Emotion Classification with Imbalance-Aware Transformers

This repository contains the implementation and experimental artefacts for an MSc dissertation on fine-grained multi-label emotion classification using the GoEmotions dataset.

Contents

• train_baseline_optimized-FINAL.py – Main training script (DeBERTa-v3-base)
• figures/ – Generated figures used in the dissertation
• tables/ – CSV tables used in appendices and analysis

Reproducibility

Due to file size constraints, trained model checkpoints are not included. All results can be reproduced by running the training script with the provided configuration.

Experiments were conducted using:

• Python 3.10+
• PyTorch
• HuggingFace Transformers
• Google Colab (GPU)

Random seeds were fixed to ensure reproducibility.