README.md

metadata

license: cc-by-nc-4.0

Synthetic Emotion Classification Dataset

Dataset Overview

This dataset contains 13,970 synthetic text samples labeled across 7 emotion classes:

• Anger
• Happiness
• Sad
• Surprise
• Hate
• Love
• Fun

The data was generated using Mistral-7B, producing diverse and realistic short-to-medium length emotional expressions.
It is suitable for benchmarking NLP models such as RNNs, BERT-based models, and LLMs in multi-class emotion detection.

---

Example

Text:

John clenched his fists, his face turning red as he paced back and forth in the room.
His eyes flashed with frustration as he muttered under his breath about the latest setback at work.

Emotion: Anger

---

Dataset Statistics

• Total samples: 13,970
• Columns: text, emotion
• Emotion classes: 7 (balanced)
• Generator model: Mistral-7B
• Data type: Fully synthetic
• File format: CSV

---

Use Cases

• Train or fine-tune emotion classifiers (LSTM, GRU, DistilBERT, RoBERTa)
• Compare traditional ML models vs transformer-based models
• Evaluate LLMs in zero-shot or few-shot emotion classification
• Augment real-world datasets with balanced emotion samples
• Educational projects in NLP and sentiment analysis

---

Notes

• This dataset is fully synthetic and contains no personally identifiable information (PII).
• Labels were auto-generated using structured LLM prompting.
• Users should check for:
  • Duplicate samples
  • Prompt-induced stylistic bias
  • Generalization limits before production use

---

Compatibility

• Hugging Face datasets library
• Kaggle notebooks
• PyTorch / TensorFlow NLP pipelines