An Implementation of Deep Exhaustive Model for Nested NER

Original paper: Sohrab, M. G., & Miwa, M. (2018). Deep Exhaustive Model for Nested Named Entity Recognition. In 2018 EMNLP

Requirements

• python 3.6.7
• pytorch 1.0.0
• numpy 1.15.3
• gensim 3.6.0
• scikit-learn 0.20.0
• joblib 0.12.5

Data Format

Our processed GENIA dataset is in ./data/.

The data format is the same as in Neural Layered Model, Ju et al. 2018 NAACL

Each line has multiple columns separated by a tab key. Each line contains

word	label1	label2	label3	...	labelN

The number of labels (N) for each word is determined by the maximum nested level in the data set. N=maximum nested level + 1 Each sentence is separated by an empty line. For example, for these two sentences, John killed Mary's husband. He was arrested last night , they contain four entities: John (PER), Mary(PER), Mary's husband(PER),He (PER). The format for these two sentences is listed as following:

John    B-PER   O   O
killed  O   O   O
Mary    B-PER   B-PER   O
's  O   I-PER   O
husband O   I-PER   O
.   O   O   O

He    B-PER   O   O
was  O   O   O
arrested  O   O   O
last  O   O   O
night  O   O   O
.  O   O   O

Pre-trained word embeddings

• Pre-trained word embeddings used here is the same as in Neural Layered Model

Setup

Download pre-trained embedding above, unzip it, and place PubMed-shuffle-win-30.bin into ./data/embedding/

Usage

Training

python3 train.py

trained best model will be saved at ./data/model/

Testing

set model_url to the url of saved model in training in main() of eval.py

python3 eval.py