SentiScope - Burmese Sentiment Analysis with XLM-RoBERTa

Overview

This repo contains a fine-tuned XLM-RoBERTa model specialized for analyzing sentiment in Burmese text.
The model classifies Burmese text into three sentiment categories:

• Positive
• Negative
• Neutral

It was trained on publicly available Burmese sentiment datasets with additional curated data and includes preprocessing to normalize encoding (e.g., Zawgyi → Unicode).

---

Model Details

• Base model: FacebookAI/xlm-roberta-base
• Task: Sentiment classification
• Languages: Burmese (my), supports multilingual inputs
• Developer: Yoon Thiri Aung (GitHub profile)
• Demo Here!

---

Uses

Usage Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "emilyyy04/burmese-sentiment-xlm-roberta"  # Replace with your repo name
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "ဒီဇာတ်လမ်းက တကယ်ကောင်းတယ်။"
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
outputs = model(**inputs)
predicted_class = torch.argmax(outputs.logits, dim=1).item()

label_map = {0: "positive", 1: "negative", 2: "neutral"}
print("Predicted Sentiment:", label_map[predicted_class])

---

Direct Use

• Sentiment classification of Burmese text from social media, reviews, comments, and other user-generated content.
• Building sentiment-aware Burmese NLP applications such as chatbots, analytics dashboards, and content moderation tools.

Limitations

• May not generalize well to domains significantly different from the training data.
• May misclassify sentences with mixed sentiments or sarcasm.
• Performance may drop for code-mixed Burmese-English text with heavy slang or informal spelling.

---

Training Details

Training Data

• Sources:

  • kalixlouiis/burmese-sentiment-analysis
  • chuuhtetnaing/myanmar-social-media-sentiment-analysis-dataset
  • Additional curated data collected and annotated by the author.

• Preprocessing:

  • Converted Zawgyi-encoded text to Unicode.
  • Cleaned and normalized text fields.
  • Tokenized using the XLM-RoBERTa tokenizer with:
    • max_length=128
    • Truncation and padding to maximum length.

Training Procedure

• Optimizer: AdamW (default in Hugging Face Trainer)
• Learning rate: 2e-5
• Batch size: 8 (train & eval)
• Epochs: 3
• Weight decay: 0.01
• Mixed precision (fp16): Enabled when training on GPU
• Metric for best model: F1 score (weighted average)
• Evaluation strategy: Per epoch
• Model selection: Best F1 score checkpoint

---

Evaluation

Metrics

The model was evaluated on a held-out validation set using accuracy, precision, recall, and F1 score.

Epoch	Val Loss	Accuracy	Precision	Recall	F1
1	0.6171	0.7859	0.7994	0.7859	0.7875
2	0.4268	0.8470	0.8465	0.8470	0.8464
3	0.4115	0.8451	0.8447	0.8451	0.8448

The final model used is the checkpoint with the highest F1 score.

Limitations

• Performance may degrade on domain-shifted data or heavy code-mixed Burmese-English text.
• Sarcasm and mixed sentiments are challenging.

License

This project is licensed under the MIT License.

---

Feel free to open issues or contribute improvements!