Fine tuning NER BERT model on Phone numbers

Beginners
1

Hello, I am new to this community. I have bit confusion regarding Fine tuning NER BERT model on Phone Numbers. According to my research fine tuning on BERT NER model. I need to create custom data with a inputs like Tag/Masks and POS tags, words and sentence_idx( i.e is word related to same sentence). Sent it to a model. then as i understood pre trained model create word_idx from Dictionary makes the model understand (word_idx belonging Tagged data(PHONE_NUMBER)) and simultaneously understand text contextual. Next time any data releated to same word_idx as well as context is tagged as Phone_number. What I have understand from BERT Model. So, I am preparing tags based on model tokenization like [‘drop’, ‘us’, ‘a’, ‘line’, ‘at’, ‘the’, ‘following’, ‘+’, ‘46’, ‘(’, ‘0’, ‘)’, ‘8’, ‘505’, ‘139’, ‘02’, ‘.’] and my tag should be some what [‘o’, ‘o’, ‘o’, ‘o’, ‘o’, ‘o’, ‘o’, ‘+’, ‘PHONE_B’, ‘(’, ‘PHONE_I’, ‘)’, ‘PHONE_I’, ‘PHONE_I’, ‘PHONE_I’, ‘PHONE_I’, ‘.’] and i will prepare POS tags based SPACY and FLAIR this is my pre processing. Then fine tuning. Is it correct procedure/understanding? I would be glad to have references and suggestion related to this.

2

Hi, trying the same. Were you able to achieve it?

3

Hi, what’s the situation?

4

For named-entity recognition (NER) you only need a list of words and corresponding labels (no need for POS tags).

See my demo notebook to fine-tune BERT for NER.