Collaborative Filtering

What is Collaborative Filtering?

Collaborative filtering (CF) makes recommendations by learning from user–item interactions (views, clicks, carts, purchases, ratings). It can be user-based (find users like you) or item-based (find items often co-interacted with) and scales with matrix factorization/embeddings from implicit feedback.

How It Works (quick)

• Signals: Implicit feedback (views, add-to-cart, purchases, dwell) with confidence weights and recency decay.
• Models: Item-item similarity, user-user similarity, or latent-factor models (ALS/BPR/Neural CF).
• Serving: For a user/session, score candidate items → filter (availability/price/size) → diversify → render.
• Cold start mitigation: Back off to content-based or popularity by segment for new users/items.

Why it Matters in E-commerce

• Personalization at scale: Tailors listings per shopper or segment.
• Cross-sell/upsell: Learns bundles and complements automatically.
• Low setup: Works even with minimal metadata if interactions exist.

Best Practices

• Prefer implicit-feedback training with confidence weighting.
• Apply recency and seasonality decay; separate weekday/holiday patterns.
• Enforce availability & size-in-stock constraints before ranking.
• Add diversity/novelty caps to avoid echo chambers.
• Use hybrid: CF + content-based + business rules.
• Measure coverage, CTR/ATC, revenue per session, returns, and catalog lift.

Challenges

• Cold start: No signals for new items/users.
• Popularity bias: Over-recommends bestsellers; hurts long tail.
• Data leakage: Don’t learn from returned/cancelled orders without weighting.
• Drift: Trends change; retrain regularly.

Examples

• PDP: “Frequently bought together” from item-item co-purchase.
• Home: “Because you viewed X” from user-sequence embeddings.
• Cart: Complementary add-ons with price and inventory checks.

Summary

Collaborative filtering turns crowd behavior into personalized, high-performing recommendations. Pair it with content signals, stock rules, and diversity to keep results fresh, fair, and profitable.