Saved in:
Bibliographic Details
Main Authors: Anuyah, Sydney, Kaushik, Mehedi Mahmud, Dai, Hao, Shiradkar, Rakesh, Durresi, Arjan, Chakraborty, Sunandan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.15429
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908780596297728
author Anuyah, Sydney
Kaushik, Mehedi Mahmud
Dai, Hao
Shiradkar, Rakesh
Durresi, Arjan
Chakraborty, Sunandan
author_facet Anuyah, Sydney
Kaushik, Mehedi Mahmud
Dai, Hao
Shiradkar, Rakesh
Durresi, Arjan
Chakraborty, Sunandan
contents Large Language Models (LLMs) generate fluent answers but can struggle with trustworthy, domain-specific reasoning. We evaluate whether domain knowledge graphs (KGs) improve Retrieval-Augmented Generation (RAG) for healthcare by constructing three PubMed-derived graphs: $\mathbb{G}_1$ (T2DM), $\mathbb{G}_2$ (Alzheimer's disease), and $\mathbb{G}_3$ (AD+T2DM). We design two probes: Probe 1 targets merged AD T2DM knowledge, while Probe 2 targets the intersection of $\mathbb{G}_1$ and $\mathbb{G}_2$. Seven instruction-tuned LLMs are tested across retrieval sources {No-RAG, $\mathbb{G}_1$, $\mathbb{G}_2$, $\mathbb{G}_1$ + $\mathbb{G}_2$, $\mathbb{G}_3$, $\mathbb{G}_1$+$\mathbb{G}_2$ + $\mathbb{G}_3$} and three decoding temperatures. Results show that scope alignment between probe and KG is decisive: precise, scope-matched retrieval (notably $\mathbb{G}_2$) yields the most consistent gains, whereas indiscriminate graph unions often introduce distractors that reduce accuracy. Larger models frequently match or exceed KG-RAG with a No-RAG baseline on Probe 1, indicating strong parametric priors, whereas smaller/mid-sized models benefit more from well-scoped retrieval. Temperature plays a secondary role; higher values rarely help. We conclude that precision-first, scope-matched KG-RAG is preferable to breadth-first unions, and we outline practical guidelines for graph selection, model sizing, and retrieval/reranking. Code and Data available here - https://github.com/sydneyanuyah/RAGComparison
format Preprint
id arxiv_https___arxiv_org_abs_2601_15429
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Domain-Specific Knowledge Graphs in RAG-Enhanced Healthcare LLMs
Anuyah, Sydney
Kaushik, Mehedi Mahmud
Dai, Hao
Shiradkar, Rakesh
Durresi, Arjan
Chakraborty, Sunandan
Computation and Language
Large Language Models (LLMs) generate fluent answers but can struggle with trustworthy, domain-specific reasoning. We evaluate whether domain knowledge graphs (KGs) improve Retrieval-Augmented Generation (RAG) for healthcare by constructing three PubMed-derived graphs: $\mathbb{G}_1$ (T2DM), $\mathbb{G}_2$ (Alzheimer's disease), and $\mathbb{G}_3$ (AD+T2DM). We design two probes: Probe 1 targets merged AD T2DM knowledge, while Probe 2 targets the intersection of $\mathbb{G}_1$ and $\mathbb{G}_2$. Seven instruction-tuned LLMs are tested across retrieval sources {No-RAG, $\mathbb{G}_1$, $\mathbb{G}_2$, $\mathbb{G}_1$ + $\mathbb{G}_2$, $\mathbb{G}_3$, $\mathbb{G}_1$+$\mathbb{G}_2$ + $\mathbb{G}_3$} and three decoding temperatures. Results show that scope alignment between probe and KG is decisive: precise, scope-matched retrieval (notably $\mathbb{G}_2$) yields the most consistent gains, whereas indiscriminate graph unions often introduce distractors that reduce accuracy. Larger models frequently match or exceed KG-RAG with a No-RAG baseline on Probe 1, indicating strong parametric priors, whereas smaller/mid-sized models benefit more from well-scoped retrieval. Temperature plays a secondary role; higher values rarely help. We conclude that precision-first, scope-matched KG-RAG is preferable to breadth-first unions, and we outline practical guidelines for graph selection, model sizing, and retrieval/reranking. Code and Data available here - https://github.com/sydneyanuyah/RAGComparison
title Domain-Specific Knowledge Graphs in RAG-Enhanced Healthcare LLMs
topic Computation and Language
url https://arxiv.org/abs/2601.15429