Collaborative Filtering

A recommendation technique that generates suggestions by analyzing the behavior patterns of many users. “Customers who bought this also bought that” is the classic formulation. Collaborative filtering requires a substantial volume of interaction data to function well, which makes it vulnerable to the cold‑start problem. It also tends to favor popular, broadly appealing items at the expense of niche or new products. For these reasons, the most effective recommendation architectures combine collaborative filtering with attribute‑based filtering, using the former for discovery when behavioral data is plentiful and the latter for relevance when it is sparse.

Example: An ecommerce site that only uses collaborative filtering will recommend a water bottle to every runner; a hybrid system that also reads shoe attributes will recommend a stability insole to a runner who bought a motion‑control shoe.