eznlp 0.2.1rc1

Easy Natural Language Processing

Easy Natural Language Processing

eznlp is a PyTorch-based package for neural natural language processing, currently supporting:

• Text classification
• Named Entity Recognition
  • Sequence tagging
  • Span classification
  • Boundary Selection
• Relation extraction
• Attribute extraction

Experiment Results

Text Classification

Dataset	Language	Our Best Imp. Acc.	Model Specification
IMDb	English	95.78	RoBERTa-base + Attention
Yelp Full	English	71.55	RoBERTa-base + Attention
Yelp 2013	English	70.80	RoBERTa-base + Attention
ChnSentiCorp	Chinese	95.83	BERT-base + Attention
THUCNews-10	Chinese	98.98	RoBERTa-base + Attention

See Text Classification for more details.

Named Entity Recognition

Dataset	Language	Our Best Imp. F1	Model Specification
CoNLL 2003	English	93.26	RoBERTa-large + LSTM + CRF
OntoNotes v5	English	91.05	RoBERTa-base + LSTM + CRF
MSRA	Chinese	96.18	BERT + LSTM + CRF
WeiboNER v2	Chinese	70.48	BERT + LSTM + CRF
ResumeNER	Chinese	95.97	BERT + LSTM + CRF
OntoNotes v5	Chinese	80.31	BERT + LSTM + CRF

See Named Entity Recognition for more details.

Relation Extraction

Dataset	Language	Our Best Imp. F1	Model Specification
CoNLL 2004	English	89.17 / 75.03	SpERT (with RoBERTa-base + LSTM)
SciERC	English	69.29 / 36.65	SpERT (with RoBERTa-base)

See Relation Extraction for more details.

Installation

With pip

$ pip install eznlp

From source

$ python setup.py sdist
$ pip install dist/eznlp-<version>.tar.gz

Running the Code

Text classification

$ python scripts/text_classification.py --dataset <dataset> [options]

Entity recognition

$ python scripts/entity_recognition.py --dataset <dataset> [options]

Relation extraction

$ python scripts/relation_extraction.py --dataset <dataset> [options]

Attribute extraction

$ python scripts/attribute_extraction.py --dataset <dataset> [options]

Citation

If you find our code useful, please cite the following paper:

@article{zhu2021framework,
  title={A Unified Framework of Medical Information Annotation and Extraction for {Chinese} Clinical Text},
  author={Zhu, Enwei and Sheng, Qilin and Yang, Huanwan and Li, Jinpeng},
  journal={Working Paper},
  year={2021}
}

Future Plans

• SoftLexicon
• Radical-Level Features
• Experiments on Chinese NER datasets
• Experiments on text classification datasets
• Focal loss (and combined to CRF?)
• Dice loss
• Relation Extraction
• Span-based models (e.g., SpERT)
• NER / RE as MRC
• Pair selection (multi-head selection; RE for flat entities)
• Data Augmentation
• LR finder for MSRA