通过知识图谱接地增强小型语言模型的推理能力
英文标题：Enhancing Small Language Models Reasoning through Knowledge Graph Grounding
尽管大型语言模型在零样本推理任务中表现优异，但其部署成本高、环境影响大，小型语言模型虽然更可持续，却在复杂逻辑接地时容易出错。本研究提出一种神经符号智能体框架，通过专用工具调用和关系图卷积网络专家提示，提升 Gemma 3 和 Llama 3.2 等小型模型的推理能力。实验表明，专家提示能带来 1.5 至 2 倍的性能增益，但系统仍受限于知识提取瓶颈和顺序推理的脆弱性，噪声事实在某些架构中甚至会产生“干扰效应”。该工作明确了低资源智能体系统中符号接地的难点，为神经符号流水线的迭代验证提供了路线图。
2.1 KB
640 B
612 B
380 B
9.1 KB
32.7 KB
22.8 KB
24.2 KB
34.7 KB
中文
[Submitted on 14 Jul 2026]
标题：通过知识图谱接地增强小型语言模型的推理能力
摘要：尽管大型语言模型（LLMs）在零样本推理方面树立了基准，但其部署仍然成本高昂且对环境造成负担。小型语言模型（SLMs）提供了一种可持续的替代方案，但在需要复杂多跳逻辑接地的任务上容易出错。我们研究了一种神经符号智能体框架，利用 CLUTRR 亲属关系基准来增强 SLM 的推理能力，具体使用了 Gemma 3（1B、4B）和 Llama 3.2（3B）。我们的方法将 SLM 转变为一个极简智能体，利用两种专用工具调用：
extract_facts用于符号三元组提取，以及get_hint用于通过关系图卷积网络（RGCN）进行专家推理。我们在两种配置下评估了这些模型：一种是有真实三元组的预言机（Oracle）场景，另一种是依赖自提取知识的现实（Realistic）场景。我们的结果表明，虽然 RGCN 提供的提示比仅基于故事的基线提供了 1.5 到 2 倍的性能提升，但系统受限于提取瓶颈和顺序演绎脆弱性，早期提取错误在多跳链中不断累积。此外，我们在特定架构中发现了一种“干扰效应”，即自生成的噪声事实尽管有专家提示，却反而降低了性能。这项工作刻画了低资源智能体系统中符号接地所面临的挑战，并为神经符号智能体流水线中的迭代验证提供了路线图。
提交历史
来自：Konstantinos Bougiatiotis [ 查看邮箱 ]
[v1] 2026年7月14日（周二）13:07:46 UTC（1,468 KB）
English / 原文
[Submitted on 14 Jul 2026]
Title:Enhancing Small Language Models Reasoning through Knowledge Graph Grounding
View a PDF of the paper titled Enhancing Small Language Models Reasoning through Knowledge Graph Grounding, by Dimitrios Kelesis and 1 other authors
Abstract:Although large language models (LLMs) have set benchmarks for zero-shot reasoning, their deployment remains cost-prohibitive and environmentally taxing. Small Language Models (SLMs) offer a sustainable alternative, but prone to errors, on tasks requiring complex, multi-hop logical grounding. We investigate a neuro-symbolic agentic framework to enhance the reasoning capabilities of SLMs, specifically Gemma 3 (1B, 4B) and Llama 3.2 (3B), using the CLUTRR kinship benchmark. Our approach transforms the SLM into a minimalist agent utilizing two specialized tool calls: extract_facts for symbolic triplet extraction and get_hint for expert reasoning via a Relational Graph Convolutional Network (RGCN). We evaluate these models across two configurations, both in an Oracle scenario with ground-truth triplets and a Realistic scenario relying on self-extracted knowledge. Our results reveal that while RGCN-derived hints provide a 1.5 - 2x performance gain over story-only baselines, the system is constrained by the extraction bottleneck and sequential deductive fragility, where early extraction errors compound over multi-hop chains. Furthermore, we identify a "distraction effect" in specific architectures where noisy, self-generated facts degrade performance despite the presence of expert hints. This work characterizes the challenges of symbolic grounding in low-resource agentic systems and provides a roadmap for iterative verification in neuro-symbolic agentic pipelines.
Submission history
From: Konstantinos Bougiatiotis [
]
[v1]
Tue, 14 Jul 2026 13:07:46 UTC (1,468 KB)