Fine-tuned BERT for Persian Comment Discrepancy Classification

This project fine-tunes a BERT model to classify Persian comments into two categories: complaints about Product discrepancy (True) and not (False). The model is trained on the Basalam Comments dataset.

🛠 Training Details

• Base Model: HooshvareLab/bert-fa-base-uncased
• Fine-Tuning Dataset: Basalam comments
• NoteBook
• Evaluation Metrics:
  • Accuracy: 95.89%
  • F1 Score: 95.62%

📥 How to Use

You can load and use the fine-tuned model as follows:

from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

def classify_comment(text):
    model_name = "alireza-2003/bert-fa-discrepancy-detection"
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        outputs = model(**inputs)
    prediction = torch.argmax(outputs.logits).item()
    
    return "Discrepancy Complaint" if prediction == 1 else "Not a Complaint"

comment = "دو تا سفارش داده بودم  یدونه ابی و یدونه قرمز ولی هردوتاش قرمز بود"
print(classify_comment(comment))

---

📝 Author: [Alireza]
📅 Last Updated: [2/16/2025]
🔗 Dataset: Kaggle Dataset