| --- |
| language: ta |
| license: apache-2.0 |
| tags: |
| - tamil |
| - emotion-classification |
| - text-classification |
| - fine-tuned |
| - multilingual |
| base_model: jusgowiturs/autotrain-tamil_emotion_11_tamilbert-2710380899 |
| pipeline_tag: text-classification |
| --- |
| |
| # Tamil Text Emotion Recognition Model |
|
|
| Fine-tuned Tamil language model for **11-class emotion classification** in Tamil text. |
| Detects: Ambiguous, Anger, Anticipation, Disgust, Fear, Joy, Love, Neutral, Sadness, Surprise, Trust. |
| Achieves ~94.5% accuracy on validation set after 6 epochs of fine-tuning. |
|
|
| ## Model Details |
|
|
| ### Model Description |
|
|
| - **Developed by:** Shanuka B Serasinghe |
| - **Shared by:** Shanuka B Serasinghe |
| - **Model type:** Text Classification (fine-tuned transformer for multi-class emotion detection) |
| - **Language(s) (NLP):** Tamil (தமிழ்) |
| - **License:** Apache-2.0 |
| - **Finetuned from model:** jusgowiturs/autotrain-tamil_emotion_11_tamilbert-2710380899 (AutoTrain-generated Tamil-BERT style checkpoint) |
| |
| ### Model Sources |
| |
| - **Repository:** https://huggingface.co/ShanukaB/Tamil_Emotion_Recognition_Model |
|
|
|
|
| ## Uses |
|
|
| ### Direct Use |
|
|
| Direct inference with Hugging Face `pipeline` for classifying Tamil sentences/comments into one of 11 emotions. |
|
|
| ### Downstream Use |
|
|
| - Building emotion-aware Tamil chatbots |
| - Tamil social media sentiment & emotion monitoring |
| - Mental health & emotional wellbeing applications in Tamil |
| - Customer support systems with emotion detection |
| - Further research/fine-tuning in low-resource Tamil NLP |
|
|
| ### Out-of-Scope Use |
|
|
| - High-stakes automated decisions (e.g. mental health diagnosis, hiring, legal) |
| - Real-time safety-critical systems without human oversight |
| - Non-Tamil languages (performance will be very poor) |
|
|
| ## Bias, Risks, and Limitations |
|
|
| - Best performance on short-to-medium informal/colloquial Tamil text (social media style) |
| - Heavy code-mixing (Tamil + English) reduces accuracy |
| - Sarcasm, irony, subtle emotions, strong dialects, or very formal/literary Tamil may be misclassified |
| - Potential biases from training data (e.g. over-representation of certain topics/styles in emotion datasets) |
| - Not robust to adversarial inputs or out-of-distribution text |
|
|
| ### Recommendations |
|
|
| - Always combine model predictions with human review in sensitive use-cases |
| - Test thoroughly on your specific domain/dialect before deployment |
| - Report issues (especially dialect or code-mixed failures) to improve future versions |
|
|
| ## How to Get Started with the Model |
|
|
| ```python |
| from transformers import pipeline |
| |
| classifier = pipeline( |
| "text-classification", |
| model="YOUR_USERNAME/YOUR_MODEL_NAME", |
| tokenizer="YOUR_USERNAME/YOUR_MODEL_NAME" |
| ) |
| |
| texts = [ |
| "இது ரொம்ப அழகா இருக்கு! 🥰🥰", |
| "என்னடா இது… மிகவும் கோபமா வருது", |
| "யாரும் இல்லாம தனிமையா ஃபீல் பண்றேன் 😔", |
| "அடேங்கப்பா! இது எப்படி சாத்தியமா? 😲" |
| ] |
| |
| for text in texts: |
| result = classifier(text)[0] |
| print(f"Text: {text}") |
| print(f"→ {result['label']} (confidence: {result['score']:.3f})\n") |
