GRISP: Guided Recurrent IRI Selection over SPARQL Skeletons

2026-04-24 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

GRISP (Guided Recurrent IRI Selection over SPARQL Skeletons) is a new SPARQL-based question-answering method for knowledge graphs, utilizing a fine-tuned small language model (SLM). The method first generates a natural-language SPARQL query skeleton using the SLM, then iteratively re-ranks and selects knowledge graph items to replace natural-language placeholders, guided by knowledge graph constraints. The SLM is jointly trained on skeleton generation and list-wise re-ranking data derived from standard question-query pairs. Evaluated on Wikidata and Freebase benchmarks, GRISP achieves superior results compared to other state-of-the-art methods in comparable settings, while being significantly faster than agentic zero-shot approaches. The system incorporates beam search for skeleton generation, a search-and-select mechanism for IRI replacement, backtracking for error recovery, and knowledge graph guidance to constrain candidate IRIs.

Key takeaway

For AI Engineers developing knowledge graph question-answering systems, GRISP offers a compelling approach to balance performance and efficiency. You should consider adopting its generate-then-retrieve architecture with a fine-tuned SLM, especially if your project has sufficient training data and requires faster inference than agentic LLM methods. This method allows for deployment on single consumer GPUs, making it accessible for many applications.

Key insights

GRISP uses a fine-tuned SLM for efficient, accurate knowledge graph question answering via iterative SPARQL skeleton generation and IRI selection.

Principles

• Generate SPARQL skeletons first, then retrieve IRIs.
• Iterative re-ranking with KG constraints improves retrieval.
• Backtracking enhances robustness in IRI selection.

Method

GRISP generates SPARQL skeletons via beam search, then iteratively replaces placeholders with IRIs using SLM-based re-ranking, guided by knowledge graph constraints, and includes backtracking for error recovery.

In practice

• Fine-tune SLMs for KGQA tasks.
• Implement beam search for skeleton generation.
• Utilize knowledge graph constraints for IRI filtering.

Topics

• GRISP
• Knowledge Graph Question Answering
• SPARQL Skeletons
• Small Language Models
• IRI Selection

Code references

• bastiscode/grasp

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Leveraging LLM-GNN Integration for Open-World Question Answering over Knowledge Graphs — Integrating GNN-predicted candidates and KG context into LLM prompts significantly improves open-world question answering over incomplete knowledge graphs.
• SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale — Dynamic, graph-based structural retrieval significantly enhances LLM agent skill selection and performance over static methods.
• Neural-Symbolic Logic Query Answering in Non-Euclidean Space — Hyperbolic GNNs with learned curvature effectively capture hierarchical logic query structures for knowledge graph reasoning.
• BLINKG: A Benchmark for LLM-Integrated Knowledge Graph Generation — BLINKG provides a standardized benchmark to evaluate Large Language Models' effectiveness in mapping data schemes to ontology concepts for Knowledge Graph generation.
• LogosKG: Hardware-Optimized Scalable and Interpretable Knowledge Graph Retrieval — LogosKG offers scalable, interpretable k-hop retrieval for large knowledge graphs via hardware-optimized symbolic operations.
• Training Azerbaijani language models on Amazon SageMaker AI — Optimizing LLM training for low-resource, morphologically rich languages requires custom tokenization and GPU memory optimizations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.