GRAFT: Gated Retrieval-Augmented Fine-Tuning for Relation Extraction

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

GRAFT (Gated Retrieval-Augmented Fine-Tuning) is a novel approach designed to improve biomedical relation extraction (RE) by addressing the challenge of missing definitional or auxiliary entity information in input texts. This method utilizes a pre-trained biomedical text retriever to augment original inputs with instance-specific textual snippets. A crucial gating mechanism ensures these retrieved snippets enhance the signal without overwhelming the original input, mitigating potential false positives or negatives. Evaluated on three standard biomedical RE datasets—CDR, BioRED, and ChemProt—GRAFT consistently demonstrated significant performance improvements, achieving up to 10 F1 points higher scores compared to robust supervised baselines, including both encoder and decoder models. The code and datasets are publicly available.

Key takeaway

For NLP Engineers developing biomedical knowledge graphs, GRAFT offers a robust solution to improve relation extraction accuracy. If you are struggling with missing contextual information, consider implementing this gated retrieval-augmented fine-tuning approach. It can significantly boost F1 scores by up to 10 points on datasets like CDR and BioRED, ensuring external data enhances rather than degrades your model's performance. You should explore the public code to integrate this method.

Key insights

Gated retrieval augmentation enhances biomedical relation extraction by selectively integrating external context without signal overwhelm.

Principles

• External context can improve RE.
• Gating prevents signal overwhelm.
• Retrieval augmentation boosts LLMs.

Method

GRAFT uses a pre-trained biomedical text retriever to augment original inputs with instance-specific snippets. A gating mechanism controls snippet integration to enhance, not overwhelm, the original signal during fine-tuning.

In practice

• Apply GRAFT to biomedical RE.
• Use gated retrieval for context.
• Evaluate on CDR, BioRED, ChemProt.

Topics

• Relation Extraction
• Biomedical NLP
• Retrieval Augmentation
• Large Language Models
• Knowledge Graphs
• Fine-tuning

Code references

• bionlproc/GRAFT-RE

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning? — LLMs offer a viable few-shot alternative for biomedical relation extraction, outperforming prior few-shot methods, especially for rare relations.
• GRAFT: Auditing Graph Neural Networks via Global Feature Attribution — GRAFT provides global, class-level feature importance explanations for GNNs, translating them into natural language rules.
• Benchmarking Retrieval Strategies for Biomedical Retrieval-Augmented Generation: A Controlled Empirical Study — Cross-Encoder Reranking significantly improves biomedical RAG performance by enhancing contextual precision.
• FineREX: Fine-Tuned NER-RE for Human Smuggling Knowledge Graphs — Domain-specific fine-tuning of LLMs significantly enhances knowledge graph construction quality and efficiency for complex, jargon-heavy legal texts.
• Reinforcement Learning Improves LLM Accuracy and Reasoning in Disease Classification from Radiology Reports — Reinforcement learning with GRPO enhances LLM accuracy and reasoning in medical text classification without explicit reasoning supervision.
• Extracting Problem and Method Sentence from Scientific Papers: A Context-enhanced Transformer Using Formulaic Expression Desensitization — Formulaic expression desensitization and context-enhanced transformers improve problem/method sentence extraction from scientific papers, outperforming baselines.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.