A behavior-based recommender matches items through overlapping user interactions rather than item attributes. It powers suggestions such as “you may also like” by learning from browsing, clicking, adding to cart, and purchasing behavior. section 1.2.2 section 9.2.2
The underlying method is collaborative filtering: users who interact with similar items provide signals for recommending additional items. A new interest in fertilizer can therefore surface soil or mulch when other fertilizer-interested users also engaged with those products. The model needs unique item IDs and user–item interactions, not names, brands, colors, or text. More interaction signals can continually improve its crowdsourced relevance estimates. section 9.2.2
Behavior-based matching is strongest when items have many interaction signals and weak when they have few or none. A hybrid recommender can use content-based matching as a fallback, then give collaborative filtering more influence as behavioral evidence accumulates, reducing the cold-start problem for new content. section 9.2.3
