BioRED task NCU-IISR submission: Preprocessing-Robust Ensemble Learning Approach for Biomedical Relation Extraction

Abstract

We propose an end-to-end system for the BioCreative VIII Challenge Track 1: BioRED Track, focusing on biomedical relation extraction. In our study, we employ an ensemble learning approach, combining the PubTator API with multiple pre-trained Bidirectional Encoder Representations from Transformers (BERT) models. A variety of preprocessing inputs are utilized, including prompt questions, entity ID pairs, and co-occurrence contexts. Special tokens and boundary tags are added to enhance model understanding. In this study, PubMedBERT and the Max Rule ensemble learning mechanism are used to combine outputs from different classifiers. In subtask 1, the method achieves a F1 score of 43%, and in subtask 2, it achieves a score of 23%, demonstrating significant advancements in biomedical relation extraction.

This article is part of the Proceedings of the BioCreative VIII Challenge and Workshop: Curation and Evaluation in the era of Generative Models.