E-motionAssistant/SInhala_Text_Emotion_Recognition_Model

Sinhala Text Emotion Recognition Model

Fine-tuned RoBERTa-style transformer for multi-class emotion classification in Sinhala text.
Detects basic emotions from Sinhala sentences/comments (e.g. social media, news).
Trained for 6 epochs on a Sinhala emotion dataset; validation accuracy 86% (modest performance – typical for initial fine-tuning in low-resource Sinhala NLP; suggest more epochs or Sinhala-pretrained base for better results).

Model Details

Model Description

• Developed by: Bimsara Serasinghe
• Shared by: Bimsara Serasinghe
• Model type: Text Classification (fine-tuned encoder-only transformer for multi-class emotion detection)
• Language(s) (NLP): Sinhala (සිංහල)
• License: Apache-2.0
• Finetuned from model: NLPC-UOM/SinBERT-large

Model Sources

• Repository: https://huggingface.co/ShanukaB/SInhala_Text_Emotion_Recognition_Model

Uses

Direct Use

Classify Sinhala text directly via Hugging Face pipeline into one of the emotion classes.

Downstream Use

• Emotion-aware Sinhala chatbots & virtual assistants
• Monitoring emotions in Sinhala social media (Facebook comments, YouTube, Twitter/X)
• Mental health & wellbeing tools for Sinhala speakers
• Customer support emotion detection in Sinhala
• Academic/research projects on low-resource Sinhala affective computing

Out-of-Scope Use

• High-stakes automated decisions (e.g. psychological diagnosis, legal judgments)
• Real-time safety-critical systems without human validation
• Non-Sinhala languages (expected very poor performance)

Recommendations

• Always pair model outputs with human review for sensitive applications (mental health, support)
• Fine-tune longer or switch to Sinhala-specific pre-trained models (e.g. SinBERT variants if available)
• Test on your target domain (e.g. news vs. casual chat) before deployment
• Report dialect/code-mixed failures to improve community versions

How to Get Started with the Model

from transformers import pipeline
import joblib  # if using saved label encoder

classifier = pipeline(
    "text-classification",
    model="YOUR_USERNAME/YOUR_MODEL_NAME",
    tokenizer="YOUR_USERNAME/YOUR_MODEL_NAME"
)

# Optional: load label encoder if uploaded to repo
# label_encoder = joblib.load("label_encoder.pkl")

texts = [
    "මම ගොඩක් සතුටින් ඉන්නවා! 😊",
    "මේක බලල බයයි වෙලා... 😨",
    "අපිට මේක ගැන කෝපයි ගොඩක්!"
]

for text in texts:
    result = classifier(text)[0]
    # If labels are "LABEL_0" etc., map manually or use saved encoder
    print(f"Text: {text}")
    print(f"→ Emotion: {result['label']} (confidence: {result['score']:.3f})\n")