期刊名称：Bulletin of the Technical Committee on Data Engineering
出版年度：2009
卷号：32
期号：04
出版社：IEEE Computer Society
摘要：Past research in recommender systems has mainly focused on improving accuracy, i.e., making each
single recommendation get as close to the user’s information need as possible. However, while this ap-
proach works well when focusing on single recommendations as atomic entities, its usefulness to the
consumer appears limited when considering entire recommendation lists along with their overall util-
ity, which often appear to provide rather an unbalanced diet to the user: Recommendation lists seldom
reflect the consumer’s entire spectrum of interest but rather hook on to small portions that appear partic-
ularly favorable with regard to accuracy optimization. We analyze the diversification issue in detail and
present a framework that is geared towards making lists as interesting and colorful as possible, trading
a minimum of accuracy in exchange for the gain in diversity. Empirical evaluations aiming for actual
user satisfaction underpin the cogency of our approach.
