Collaborative Filtering

What Collaborative Filtering is

Collaborative filtering uses only the interaction matrix — who interacted with what — without needing to understand item content at all.

How it works

Neighbourhood methods compare user or item vectors directly; matrix factorisation learns latent user and item embeddings whose dot product predicts interaction strength.

Why it matters

It is the classic recommender technique and still a strong baseline, and its cold-start weakness is the reason hybrid systems exist.

Common uses

• →Retail recommendations
• →Music and video discovery
• →Cross-sell in banking