About This Project

This is an Emotion Classification System built using a DistilBERT Transformer model fine-tuned on over 100,000 sentences.


đź“„Details:

  • Model: DistilBERT (66 Million Parameters).
  • Tokenizer: DistilertTokenizerFast.
  • Dataset: A custom dataset by fusing 3 datasets (106k samples). Click here for the dataset.
  • Performance: Achieved ~89% accuracy compared to ~82% accuracy of a Logistic Regression model which used a TF-IDF vectorizer.
  • GPU: NVIDIA GeForce RTC 4060Ti.
GitHub