distilbert-moroccan-pii-classifier

🇲🇦 DistilBERT Moroccan PII & CV Entity Extraction

Model Details

Model Description

This model is a fine-tuned version of DistilBERT for Named Entity Recognition (NER) focused on PII detection and CV parsing in the Moroccan context.

It is designed to extract structured information from multilingual text including Darija (Moroccan Arabic), Arabic, and French, commonly found in resumes and semi-structured documents.

• Developed by: Youssef Lamaachi
• Model type: Token Classification (NER)
• Language(s): Arabic, French, English (Darija included)
• License: Apache 2.0
• Finetuned from model: distilbert-base-uncased

Uses

Direct Use

This model can be directly used for:

• Extracting PII from text
• Parsing CVs and resumes
• Structuring candidate information
• Data anonymization pipelines

---

Downstream Use

• HRTech platforms
• Applicant Tracking Systems (ATS)
• Document processing pipelines
• AI-powered recruitment tools

---

Out-of-Scope Use

• Surveillance or mass tracking systems
• Legal or medical critical decision-making
• Highly noisy OCR without preprocessing

---

Bias, Risks, and Limitations

• Model is trained on Moroccan CV-style data → may not generalize globally
• May struggle with:
  • Informal Darija spelling variations
  • OCR errors or noisy inputs
• Potential bias depending on dataset annotation quality

---

Recommendations

• Use preprocessing (clean text / OCR correction)
• Fine-tune further for other domains if needed
• Avoid sensitive or unethical use cases

---

How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

tokenizer = AutoTokenizer.from_pretrained("lamaachi/distilbert-moroccan-pii-classifier")
model = AutoModelForTokenClassification.from_pretrained("your-username/distilbert-moroccan-pii-classifier")

nlp = pipeline("ner", model=model, tokenizer=tokenizer)

text = "Smiya dyali Youssef, num 0612345678, email: [email protected]"
result = nlp(text)

print(result)