Problem. Build and compare deep learning models for text classification to analyze sentiment in book reviews, supporting the PyBooks recommendation engine.

Data. Custom book review samples and fetch_20newsgroups dataset (categories: rec.autos, sci.med, comp.graphics).

Approach.

• Implemented multiple deep learning models for text classification:
  • CNN — captures local word patterns using 1D convolutions.
  • RNN — sequential modeling for contextual dependencies.
  • LSTM — handles long-term dependencies and vanishing gradients.
  • GRU — simplified recurrent structure for efficient training.
• Used embeddings to represent words numerically.
• Compared architecture behaviors without focusing on raw metrics.

Results (qualitative).

• CNN effectively captured localized text patterns.
• LSTM and GRU demonstrated smoother learning for sequential context.
• RNN served as a conceptual baseline for understanding recurrence.

What I Learned.

• Fundamental architecture differences between CNNs and RNN-family models for NLP.
• How embeddings and tokenization shape model performance.
• Practical trade-offs between accuracy, complexity, and interpretability in deep learning.

git clone https://github.com/Joe-Naz01/text-classification-nlp.git
cd text-classification-nlp
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt
jupyter notebook