Combining Explicit Knowledge with Large-language Model Helps Medical Data Integration

5. April 2023

Short Paper with Emory University and Google Research Accepted at SIGIR-2023

Medical decision-making processes can be enhanced by the support of comprehensive biomedical knowledge bases, which can be obtained by fusing biomedical knowledge graphs constructed from different sources via a uniform index system.
The approach proposed in the paper addresses this task by leveraging large language models and hierarchy-oriented prompts techniques, to automatically fuse multiple biomedical KGs into a standard hierarchical index system, even with only limited labeled data.
Performance comparison on the newly constructed benchmark datasets demonstrates the effectiveness of the proposed method.

Jiaying Lu, Jiaming Shen, Bo Xiong, Wenjing Ma, Steffen Staab, and Carl Yang. “HiPrompt: Few-Shot Biomedical Knowledge Fusion via Hierarchy-Oriented Prompting” SIGIR (2023).

Zum Seitenanfang