select knowledge graph items to iteratively replace the natural-language placeholders using knowledge graph constraints. The SLM is jointly trained on skeleton generation and list-wise re-ranking data generated from standard question-query pairs. We evaluate the method on common Wikidata and Freebase benchmarks, and achieve better results than other state-of-the-art methods in a comparable setting.