Repository files navigation

This repo contains our code and pre-processed distantly/weakly labeled data for paperBOND: BERT-Assisted Open-Domain Name Entity Recognition with Distant Supervision (KDD2020)

The reuslts (entity-level F1 score) are summarized as follows:

• Full Supervision: Roberta Finetuning/BiLSTM CRF
• Previous SOTA: BiLSTM-CRF/AutoNER/LR-CRF/KALM/CONNET

We release five open-domain distantly/weakly labeled NER datasets here:dataset. For gazetteers information and distant label generation code, please directly emailcliang73@gatech.edu.

Python 3.7, Pytorch 1.3, Hugging Face Transformers v2.3.0.

We provides the training scripts for all five open-domain distantly/weakly labeled NER datasets inscripts. E.g., for BOND training and evaluation on CoNLL03

cd BOND./scripts/conll_self_training.sh

For Stage I training and evaluation on CoNLL03

cd BOND./scripts/conll_baseline.sh

The test reuslts (entity-level F1 score) are summarized as follows:

Please cite the following paper if you are using our datasets/tool. Thanks!

@inproceedings{liang2020bond,  title={BOND: Bert-Assisted Open-Domain Named Entity Recognition with Distant Supervision},  author={Liang, Chen and Yu, Yue and Jiang, Haoming and Er, Siawpeng and Wang, Ruijia and Zhao, Tuo and Zhang, Chao},  booktitle={ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},  year={2020}}