Reproduce “Enhancing Biomedical Relation Extraction with Directionality” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Po-Ting Lai, Chih-Hsuan Wei, Shubo Tian, Robert Leaman, Zhiyong Lu
Researchers at the National Library of Medicine enriched the BioRED corpus with 10,864 directionality annotations and developed a multi-task language model for joint extraction of relation type, novelty, and directionality. This model achieved an F1-score of 75.34% for Entity Pair and 48.62% for Relation Type + Directionality on BioRED, and 71.4% on BC5CDR, surpassing previous state-of-the-art methods and large language models.
Key insights
- Providing explicit directionality annotations is essential for building models that can create actionable and mechanistically accurate biological knowledge graphs.
- A multi-task learning framework, jointly predicting relation type, novelty, and directionality, allows for shared learning signals that improve overall performance.
- Combining context chunking strategies with soft-prompt learning effectively addresses the challenges of long-document processing and enhances the performance of PLM-based relation extraction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.