Project 3: Movie Recommendation System

Build a recommendation system to suggest movies to users.

Starter code: Run starter.py after downloading MovieLens data to data/.

Difficulty

Intermediate

Time Estimate

5-7 days

Skills You'll Practice

• Collaborative Filtering
• Content-Based Filtering
• Matrix Factorization
• Recommendation Metrics

Learning Objectives

By completing this project, you will learn to:

• Implement collaborative filtering
• Build content-based recommendations
• Use matrix factorization
• Evaluate recommendation systems
• Handle cold-start problem

Dataset

MovieLens Dataset

• MovieLens Dataset
• User ratings for movies
• Movie metadata (genres, year, etc.)
• Start with 100K dataset

Project Steps

Step 1: Load and Explore Data

• Load ratings and movie data
• Analyze rating distribution
• Check for sparsity
• Explore user and movie statistics

Step 2: Collaborative Filtering

• User-based collaborative filtering
• Item-based collaborative filtering
• Calculate similarity matrices
• Generate recommendations

Step 3: Matrix Factorization

• Implement SVD (Singular Value Decomposition)
• Use scikit-surprise library
• Tune hyperparameters
• Generate recommendations

Step 4: Content-Based Filtering

• Extract movie features (genres, etc.)
• Calculate similarity between movies
• Recommend similar movies
• Combine with collaborative filtering

Step 5: Hybrid Approach

• Combine collaborative and content-based
• Weight different approaches
• Improve recommendation quality

Step 6: Evaluation

• Split data into train/test
• Calculate RMSE, MAE
• Use precision@k, recall@k
• Evaluate on test set

Expected Deliverables

1. Jupyter Notebook with all approaches
2. Recommendation Function that takes user ID and returns recommendations
3. Evaluation Report comparing different methods
4. Demo showing recommendations for sample users

Evaluation Metrics

• RMSE: Root Mean Squared Error
• MAE: Mean Absolute Error
• Precision@K: Top-K precision
• Recall@K: Top-K recall
• Diversity: Variety of recommendations

Approaches to Implement

1. User-Based CF: Find similar users
2. Item-Based CF: Find similar items
3. Matrix Factorization: SVD, NMF
4. Content-Based: Based on movie features
5. Hybrid: Combine multiple approaches

Tips

• Start with simple collaborative filtering
• Use scikit-surprise for matrix factorization
• Handle cold-start problem (new users/movies)
• Consider popularity bias
• Evaluate on held-out test set

Resources

• MovieLens Dataset
• scikit-surprise
• Recommender Systems Guide

Extensions

• Real-time recommendations
• Handle implicit feedback
• Deep learning approaches
• Deploy as web service
• Build web interface

Next Steps

After completing this project:

• Try larger datasets (1M, 10M ratings)
• Experiment with deep learning
• Move to Project 4: Fraud Detection