Collaborative vs Content-Based Filtering

Collaborative filtering finds patterns in user-item interactions without understanding item content. There are two main approaches:

User-based CF identifies users similar to you and recommends their favorites. "Users who are similar to you also liked..."

Item-based CF finds items that are frequently co-interacted with. "Users who liked this also liked..."

Matrix Factorization (SVD, ALS) learns latent factors representing user preferences and item characteristics. Netflix famously improved recommendations by 10% using matrix factorization during the Netflix Prize.

The key advantage: collaborative filtering can surface unexpected discoveries because it doesn't require understanding why items are similar.

Content-based systems analyze item features (genre, tags, descriptions, images) and match them to user preference profiles built from their history.

Production systems typically combine multiple approaches:

Weighted hybrid — Score items with multiple models, combine scores with learned weights.

Switching hybrid — Use different models for different situations (content-based for new users, collaborative for established users).

Feature augmentation — Use content-based features as input to collaborative models.

Cascade — Use one model to generate candidates, another to rank them.

For example, Netflix combines collaborative filtering for discovery, content-based for coverage, and knowledge-based rules for business constraints (e.g., regional content availability).

Select based on your data and business needs:

Most successful recommendation systems evolve: start simple, measure, and add complexity where it provides measurable lift.