Ambareeshkumar/BERT-Tamil

Tamil Named Entity Recognition

Fine-tuning bert-base-multilingual-cased on Wikiann dataset for performing NER on Tamil language.

Label ID and its corresponding label name

Label ID	Label Name
0	O
1	B-PER
2	I-PER
3	B-ORG
4	I-ORG
5	B-LOC
6	I-LOC

Results

Step	Training Loss	Validation Loss	Overall Precision	Overall Recall	Overall F1	Overall Accuracy	Loc F1	Org F1	Per F1
1000	0.386900	0.300006	0.833469	0.824748	0.829086	0.912857	0.835343	0.781625	0.867752
2000	0.210200	0.251389	0.845455	0.842052	0.843750	0.924861	0.851711	0.790198	0.886515
3000	0.140000	0.264964	0.866952	0.856137	0.861510	0.930141	0.874872	0.818150	0.885203
4000	0.095400	0.298542	0.860871	0.882696	0.871647	0.935692	0.881348	0.829285	0.899245
5000	0.062200	0.296011	0.871805	0.878471	0.875125	0.938806	0.875434	0.850889	0.898148
6000	0.042200	0.320418	0.868416	0.879074	0.873713	0.937497	0.877588	0.833611	0.907737

Example

from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("Ambareeshkumar/BERT-Tamil")
model = AutoModelForTokenClassification.from_pretrained("Ambareeshkumar/BERT-Tamil")
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
example = "இந்திய"
ner_results = nlp(example)
ner_results