# Collaborative Filtering
**Entity class:** Recommendation method / social preference inference
## Definition and history
Collaborative filtering uses patterns of evaluation or behavior across people to predict which items a particular person may find relevant. [[articles/Intelligence Inside|Intelligence Inside]] traces the method from Tapestry's reader-writable annotation filters through GroupLens and Ringo to commercial item-based recommendation. The shared problem is selection from a larger available set; the systems differ in inputs, transparency, and what users can edit. A recommendation influences the next encounter, making later observations partly dependent on earlier selections.
## Mechanism and limits
Social annotations, explicit ratings, purchase histories, and item similarities are different forms of evidence. The method can expand discovery by connecting a reader to judgments made elsewhere, while it can also concentrate attention through recursive exposure. Tapestry is particularly instructive because its selection rule was inspectable to the reader rather than merely an opaque institutional score.
## Relationships
**Early systems:** [[wiki/Tapestry|Tapestry]], [[wiki/GroupLens|GroupLens]], and [[wiki/Ringo Recommender|Ringo Recommender]]. **Commercial method:** [[wiki/Item-Based Collaborative Filtering|Item-Based Collaborative Filtering]]. **Governance:** [[wiki/Reader-Inspectable Selection|Reader-Inspectable Selection]]. **Article route:** [[articles/Intelligence Inside|Intelligence Inside]].
## Sources
[Xerox PARC, Tapestry paper (1992)](https://www.bitsavers.org/pdf/xerox/parc/techReports/CSL-92-10_Using_Collaborative_Filtering_to_Weave_an_Information_Tapestry.pdf); [[articles/Intelligence Inside|Intelligence Inside]].