模型:

flair/ner-multi

英文

4-语言NER模型(英语、德语、荷兰语和西班牙语)

这是一个标准的4类NER模型,适用于四种CoNLL-03语言, Flair 进行了一些修改,并且也能应用于类似法语的相关语言。

F1-Score: 92.16(CoNLL-03英文),87.33(修订版CoNLL-03德语),88.96(CoNLL-03荷兰语),86.65(CoNLL-03西班牙语)

预测4个标签:

tag meaning
PER person name
LOC location name
ORG organization name
MISC other name

基于 Flair embeddings 和LSTM-CRF。

在Flair中使用的演示

需要: Flair (pip install flair)

from flair.data import Sentence
from flair.models import SequenceTagger

# load tagger
tagger = SequenceTagger.load("flair/ner-multi")

# make example sentence in any of the four languages
sentence = Sentence("George Washington ging nach Washington")

# predict NER tags
tagger.predict(sentence)

# print sentence
print(sentence)

# print predicted NER spans
print('The following NER tags are found:')
# iterate over entities and print
for entity in sentence.get_spans('ner'):
    print(entity)

这将产生以下输出:

Span [1,2]: "George Washington"   [− Labels: PER (0.9977)]
Span [5]: "Washington"   [− Labels: LOC (0.9895)]

因此,在句子“George Washington ging nach Washington”中找到了实体“George Washington”(标记为人物)和“Washington”(标记为位置)。

训练:用于训练该模型的脚本

使用以下Flair脚本进行了模型训练:

from flair.data import Corpus
from flair.datasets import CONLL_03, CONLL_03_GERMAN, CONLL_03_DUTCH, CONLL_03_SPANISH
from flair.embeddings import WordEmbeddings, StackedEmbeddings, FlairEmbeddings

# 1. get the multi-language corpus
corpus: Corpus = MultiCorpus([
    CONLL_03(),         # English corpus
    CONLL_03_GERMAN(),  # German corpus
    CONLL_03_DUTCH(),   # Dutch corpus
    CONLL_03_SPANISH(), # Spanish corpus
    ])

# 2. what tag do we want to predict?
tag_type = 'ner'

# 3. make the tag dictionary from the corpus
tag_dictionary = corpus.make_tag_dictionary(tag_type=tag_type)

# 4. initialize each embedding we use
embedding_types = [

    # GloVe embeddings
    WordEmbeddings('glove'),

    # FastText embeddings
    WordEmbeddings('de'),

    # contextual string embeddings, forward
    FlairEmbeddings('multi-forward'),

    # contextual string embeddings, backward
    FlairEmbeddings('multi-backward'),
]

# embedding stack consists of Flair and GloVe embeddings
embeddings = StackedEmbeddings(embeddings=embedding_types)

# 5. initialize sequence tagger
from flair.models import SequenceTagger

tagger = SequenceTagger(hidden_size=256,
                        embeddings=embeddings,
                        tag_dictionary=tag_dictionary,
                        tag_type=tag_type)

# 6. initialize trainer
from flair.trainers import ModelTrainer

trainer = ModelTrainer(tagger, corpus)

# 7. run training
trainer.train('resources/taggers/ner-multi',
              train_with_dev=True,
              max_epochs=150)

引用

使用该模型时,请引用以下论文。

@misc{akbik2019multilingual,
  title={Multilingual sequence labeling with one model},
  author={Akbik, Alan and Bergmann, Tanja and Vollgraf, Roland}
  booktitle = {{NLDL} 2019, Northern Lights Deep Learning Workshop},
  year      = {2019}
}
@inproceedings{akbik2018coling,
  title={Contextual String Embeddings for Sequence Labeling},
  author={Akbik, Alan and Blythe, Duncan and Vollgraf, Roland},
  booktitle = {{COLING} 2018, 27th International Conference on Computational Linguistics},
  pages     = {1638--1649},
  year      = {2018}
}