QomSSLab/Anonymizer-v3

This repository hosts an XLM-RoBERTa token-classification head trained.

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

model_id = "QomSSLab/Anonymizer-v3"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForTokenClassification.from_pretrained(model_id)
tagger = pipeline("token-classification", model=model, tokenizer=tokenizer, aggregation_strategy="simple")

text = "مثال از یک ورودی فارسی"
for entity in tagger(text):
    print(entity)

Labels

• ACOUNT
• ADDRESS
• AMOUNT
• DATE
• DOCUMENT_ID
• ID
• JOB
• O
• ORG
• ORG_BRANCH
• PERSON
• PHONEـNUMBER
• PLATEـNUMBER

Metrics

Validation Metrics

• Precision: 0.9336
• Recall: 0.9508
• F1: 0.9421
• Accuracy: 0.9923

Per-label Breakdown

Label	Precision	Recall	F1	Support
ACOUNT	1.0000	0.4750	0.6441	40
ADDRESS	0.9987	0.9993	0.9990	1519
AMOUNT	0.9941	1.0000	0.9970	168
DATE	0.9836	0.9938	0.9887	484
DOCUMENT_ID	0.9660	0.9906	0.9781	1378
ID	0.9427	1.0000	0.9705	181
JOB	1.0000	0.4286	0.6000	35
O	0.9960	0.9982	0.9971	16164
ORG	0.7961	0.9213	0.8542	89
ORG_BRANCH	0.9911	0.9321	0.9607	839
PERSON	0.9924	0.9931	0.9928	1454
PHONEـNUMBER	1.0000	0.9417	0.9700	206
PLATEـNUMBER	1.0000	1.0000	1.0000	0