China-ChallengeHub/bert4ner.pytorch

BERT models for Named Entity Recognition

Python

32 commits

updated Jun 1, 2021

See the code

README

bert4ner.pytorch

A PyTorch implementation of Bert models for Named Entity Recognition(NER).

Quick start

The training arguments can be found in args/. To train CRF models, the command should be like

$ python run_ner_crf.py args/bert_crf-clue_ner.json

Requirements

RTX3090, Driver Version: 455.38, CUDA Version: 11.1

apex==0.1
numpy==1.19.4
pandas==1.1.4
scipy==1.5.4
seqeval==1.2.2
torch==1.7.0
torchtyping==0.1.1
tqdm==4.49.0
transformers==4.5.0
allennlp==2.4.0

Datasets

Datasets used in this repository can be found in OYE93/Chinese-NLP-Corpus - Github and CLUEbenchmark.

Models

Pretrained language model(PLM) including

BERT: e.g. hfl/chinese-roberta-wwm-ext
NEZHA: e.g. nghuyong/ernie-1.0
FLAT

Named Entity Recognition Prediction Header including

CRF: Bidirectional LSTM-CRF Models for Sequence Tagging
SPAN
Biaffine: Named Entity Recognition as Dependency Parsing

Benchmarks

Models

The performance on dev set of CLUENER2000(F1 score).

entitiy	Human Performance	Bert-crf	Bert-span	Bert-biaffine	Bert-span-v2
Address	43.04	64.30	65.76	65.47	69.51
Book	71.70	82.58	83.61	81.23	82.84
Company	49.32	80.82	83.11	81.96	85.36
Game	80.39	84.93	85.29	84.06	85.66
Government	79.27	82.92	83.07	81.25	86.02
Movie	63.21	83.06	83.11	83.62	87.26
Person Name	74.49	88.87	88.87	87.38	89.58
Organization	65.41	80.27	82.48	80.56	84.58
Position	55.38	79.51	80.69	80.76	80.24
Scene	51.85	76.56	72.37	74.07	75.25
Overall@Macro	63.41	80.38	80.84	79.89	82.63
Overall@Micro	/	80.26	81.00	80.08	82.72

entitiy	Bert-span-v2	Ernie-span-v2
Address	69.51	67.16
Book	82.84	81.82
Company	85.36	84.11
Game	85.66	87.39
Government	86.02	84.48
Movie	87.26	86.92
Person Name	89.58	89.37
Organization	84.58	83.67
Position	80.24	81.26
Scene	75.25	73.60
Overall@Macro	82.63	81.98
Overall@Micro	82.72	82.19

Datasets

Reference

Similar repos

lonePatient/BERT-NER-Pytorch

Chinese NER(Named Entity Recognition) using BERT(Softmax, CRF, Span)

2,236

CLUEbenchmark/CLUENER2020

CLUENER2020 中文细粒度命名实体识别 Fine Grained Named Entity Recognition

1,523

FantasticCode2019/NER_center

收集了一些NER的sota模型

dslim/bert-base-NER

Model

bert-base-NER

743

kaiyinzhou/BERT-NER

Use Google's BERT for named entity recognition （CoNLL-2003 as the dataset）.

1,275

kamalkraj/BERT-NER

Pytorch-Named-Entity-Recognition-with-BERT

1,250

taishan1994/awesome-chinese-ner

中文命名实体识别。包含目前最新的中文命名实体识别论文、中文实体识别相关工具、数据集，以及中文预训练模型、词向量、实体识别综述等。

769

owenliang/ner-bert-crf

NER Model

China-ChallengeHub/bert4ner.pytorch

BERT models for Named Entity Recognition

Python

32 commits

updated Jun 1, 2021

See the code

README

bert4ner.pytorch

A PyTorch implementation of Bert models for Named Entity Recognition(NER).

Quick start

The training arguments can be found in args/. To train CRF models, the command should be like

$ python run_ner_crf.py args/bert_crf-clue_ner.json

Requirements

RTX3090, Driver Version: 455.38, CUDA Version: 11.1

apex==0.1
numpy==1.19.4
pandas==1.1.4
scipy==1.5.4
seqeval==1.2.2
torch==1.7.0
torchtyping==0.1.1
tqdm==4.49.0
transformers==4.5.0
allennlp==2.4.0

Datasets

Datasets used in this repository can be found in OYE93/Chinese-NLP-Corpus - Github and CLUEbenchmark.

Models

Pretrained language model(PLM) including

BERT: e.g. hfl/chinese-roberta-wwm-ext
NEZHA: e.g. nghuyong/ernie-1.0
FLAT

Named Entity Recognition Prediction Header including

CRF: Bidirectional LSTM-CRF Models for Sequence Tagging
SPAN
Biaffine: Named Entity Recognition as Dependency Parsing

Benchmarks

Models

The performance on dev set of CLUENER2000(F1 score).

entitiy	Human Performance	Bert-crf	Bert-span	Bert-biaffine	Bert-span-v2
Address	43.04	64.30	65.76	65.47	69.51
Book	71.70	82.58	83.61	81.23	82.84
Company	49.32	80.82	83.11	81.96	85.36
Game	80.39	84.93	85.29	84.06	85.66
Government	79.27	82.92	83.07	81.25	86.02
Movie	63.21	83.06	83.11	83.62	87.26
Person Name	74.49	88.87	88.87	87.38	89.58
Organization	65.41	80.27	82.48	80.56	84.58
Position	55.38	79.51	80.69	80.76	80.24
Scene	51.85	76.56	72.37	74.07	75.25
Overall@Macro	63.41	80.38	80.84	79.89	82.63
Overall@Micro	/	80.26	81.00	80.08	82.72

entitiy	Bert-span-v2	Ernie-span-v2
Address	69.51	67.16
Book	82.84	81.82
Company	85.36	84.11
Game	85.66	87.39
Government	86.02	84.48
Movie	87.26	86.92
Person Name	89.58	89.37
Organization	84.58	83.67
Position	80.24	81.26
Scene	75.25	73.60
Overall@Macro	82.63	81.98
Overall@Micro	82.72	82.19

Datasets

Reference

Similar repos

lonePatient/BERT-NER-Pytorch

Chinese NER(Named Entity Recognition) using BERT(Softmax, CRF, Span)

2,236

CLUEbenchmark/CLUENER2020

CLUENER2020 中文细粒度命名实体识别 Fine Grained Named Entity Recognition

1,523

FantasticCode2019/NER_center

收集了一些NER的sota模型

dslim/bert-base-NER

Model

bert-base-NER

743

kaiyinzhou/BERT-NER

Use Google's BERT for named entity recognition （CoNLL-2003 as the dataset）.

1,275

kamalkraj/BERT-NER

Pytorch-Named-Entity-Recognition-with-BERT

1,250

taishan1994/awesome-chinese-ner

中文命名实体识别。包含目前最新的中文命名实体识别论文、中文实体识别相关工具、数据集，以及中文预训练模型、词向量、实体识别综述等。

769

owenliang/ner-bert-crf

NER Model