BeLink: Biomedical Entity Linking Meets Generative Re-Ranking

2026-05-21

AI Summary

This work addresses the challenges of computational inefficiency and deployment difficulty in biomedical entity linking (BEL) when leveraging large language models. It introduces, for the first time, an instruction-tuned open-source generative model into the BEL re-ranking stage and proposes an ensemble instruction-tuning strategy to enable efficient and accurate candidate entity selection. The approach is integrated into BeLink, a modular end-to-end BEL system. Evaluated across multiple BEL benchmarks, the method improves linking accuracy by 3%–24% over state-of-the-art techniques while substantially reducing inference time, thereby achieving a superior balance between precision and efficiency and enhancing the practical feasibility of BEL in real-world applications.

Abstract

Despite recent progress, Biomedical Entity Linking (BEL) with large language models (LLMs) remains computationally inefficient and challenging to deploy in practical settings. In this work, we demonstrate that instruction-tuning of open-source generative models can offer an effective solution when applied at the re-ranking stage of the BEL pipeline. We propose a set-wise instruction-tuning formulation that enables fast and accurate candidate selection. Our method demonstrates strong performance on multiple BEL benchmarks, yielding significant improvements in linking accuracy (3%-24%) while reducing inference time compared to the state-of-the-art. We integrate our generative re-ranker into BeLink, a modular, end-to-end system designed for practical real-world BEL applications.

Problem

• Biomedical Entity Linking
• large language models
• computational efficiency
• practical deployment

Innovation

• Biomedical Entity Linking
• Generative Re-Ranking
• Instruction Tuning
• Open-Source LLMs
• Candidate Re-Ranking