MARS: Multi-hop Adaptive Retrieval and SPARQL Generation for KGQA

2026-07-16 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

MARS is a scalable Knowledge Graph Question Answering (KGQA) approach that integrates Large Language Models (LLMs) with knowledge graphs (KGs) without requiring LLM fine-tuning. It addresses LLM hallucination in knowledge-intensive tasks by employing a structured, multi-hop retrieval procedure. MARS links question entities to the KG and iteratively retrieves relevant next-hop information, adaptively deciding whether to continue graph traversal or generate the final SPARQL query. This method offers a more predictable pipeline than fully agentic approaches. Evaluated on three established KGQA benchmarks, including multilingual settings, MARS achieves competitive performance against state-of-the-art methods while remaining efficient and scalable.

Key takeaway

For Machine Learning Engineers developing reliable knowledge-intensive applications, MARS offers a compelling alternative to costly LLM fine-tuning. You should consider integrating this structured, multi-hop retrieval approach to leverage explicit symbolic knowledge from KGs, ensuring grounded responses and reducing hallucination. This method provides a predictable and scalable pathway for building efficient Knowledge Graph Question Answering systems.

Key insights

MARS enables reliable KGQA by combining LLMs with structured, adaptive multi-hop retrieval over KGs without fine-tuning.

Principles

• KGs enhance LLM reliability for knowledge tasks.
• Structured retrieval improves predictability over agentic exploration.
• Adaptive retrieval depth optimizes query generation.

Method

MARS links question entities to a KG, iteratively retrieves next-hop data, then decides to continue traversal or generate a SPARQL query.

In practice

• Apply for knowledge-intensive tasks needing grounded information.
• Use for scalable and efficient KGQA.
• Implement in multilingual question answering systems.

Topics

• Knowledge Graph Question Answering
• Large Language Models
• SPARQL Generation
• Multi-hop Retrieval
• Knowledge Graphs
• Natural Language Processing

Code references

• dice-group/mars-kgqa

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

Related on AIssential

• RSF-GLLM: Bridging the Semantic Gap in Multi-Hop Knowledge Graph QA via Recurrent Soft-Flow and Decoupled LLM Generation — RSF-GLLM decouples differentiable graph reasoning from LLM answer generation to bridge semantic gaps in multi-hop KG QA.
• SAGA: Schema-Aware Grounding for Agentic Text-to-SPARQL Generation — Schema-aware grounding significantly enhances Text-to-SPARQL generation by preventing type-blind query construction in LLM agents.
• GRASP: Graph Agentic Search over Propositions for Multi-hop Question Answering — GRASP optimizes multi-hop QA by combining a hierarchical proposition-based graph with dependency-aware agentic planning for efficiency and accuracy.
• Simorgh at SemEval-2026 task 7: Region-Aware Hybrid Retrieval for Low-Resource Cultural Reasoning in Multilingual Question Answering — A region-aware hybrid retrieval system enhances multilingual QA by prioritizing culturally relevant evidence for efficient, stable inference.
• Executable Schema Contracts: From Automatic Ingestion to Multi-Source Retrieval — An automatically induced, executable schema can unify knowledge graph ingestion and multi-source query-time retrieval.
• QUBO-Optimized Evidence Selection for Retrieval-Augmented Question Answering with Unconventional Solvers — Formulating RAG evidence selection as a QUBO problem enables efficient, structured context retrieval separate from LLM answer generation.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.