Small Language Models Enhanced with Knowledge Graphs for Improved Reasoning

Researchers have developed a neuro-symbolic agentic framework to improve the reasoning abilities of small language models (SLMs) like Gemma 3 and Llama 3.2. This framework uses tool calls for symbolic triplet extraction and expert reasoning via a Relational Graph Convolutional Network (RGCN). While hints from the RGCN improved performance by 1.5-2x over baseline models, the system's effectiveness was limited by the knowledge extraction process and sequential reasoning fragility. The study also identified a "distraction effect" where noisy, self-generated facts could degrade performance. AI

IMPACT This research offers a potential path to more efficient and capable small language models, reducing reliance on costly large models.

RANK_REASON The cluster contains an academic paper detailing a new methodology for enhancing small language models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• CLUTRR
• Gemma 3
• Hugging Face
• Konstantinos Bougiatiotis
• Llama 3.2

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →