2000 character limit reached
Reducing offline evaluation bias of collaborative filtering algorithms (1506.04135v1)
Published 12 Jun 2015 in cs.IR, cs.LG, and stat.ML
Abstract: Recommendation systems have been integrated into the majority of large online systems to filter and rank information according to user profiles. It thus influences the way users interact with the system and, as a consequence, bias the evaluation of the performance of a recommendation algorithm computed using historical data (via offline evaluation). This paper presents a new application of a weighted offline evaluation to reduce this bias for collaborative filtering algorithms.
