Tempclin Biobertpt All

🚀 Portuguese NER - TempClinBr - BioBERTpt(all)

This model is trained with BioBERTpt(all) using the TempClinBr corpus. It aims to perform named - entity recognition (NER) on Portuguese clinical texts.

✨ Features

• Widget Examples:
  • "Significant dyspnea on exertion + chest pain like a heavy weight on exertion."
  • "Obese, with hypertension, chronic heart failure. # Myocardial scintigraphy to evaluate angina. Slight edema on the left side of the mitral valve."
  • "Mitral valve plasty (Insufficiency), CRM Saf - 2Mg and Saf - 3MG (09/03/16). Denies palpitations."
  • "Usage: Aspirin (AAS) 100mg - once a day; Metoprolol 25mg - once a day; FSM - once a day; Levothyroxine 175mcg - once a day; Simvastatin 40mg - once a day; Fluoxetine 20mg - once a day."
  • "Reports improvement of dyspnea after the procedure, but still has it with mild exertion."
• Datasets: Utilizes the TempClinBr dataset for training and evaluation.

📚 Documentation

Metrics

                   precision    recall  f1-score   support

           0       0.75      0.90      0.82       291
           1       0.77      1.00      0.87        33
           2       1.00      0.25      0.40        28
           3       0.90      0.99      0.94        71
           4       0.79      0.91      0.85       112
           5       0.72      0.83      0.77       420
           6       0.62      0.45      0.53        11
           7       0.96      0.85      0.91      2236
           8       0.61      0.67      0.64        78
           9       0.61      0.98      0.76       124
          10       0.81      0.87      0.84       503
          11       0.67      0.60      0.63        10

    accuracy                           0.86      3917
   macro avg       0.77      0.78      0.74      3917
weighted avg       0.87      0.86      0.86      3917

F1:  0.8588744393393593 Accuracy:  0.8565228491192239

Parameters

device = cuda (Colab)
nclasses = len(tag2id)
nepochs = 50 => stopped at 9
batch_size = 16
batch_status = 32
learning_rate = 3e-5

early_stop = 5
max_length = 256
write_path = 'model'

Evaluation on the Test Set - TempClinBr

Note: The evaluation includes the tag "O" (label 7). If necessary, calculate the average without this tag.

tag2id ={'B-Tratamento': 0,
 'I-Teste': 1,
 'I-Ocorrencia': 2,
 'B-Evidencia': 3,
 'B-Teste': 4,
 'I-Problema': 5,
 'B-DepartamentoClinico': 6,
 'O': 7,
 'I-Tratamento': 8,
 'B-Ocorrencia': 9,
 'B-Problema': 10,
 'I-DepartamentoClinico': 11,
 '<pad>': 12}

               precision    recall  f1-score   support

           0       0.82      0.92      0.87       261
           1       0.81      0.58      0.67        99
           2       0.56      0.20      0.29        51
           3       1.00      0.94      0.97       128
           4       0.81      0.86      0.83       194
           5       0.81      0.87      0.84       645
           6       0.96      0.80      0.87        30
           7       0.95      0.90      0.93      2431
           8       0.73      0.81      0.77       146
           9       0.74      0.88      0.80       146
          10       0.87      0.95      0.91       713
          11       0.83      0.71      0.77        14
          12       0.00      0.00      0.00         0

    accuracy                           0.89      4858
   macro avg       0.76      0.72      0.73      4858
weighted avg       0.89      0.89      0.89      4858

Citation

Citation information will be available soon.