NLL-IE

Code for EMNLP 2021 paper Learning from Noisy Labels for Entity-Centric Information Extraction.

Requirements

• PyTorch >= 1.8.1
• Transformers >= 3.4.0
• wandb
• ujson
• tqdm
• truecase
• seqeval

Dataset

The TACRED dataset can be obtained from this link. The TACREV dataset can be obtained following the instructions in tacrev. The original CoNLL dataset can be obtained from this link. The revised CoNLL test dataset can be obtained from this link. The expected structure of files is:

NLL-IE
 |-- re
 |    |-- data
 |    |    |-- train.json        
 |    |    |-- dev.json
 |    |    |-- test.json
 |    |    |-- dev_rev.json
 |    |    |-- test_rev.json
 |-- ner
 |    |-- data
 |    |    |-- train.txt     
 |    |    |-- dev.txt
 |    |    |-- test.txt
 |    |    |-- conllpp_test.txt

Training and Evaluation

Train the RE/NER model on with the following command:

>> python train.py

The training loss and evaluation results on the dev set are synced to the wandb dashboard.