roberta-emotion-7class

RoBERTa Emotion Classifier (7-class)

Fine-tuned RoBERTa model for emotion classification on 7 emotions: happy, sad, angry, fear, disgust, surprise, neutral.

Model Details

• Developed by: VanshajR
• Base Model: roberta-base (125M parameters)
• Task: Multi-class emotion classification
• Dataset: GoEmotions (27 emotions mapped to 7)
• Training Samples: ~58,000
• Language: English
• License: MIT

Performance

Evaluated on GoEmotions test set:

Per-Class Performance

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("VanshajR/roberta-emotion-7class")
model = AutoModelForSequenceClassification.from_pretrained("VanshajR/roberta-emotion-7class")

# Classify emotion
text = "I'm so excited about this project!"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=128)

with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_class = torch.argmax(predictions, dim=-1).item()

# Emotion labels
emotions = ["happy", "sad", "angry", "fear", "disgust", "surprise", "neutral"]
print(f"Predicted emotion: {emotions[predicted_class]}")
print(f"Confidence: {predictions[0][predicted_class].item():.2%}")

Training Details

Training Data

• Dataset: GoEmotions (Google Research)
• Emotion Mapping: 27 fine-grained emotions → 7 basic emotions
• Training Samples: ~58,000 Reddit comments
• Preprocessing: Truncation to 128 tokens, lowercase normalization

Training Procedure

• Optimizer: AdamW (lr=2e-5, weight_decay=0.01)
• Batch Size: 16 (train), 32 (eval)
• Epochs: 3
• Max Length: 128 tokens
• Training Regime: fp32

Compute Infrastructure

• Hardware: NVIDIA RTX 3070 (8GB VRAM)
• Training Time: ~2 hours
• Framework: PyTorch 2.1.0, Transformers 4.35.0

Limitations and Bias

• Language: English only
• Domain: Primarily trained on Reddit comments (may not generalize to formal text)
• Class Imbalance: Better performance on frequent emotions (happy, neutral) vs rare emotions (fear, disgust)
• Subjective Task: Human annotators often disagree on emotions (~25-30% disagreement rate)

Intended Use

✅ Recommended:

• Emotion detection in conversational text
• Evaluating emotion-controlled text generation
• Research on emotion understanding in dialogue
• Sentiment analysis applications

❌ Not Recommended:

• Clinical diagnosis or mental health assessment
• High-stakes decision making
• Non-English languages

Citation

@misc{vanshajr2024roberta,
  author = {Vanshaj R},
  title = {RoBERTa Emotion Classifier for 7-Class Emotion Detection},
  year = {2024},
  publisher = {HuggingFace},
  url = {https://huggingface.co/VanshajR/roberta-emotion-7class}
}

Related Work

Part of the Emotion-Controlled Response Generation project:

• 🔗 GitHub Repository
• 🔗 GPT-2 Emotion-Conditioned Model
• 📄 Full Project Report