Protecting the Privacy of Observable Behavior in Distributed Recommender Systems
Douglas W. Oard, Anton Leuski, Stuart G. Stubblebine · 2003
this paper we use U = user, I = item, B = behavior, R = recommendation, and F = feature. Centralized recommender systems based on implicit feedback often map from a UIB array of observations to a UI matrix of recommendations. This matrix is then used to form either an I matrix of item similarity that can be used as a basis for "cross-selling," or U matrix of user similarity that can be used to find users with similar tastes, with which recommendations can then be constructed using the UI matrix