README.md

---
language: 
- en
tags:
- bert
- pytorch
- en
- ner
license: "apache-2.0"
---

# BERT for English Named Entity Recognition(bert4ner) Model
英文实体识别模型

`bert4ner-base-uncased` evaluate CoNLL-2003 test data：

The overall performance of BERT on CoNLL-2003 **test**:

|              | Accuracy  | Recall    | F1  |
| ------------ | ------------------ | ------------------ | ------------------ |
| BertSoftmax | 0.8956     | 0.9132   | 0.9043     |

在CoNLL-2003的测试集上达到接近SOTA水平。

BertSoftmax的网络结构(原生BERT)。

本项目开源在实体识别项目：[nerpy](https://github.com/shibing624/nerpy)，可支持bert4ner模型，通过如下命令调用：

#### 英文实体识别：

```shell
>>> from nerpy import NERModel
>>> model = NERModel("bert", "shibing624/bert4ner-base-uncased")
>>> predictions, raw_outputs, entities = model.predict(["AL-AIN, United Arab Emirates 1996-12-06"], split_on_space=True)
entities:  [('AL-AIN,', 'LOC'), ('United Arab Emirates', 'LOC')]
```

模型文件组成：
```
bert4ner-base-uncased
    ├── config.json
    ├── model_args.json
    ├── pytorch_model.bin
    ├── special_tokens_map.json
    ├── tokenizer_config.json
    └── vocab.txt
```

## Usage (HuggingFace Transformers)
Without [nerpy](https://github.com/shibing624/nerpy), you can use the model like this: 

First, you pass your input through the transformer model, then you have to apply the bio tag to get the entity words.

Install package:
```
pip install transformers seqeval
```

```python
import os
import torch
from transformers import AutoTokenizer, AutoModelForTokenClassification
from seqeval.metrics.sequence_labeling import get_entities

os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained("../bert4ner-base-uncased")
model = AutoModelForTokenClassification.from_pretrained("../bert4ner-base-uncased")
label_list = ["E-ORG", "E-LOC", "S-MISC", "I-MISC", "S-PER", "E-PER", "B-MISC", "O", "S-LOC",
              "E-MISC", "B-ORG", "S-ORG", "I-ORG", "B-LOC", "I-LOC", "B-PER", "I-PER"]

sentence = "AL-AIN, United Arab Emirates 1996-12-06"


def get_entity(sentence):
    tokens = tokenizer.tokenize(sentence)
    inputs = tokenizer.encode(sentence, return_tensors="pt")
    with torch.no_grad():
        outputs = model(inputs).logits
    predictions = torch.argmax(outputs, dim=2)
    word_tags = [(token, label_list[prediction]) for token, prediction in zip(tokens, predictions[0].numpy()[1:-1])]
    print(sentence)
    print(word_tags)

    pred_labels = [i[1] for i in word_tags]
    entities = []
    line_entities = get_entities(pred_labels)
    for i in line_entities:
        word = tokens[i[1]: i[2] + 1]
        entity_type = i[0]
        entities.append((word, entity_type))

    print("Sentence entity:")
    print(entities)


get_entity(sentence)
```


### 数据集

#### 实体识别数据集


| 数据集 | 语料 | 下载链接 | 文件大小 |
| :------- | :--------- | :---------: | :---------: |
| **`CNER中文实体识别数据集`** | CNER(12万字) | [CNER github](https://github.com/shibing624/nerpy/tree/main/examples/data/cner)| 1.1MB |
| **`PEOPLE中文实体识别数据集`** | 人民日报数据集（200万字） | [PEOPLE github](https://github.com/shibing624/nerpy/tree/main/examples/data/people)| 12.8MB |
| **`CoNLL03英文实体识别数据集`** | CoNLL-2003数据集（22万字） | [CoNLL03 github](https://github.com/shibing624/nerpy/tree/main/examples/data/conll03)| 1.7MB |


### input format

Input format (prefer BIOES tag scheme), with each character its label for one line. Sentences are splited with a null line.

```text
EU	S-ORG
rejects	O
German	S-MISC
call	O
to	O
boycott	O
British	S-MISC
lamb	O
.	O

Peter	B-PER
Blackburn	E-PER
```


如果需要训练bert4ner，请参考[https://github.com/shibing624/nerpy/tree/main/examples](https://github.com/shibing624/nerpy/tree/main/examples)


## Citation

```latex
@software{nerpy,
  author = {Xu Ming},
  title = {nerpy: Named Entity Recognition toolkit},
  year = {2022},
  url = {https://github.com/shibing624/nerpy},
}
```