arXiv cs.CL· Nikit Srivastava, Daniel Vollmers, Ren\'e Speck, Nikolaos Karalis, Hamada M. Zahera, Axel-Cyrille Ngonga Ngomo·· 13 小时前AI 评分31
MARS：面向 KGQA 的多跳自适应检索与 SPARQL 生成方法
MARS: Multi-hop Adaptive Retrieval and SPARQL Generation for KGQA
AI 导读
MARS 是一种无需模型微调的可扩展知识图谱问答（KGQA）方法，通过结构化检索将问题实体链接到 KG 并迭代获取下一跳信息，每步自行决定继续遍历还是生成最终 SPARQL 查询。该方法在三个 KGQA 基准上跨多个 LLM 及多语言设置评估，性能与 SOTA 方法相当，同时保持高效可扩展，代码与资源已公开。
正文 · 原文
Abstract:Large language models (LLMs) have demonstrated strong reasoning performance, but their tendency to hallucinate limits their reliability in knowledge-intensive tasks requiring up-to-date and grounded information. Combining knowledge graphs (KGs) with LLMs facilitates the use of explicit symbolic knowledge that can be continuously updated without costly fine-tuning, while benefiting from rapidly advancing LLM reasoning. We propose MARS, a scalable knowledge graph question answering (KGQA) approach that requires no model fine-tuning. Rather than relying on open-ended agentic exploration, MARS performs a structured retrieval procedure that links question entities to the KG and iteratively retrieves relevant next-hop information. At each step, MARS decides whether to continue graph traversal or to generate the final SPARQL query, allowing the model to adapt the retrieval depth to the question while keeping the overall pipeline more predictable than fully agentic approaches. We evaluate MARS on three established KGQA benchmarks across several LLMs and settings, including multilingual evaluation, and provide insights through ablation studies and error analysis. Our approach achieves competitive performance relative to state-of-the-art methods while remaining efficient and scalable. The evaluation results, code and resources are publicly available: this https URL.
| Comments: | EKAW 2026 (this https URL) |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.14561 [cs.CL] |
| (or arXiv:2607.14561v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.14561 arXiv-issued DOI via DataCite |
|
| Related DOI: | https://doi.org/10.1007/978-3-032-41614-8_9
DOI(s) linking to related resources |
Submission history
From: Nikit Srivastava [view email]
[v1]
Thu, 16 Jul 2026 04:40:05 UTC (98 KB)
[v2]
Mon, 27 Jul 2026 09:22:09 UTC (98 KB)
[v3]
Wed, 21 Oct 2026 08:59:02 UTC (98 KB)
来源：arXiv cs.CL · arxiv.org