Collaborative filtering recommends products or content by finding users with similar behavior and using their preferences to predict what someone may want next. It gives a personalization effort a practical starting point when user-behavior data already exists, but it struggles with new users who have little or no interaction history. section 2.2.1 section 4.4.1
Two common approaches organize the comparison differently:
- User-based filtering compares users and recommends items favored by people with similar preferences. section 4.4.1
- Item-based filtering recommends products that are often purchased or viewed together. section 4.4.1
- The method can uncover relationships without collecting explicit preference data, although content-based or hybrid filtering can help address cold-start cases. section 4.4.1 section 4.5
From shared purchases to testing
If User A and User B both buy the same shoes, the system can recommend User A’s outfit purchase to User B. A product team can compare user-based and item-based filtering with A/B tests, then refine recommendations using clicks, purchases, browsing time, and explicit thumbs-up or thumbs-down feedback. section 4.4.1 section 4.4.2
