Mizo Sentiment Analysis (DistilBERT)
A fine-tuned distilbert-base-multilingual-cased model for binary sentiment classification (Positive vs. Negative) in the Mizo (lus) language.
Model Summary
- Model Architecture: DistilBERT (
distilbert-base-multilingual-cased) - Language: Mizo / Lushai (
lus) - Task: Binary Sentiment Analysis / Text Classification (
POSITIVE,NEGATIVE) - Parameters: ~134M
Performance & Evaluation
Evaluated on a held-out stratified validation set (583 samples):
| Metric | Score |
|---|---|
| Accuracy | 92.62% |
| F1 Score | 93.51% |
| Precision | 92.54% |
| Recall | 94.51% |
| Validation Loss | 0.2279 |
Dataset & Preprocessing
The model was fine-tuned on a merged collection of:
Neutral annotations were excluded to build a robust binary sentiment classifier. The dataset was deduplicated and stratified (3,301 train / 583 validation samples).
How to Use
from transformers import pipeline
# Load pipeline
classifier = pipeline("sentiment-analysis", model="<repo_id>")
# Run inference
sample_text = "Ka lawm e, a ropui hle mai."
result = classifier(sample_text)
print(result)
# Output: [{'label': 'POSITIVE', 'score': 0.9900}]
- Downloads last month
- 2