pierreguillou's picture
Update README.md
17c0eab
|
raw
history blame
5.32 kB
metadata
language:
  - pt
tags:
  - generated_from_trainer
datasets:
  - lener_br
metrics:
  - precision
  - recall
  - f1
  - accuracy
model-index:
  - name: checkpoints
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: lener_br
          type: lener_br
        metrics:
          - name: F1
            type: f1
            value: 0.8716487228203504
          - name: Precision
            type: precision
            value: 0.8559286898839138
          - name: Recall
            type: recall
            value: 0.8879569892473118
          - name: Accuracy
            type: accuracy
            value: 0.9755893153732458
          - name: Loss
            type: loss
            value: 0.1133928969502449
widget:
  - text: >-
      Acrescento que não há de se falar em violação do artigo 114, § 3º, da
      Constituição Federal, posto que referido dispositivo revela-se
      impertinente, tratando da possibilidade de ajuizamento de dissídio
      coletivo pelo Ministério Público do Trabalho nos casos de greve em
      atividade essencial.

(BERT base) NER model in the legal domain in Portuguese (LeNER-Br)

ner-bert-base-portuguese-cased-lenerbr is a NER model (token classification) in the legal domain in Portuguese that was finetuned on 16/12/2021 in Google Colab from the model BERTimbau base on the dataset LeNER_br by using a NER objective.

Note: due to the small size of BERTimbau base and finetuning dataset, the model overfitted before to reach the end of training. Here are the overall final metrics on the validation dataset (note: see the paragraph "Validation metrics by Named Entity" to get detailed metrics):

  • f1: 0.8716487228203504
  • precision: 0.8559286898839138
  • recall: 0.8879569892473118
  • accuracy: 0.9755893153732458
  • loss: 0.1133928969502449

Widget & APP

You can test this model into the widget of this page.

Using the model for inference in production

# install pytorch: check https://pytorch.org/
# !pip install transformers 
from transformers import AutoModelForTokenClassification, AutoTokenizer
import torch

# parameters
model_name = "ner-bert-base-portuguese-cased-lenebr"
model = AutoModelForTokenClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

input_text = "EMENTA: APELAÇÃO CÍVEL - AÇÃO DE INDENIZAÇÃO POR DANOS MORAIS - PRELIMINAR - ARGUIDA PELO MINISTÉRIO PÚBLICO EM GRAU RECURSAL - NULIDADE - AUSÊNCIA DE IN- TERVENÇÃO DO PARQUET NA INSTÂNCIA A QUO - PRESENÇA DE INCAPAZ - PREJUÍZO EXISTENTE - PRELIMINAR ACOLHIDA - NULIDADE RECONHECIDA."

# tokenization
inputs = tokenizer(input_text, max_length=512, truncation=True, return_tensors="pt")
tokens = inputs.tokens()

# get predictions
outputs = model(**inputs).logits
predictions = torch.argmax(outputs, dim=2)

# print predictions
for token, prediction in zip(tokens, predictions[0].numpy()):
    print((token, model.config.id2label[prediction]))

You can use pipeline, too. However, it seems to have an issue regarding to the max_length of the input sequence.

!pip install transformers
import transformers
from transformers import pipeline

model_name = "ner-bert-base-portuguese-cased-lenebr"

ner = pipeline(
    "ner",
    model=model_name
) 

ner(input_text)

Training procedure

Training results

Num examples = 7828
Num Epochs = 3
Instantaneous batch size per device = 8
Total train batch size (w. parallel, distributed & accumulation) = 8
Gradient Accumulation steps = 1
Total optimization steps = 2937

Step	Training Loss	Validation Loss	Precision	Recall    	F1      	Accuracy
290 	0.315100     	0.141881       	0.764542 	0.709462  	0.735973	0.960550
580	 0.089100	     0.137700       	0.729155 	0.810538  	0.767695	0.959940
870	 0.071700	     0.122069       	0.780277 	0.872903  	0.823995	0.967955
1160	0.047500     	0.125950       	0.800312 	0.881720  	0.839046	0.968367
1450	0.034900     	0.129228       	0.763666 	0.910323  	0.830570	0.969068
1740	0.036100     	0.113393       	0.855929 	0.887957  	0.871649	0.975589
2030	0.037800     	0.121275       	0.817230 	0.889462  	0.851818	0.970393
2320	0.018700     	0.115745       	0.836066 	0.877419  	0.856243	0.973136
2610	0.017100     	0.118826       	0.822488 	0.888817  	0.854367	0.973471

Validation metrics by Named Entity

Num examples = 1177

{'JURISPRUDENCIA': {'f1': 0.6641509433962263,
  'number': 657,
  'precision': 0.6586826347305389,
  'recall': 0.669710806697108},
 'LEGISLACAO': {'f1': 0.8489082969432314,
  'number': 571,
  'precision': 0.8466898954703833,
  'recall': 0.851138353765324},
 'LOCAL': {'f1': 0.8066037735849058,
  'number': 194,
  'precision': 0.7434782608695653,
  'recall': 0.8814432989690721},
 'ORGANIZACAO': {'f1': 0.8540462427745664,
  'number': 1340,
  'precision': 0.8277310924369747,
  'recall': 0.8820895522388059},
 'PESSOA': {'f1': 0.9845722300140253,
  'number': 1072,
  'precision': 0.9868791002811621,
  'recall': 0.9822761194029851},
 'TEMPO': {'f1': 0.9527794381350867,
  'number': 816,
  'precision': 0.9299883313885647,
  'recall': 0.9767156862745098},
 'overall_accuracy': 0.9755893153732458,
 'overall_f1': 0.8716487228203504,
 'overall_precision': 0.8559286898839138,
 'overall_recall': 0.8879569892473118}