BERT-Relation-Extraction

Why we included this project

This repo is a working PyTorch implementation of the 'Matching the Blanks' relation-learning approach from ACL 2019, and it goes beyond the paper's plain BERT by also supporting ALBERT and BioBERT. The training pipeline pairs entities found in raw text, pre-trains with the matching-the-blanks objective, then fine-tunes on SemEval2010 Task 8; a small inference script lets you tag entities in a sentence and get back a predicted relation such as Cause-Effect. BioBERT support is the standout for anyone working on biomedical text, where domain-specific pretraining tends to help. It is a research codebase rather than a turnkey tool, so plan to adapt the scripts to your own data and label set, but the entity-pairing logic and fine-tuning flow are a solid foundation to build on.