Relation-aware dynamic knowledge subgraph enhanced large language models for medical multiple-choice question answering
4 affiliations
- 1Gansu Police College, Lanzhou, China
- 2Police Embodied Intelligence and Unmanned Systems Innovation Center, Lanzhou, China
- 3The Second Hospital of Lanzhou University, Lanzhou, China
- 4Gansu Provincial Public Security Department, Lanzhou, China
Abstract
Medical multiple-choice question answering requires models to simultaneously understand question semantics, candidate-answer differences, and structured relations among medical entities in complex clinical contexts. To address the limitations that existing knowledge-enhanced methods often adopt question-level unified modeling and thus struggle to characterize local relational differences among different candidate options, while conventional low-rank adaptation lacks relation-pattern adaptivity, this paper proposes a knowledge-enhanced large language model framework that combines option-level knowledge subgraph representation fusion with relation-aware dynamic LoRA adaptation. The proposed method first constructs a local medical knowledge subgraph for the question stem and each candidate answer separately, and then injects structured knowledge into candidate representations through a graph encoder and gated fusion mechanism. Subsequently, multiple learned LoRA expert branches are dynamically combined through candidate-conditioned routing based on the relational information encoded in the enhanced candidate representations, enabling the model to obtain more suitable parameter update paths. The experimental results are reported as the mean and standard deviation over three random-seed runs. On the MedQA dataset, the proposed method achieves Accuracy scores of 62.90%±0.82% and 62.84%±1.04%, together with F1-Score values of 62.97%±1.07% and 62.79%±0.78%, using Llama3 8B and Qwen 7B, respectively. On the MedMCQA dataset, the corresponding Accuracy scores are 61.04%±0.30% and 58.59%±0.65%, while the F1-Score values reach 61.21%±0.54% and 57.77%±0.57%, respectively. Experimental results show that the proposed method can effectively improve knowledge understanding, candidate discrimination, and reasoning stability in medical multiple-choice question answering.
Related
- MEMOIR-VLM—a multimodal vision-language model for Alzheimer's disease classification and question answering
- Answering clinicians' questions over trial evidence tables with verifiable, feedback-driven language models
- Entity-centric evaluation of large language model responses for medical question-answering tasks
- SportRGDET: Scene prior and relation graph enhanced detection transformer for sports scenes
- Towards a knowledge-enhanced single-cell foundation model
BibTeX
@article{ma2026relationaware,
title = {{Relation-aware dynamic knowledge subgraph enhanced large language models for medical multiple-choice question answering}},
author = {Yu Ma and Tianning Di and Jiangpeng Xu and Yue Wu},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0349628},
url = {https://doi.org/10.1371/journal.pone.0349628},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Relation-aware dynamic knowledge subgraph enhanced large language models for medical multiple-choice question answering
AU - Yu Ma
AU - Tianning Di
AU - Jiangpeng Xu
AU - Yue Wu
PY - 2026
DA - 2026/10/16
JO - PLOS One
DO - 10.1371/journal.pone.0349628
UR - https://doi.org/10.1371/journal.pone.0349628
AB - Medical multiple-choice question answering requires models to simultaneously understand question semantics, candidate-answer differences, and structured relations among medical entities in complex clinical contexts. To address the limitations that existing knowledge-enhanced methods often adopt question-level unified modeling and thus struggle to characterize local relational differences among different candidate options, while conventional low-rank adaptation lacks relation-pattern adaptivity, this paper proposes a knowledge-enhanced large language model framework that combines option-level knowledge subgraph representation fusion with relation-aware dynamic LoRA adaptation. The proposed method first constructs a local medical knowledge subgraph for the question stem and each candidate answer separately, and then injects structured knowledge into candidate representations through a graph encoder and gated fusion mechanism. Subsequently, multiple learned LoRA expert branches are dynamically combined through candidate-conditioned routing based on the relational information encoded in the enhanced candidate representations, enabling the model to obtain more suitable parameter update paths. The experimental results are reported as the mean and standard deviation over three random-seed runs. On the MedQA dataset, the proposed method achieves Accuracy scores of 62.90%±0.82% and 62.84%±1.04%, together with F1-Score values of 62.97%±1.07% and 62.79%±0.78%, using Llama3 8B and Qwen 7B, respectively. On the MedMCQA dataset, the corresponding Accuracy scores are 61.04%±0.30% and 58.59%±0.65%, while the F1-Score values reach 61.21%±0.54% and 57.77%±0.57%, respectively. Experimental results show that the proposed method can effectively improve knowledge understanding, candidate discrimination, and reasoning stability in medical multiple-choice question answering.
SN - 1932-6203
ER -