README.md

metadata

license: apache-2.0
language:
  - ko
base_model:
  - beomi/kcbert-base
pipeline_tag: token-classification
tags:
  - Korean
  - PII
  - KoreanPII
  - PIIMasking
  - Anonymization
  - Privacy

Korean-PII-Masking-BERT

GitHub Repository: alphagyuu/Korean-PII-Masking-BERT

Korean-PII-Masking-BERT is a token classification model fine-tuned on KcBERT’s TokenClassifier using a processed version of "Korean SNS" dataset from AI-Hub.

🖥️ Python Implementation

• Tokenizer:

  BertTokenizer.from_pretrained('beomi/kcbert-base', do_lower_case=False)
  

• Model:

  TFBertForTokenClassification.from_pretrained('alphagyuu/Korean-PII-Masking-BertForTokenClassification', num_labels=len(tag2idx))
  

• LabelMap:

  LabelMAP = {
    'O': 'LABEL0',
    'B-URL': 'LABEL1',
    'I-URL': 'LABEL2',
    'B-계정': 'LABEL3',
    'I-계정': 'LABEL4',
    'B-금융': 'LABEL5',
    'I-금융': 'LABEL6',
    'B-번호': 'LABEL7',
    'I-번호': 'LABEL8',
    'B-소속': 'LABEL9',
    'I-소속': 'LABEL10',
    'B-신원': 'LABEL11',
    'I-신원': 'LABEL12',
    'B-이름': 'LABEL13',
    'I-이름': 'LABEL14',
    'B-주소': 'LABEL15',
    'I-주소': 'LABEL16'
  }