Sentiment Analysis with Fine-Tuned DistilBERT

Fine-tuning a pre-trained transformer model for movie review sentiment classification.

Results

• Accuracy: 85.5%
• Base model: distilbert-base-uncased
• Dataset: IMDB (25,000 movie reviews)
• Training: 10 epochs using HuggingFace Trainer API

Steps Taken

1. Loaded pre-trained DistilBERT model from HuggingFace
2. Prepared IMDB dataset with proper tokenization
3. Fine-tuned using transfer learning on sentiment classification task
4. Evaluated on held-out test set

Technologies

• Python
• HuggingFace Transformers
• PyTorch
• Google Colab (free GPU)

Key Learnings

• Transfer learning dramatically reduces training time and data requirements
• Pre-trained language models adapt quickly to domain-specific tasks
• HuggingFace Transformers library makes state-of-the-art NLP accessible

How to Run

Open the notebook in Google Colab and run all cells.