Modelpublic

ilkayO/yildizsezar-convbert

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes19downloads

Model Card

<div align="center"> <img src="https://huggingface.co/ilkayO/yildizsezar-convbert/resolve/main/yildizsezar.png" width="220" alt="YildizSezar Logo"/> <p style="color: #6b7280; font-size: 16px; font-style: italic; margin-top: 5px; font-family: sans-serif;"> "Geldim, Gördüm, Yıldızladım." </p> </div>

🌟 YıldızSezar: Turkish E-Commerce Review ConvBERT

![Hugging Face Spaces](https://huggingface.co/spaces/ilkayO/YildizSezar-Demo) ![Open In Colab](https://colab.research.google.com/github/ilkay-onay/YildizSezar-Review-Classifier/blob/main/YildizSezar_Demo.ipynb) ![GitHub](https://github.com/ilkay-onay/YildizSezar-Review-Classifier)

YıldızSezar is a fine-tuned version of dbmdz/convbert-base-turkish-mc4-uncased, specifically engineered to predict 1-to-5 star ratings directly from morphologically complex Turkish e-commerce customer reviews.

This model serves as the official implementation for my B.Sc. Thesis Project, which was successfully published in a peer-reviewed journal. 📄 Paper: A Star Rating-Based Approach in BERT-Based Sentiment Analysis of Customer Feedback

🚀 Model Details

—Architecture: ConvBERT (Turkish)
—Task: Multi-class Sentiment Analysis
—Classes: 5 (1 Yildiz, 2 Yildiz, 3 Yildiz, 4 Yildiz, 5 Yildiz)
—Language: Turkish
—License: Apache 2.0 (Commercially friendly)

📊 Performance & Training

Real-world e-commerce data often suffers from extreme class imbalance. To overcome this, the model was trained on a highly robust dataset augmented with a scalable synthetic data generation pipeline via LLaMA-8B-DPO (generating over 900,000 synthetic review samples).

—Accuracy: 89.96%
—Macro F1-Score: 0.799
—Dataset: ilkayO/yildizsezar-turkish-reviews

💻 Quick Usage

You can easily integrate this model into your Python projects:

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "ilkayO/yildizsezar-convbert"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

text = "Ürünün kalitesi beklediğimden çok daha iyi çıktı."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=256)

with torch.no_grad():
    outputs = model(**inputs)
    predicted_class_id = torch.argmax(outputs.logits, dim=-1).item()

# Assuming id2label is properly configured in config.json
print(f"Predicted Rating: {model.config.id2label[str(predicted_class_id)]}")

🤝 Commercial Integration & Consulting

This model is open-sourced under the Apache 2.0 license, meaning you are free to use, modify, and integrate it into your commercial products.

If your company is looking to integrate advanced NLP systems, requires custom fine-tuning on internal data, or if you are interested in having me join your AI team, feel free to reach out! 📧 Contact: ilkayonay2001@gmail.com | LinkedIn