# Item-Based Collaborative Filtering
**Entity class:** Recommendation method / item similarities
## Definition and history
Item-based collaborative filtering estimates a user's likely interest in an item from relationships among items they and others have engaged with. [[articles/Intelligence Inside|Intelligence Inside]] cites Amazon's account of deploying this method from 1998. Its computational unit differs from social annotation rules in Tapestry, while both alter the selection of what a person encounters next. The method can scale recommendations across a large catalog with reusable item-to-item relationships.
## Mechanism and limits
An item's predicted relevance is conditional on a history already shaped by prior exposure and choice. Purchase or interaction records become input to future selection, which can reinforce the distribution of attention. The article's contrast concerns the visibility and writability of selection rules, not a claim that collaborative recommendations have only one implementation.
## Relationships
**Commercial implementation:** Amazon. **Method family:** [[wiki/Collaborative Filtering|Collaborative Filtering]]. **Historical comparison:** [[wiki/Tapestry|Tapestry]] and [[wiki/GroupLens|GroupLens]]. **Article route:** [[articles/Intelligence Inside|Intelligence Inside]].
## Sources
[Amazon Science, *Two Decades of Recommender Systems at Amazon*](https://cdn.amazon.science/76/9e/7eac89c14a838746e91dde0a5e9f/two-decades-of-recommender-systems-at-amazon.pdf); [[articles/Intelligence Inside|Intelligence Inside]].