This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
MedKGent: A Large Language Model Agent Framework for Constructing Temporally Evolving Medical Knowledge Graph
Created by
Haebom
Author
Duzhen Zhang, Zixiao Wang, Zhong-Zhi Li, Yahan Yu, Shuncheng Jia, Jiahua Dong, Haotian Xu, Xing Wu, Yingying Zhang, Tielin Zhang, Jie Yang, Xiuying Chen, Le Song
Outline
This paper presents MedKGent, an LLM agent framework, as a way to efficiently build a knowledge graph (KG) to address the rapidly increasing volume of medical literature. Using over 10 million PubMed abstracts published from 1975 to 2023, we simulate the daily generation of medical knowledge across time periods. The KG is incrementally built using two specialized agents—an extraction agent and a composition agent—based on the Qwen2.5-32B-Instruct model. The extraction agent identifies knowledge triples and assigns confidence scores, while the composition agent integrates triples and resolves conflicts using confidence scores and timestamps. The final KG contains 156,275 entities and 2,971,384 relational triples, achieving an accuracy of approximately 90%. Experimental results using seven medical question-answering benchmarks demonstrate improved performance compared to existing methods, suggesting its potential for use in diverse applications such as drug repurposing.
Takeaways, Limitations
•
Takeaways:
◦
A method for constructing a medical knowledge graph considering temporal evolution based on LLM agents is presented.
◦
Achieves higher accuracy (approximately 90%) than existing methods.
◦
Performance improvements have been demonstrated in a variety of applications, including medical question answering and drug repurposing.
◦
Improving the quality of KG by extracting and integrating knowledge using confidence scores.
•
Limitations:
◦
There are some dependencies on the Qwen2.5-32B-Instruct model. Generalizability to other LLM models needs to be verified.
◦
Computational resource consumption due to the use of large datasets (over 10 million PubMed abstracts).
◦
Further research is needed on the evolution and maintenance of knowledge graphs over long periods of time.
◦
Further research is needed to determine generalizability across various medical fields.
