Sadman4701/Apricity-Final

Emotion DeBERTa – 5-Class Emotion Classifier

Model Description

Emotion DeBERTa – 5-Class Emotion Classifier

Model Description

This model is a fine-tuned version of DeBERTa-v3-base for emotion classification. It predicts one of five emotional states from input text:

• anger
• fear
• joy
• sadness
• surprise

The model was trained as part of a university capstone project focused on building an emotion-aware mental healthcare companion.

Base Model

• microsoft/deberta-v3-base

Training Details

• Task: Text-based emotion classification
• Architecture: DeBERTa encoder with a custom classification head
• Number of labels: 5
• Training method: Supervised fine-tuning
• Output: Single-label emotion prediction

The model was originally trained using a custom PyTorch class and later converted into Hugging Face format for deployment and reproducibility.

Intended Use

This model is designed for:

• Emotion-aware chat applications
• Mental health companion systems
• Sentiment and emotional analysis in academic projects
• Research and educational purposes

It is not intended for clinical diagnosis or professional mental health decisions.

Limitations

• Trained on a limited dataset
• May not generalize well to:
  • Slang-heavy text
  • Code-mixed or multilingual inputs
  • Highly sarcastic or ambiguous sentences
• Predictions should be treated as probabilistic, not factual

Example Usage

from transformers import pipeline

classifier = pipeline(
    task="text-classification",
    model="Sadman4701/Apricity-Final",
    return_all_scores=True
)

text = "I feel scared but also strangely hopeful about the future."

outputs = classifier(text)

THRESHOLD = 0.5 #change it according to your preferences
predicted_emotions = [
    o["label"] for o in outputs[0] if o["score"] >= THRESHOLD
]

print(predicted_emotions)