Implementation of Product Recommendation System (Collaborative Filtering)
Collaborative Filtering — the most common approach for product recommendations: "users similar to you bought this". Doesn't require product descriptions — only interaction history. Works for any domain, but requires sufficient transaction volume (>50K) and struggles with cold start.
ALS (Alternating Least Squares) Matrix Factorization
Collaborative Filtering with sufficient data volume: NDCG@10 of 0.25-0.45. Key hyperparameters: factors=64-128, iterations=15-30, regularization=0.001-0.1. Event weighting: view=1, add to cart=3, purchase=5, repeat purchase=8.
ALS on 1M users × 100K products trains in 5-15 minutes on CPU (8 threads). Minimum volume for successful training: 50K unique user-item pairs.