multilingual-sentiment-xlm-roberta

🌍 Multilingual Sentiment Classifier (XLM-RoBERTa)

This model is a fine-tuned version of xlm-roberta-base for multilingual sentiment classification across English, German, and Italian.

We built this model to classify sentiment into:

• 0 → Negative
• 1 → Neutral
• 2 → Positive

---

✍️ How We Built It

This model was fine-tuned using the Amazon Reviews Multilingual Dataset, specifically on the English, German, and Italian subsets.
Training was done using PyTorch and Hugging Face Transformers.

Preprocessing

• Texts were tokenized using XLMRobertaTokenizer
• Labels were mapped to integers (negative: 0, neutral: 1, positive: 2)
• Dataset was split into train/test/validation using an 80/10/10 ratio

Training

• Model: xlm-roberta-base
• Epochs: 2
• Optimizer: AdamW
• Batch size: 8
• Evaluation metric: Macro F1-score
• Hardware: Google Colab GPU

---

🔍 Example Usage

from transformers import pipeline

classifier = pipeline("sentiment-analysis", model="subba5076/multilingual-sentiment-xlm-roberta")

classifier("Der Film war unglaublich schön.")  # German
classifier("This phone is terrible.")           # English
classifier("È stato un buon acquisto.")         # Italian

📊 Evaluation Macro F1-score on test set: 0.81 Confusion matrix and training curves can be shared in future updates.

👨‍💻 Authors This project was developed as part of a team NLP assignment.

Team Members:

Subrahmanya Rajesh Nayak @subba5076

Rim Tafech

🪪 License This model is licensed under the MIT License.