Multi-Label Movie Genre Classification
Oct 2025
Finished
Academic project for multi-label movie genre classification (18 classes: Action, Comedy, Drama, Horror, Sci-Fi, etc.) using title and plot description. Implements and compares three methodologies: (1) Classical Logistic Regression with MultiLabelBinarizer (scikit-learn), (2) Transformer models — DistilBERT and RoBERTa — fine-tuned for multi-label classification with HuggingFace Trainer, 50 epochs, early stopping, and F1 metric, (3) Gemini (Google Generative AI models) with few-shot prompting: a prompt is built with train set examples and structured JSON classification is requested using Enum types for genres. Batch processing with retries on API errors. Includes analysis notebooks with F1-per-epoch visualizations and cross-method comparisons.
AI
HuggingFace
Jupyter
NumPy
Pandas
Plotly
Python
Sklearn
