Exploring the Impact of Users' Preference Diversity on Recommender System Performance

摘要

Recommender systems present an effective alternative to subject access in the domain of reading for leisure. They are particular valuable for the discovery of novel and serendipitous finds. In the recommender system evaluation literature, a trade-off has been recognized between accuracy and surpriseon-obviousness, and more recently, that between accuracy and diversity of the recommendation set. It is argued in this paper that a proper balance between accuracy and diversity might lie in users' 'preference diversity,' a construct we propose to represent how wide a user's reading interests are. Users with more diverse interests might appreciate more novel and diverse set of recommendation. Drawing from marketing literature and our own empirical studies, this paper discusses how the motivation for diversity might influence users' preference for cultural goods such as books and music, and more specifically their responses to recommendations. Future study is needed to further examine the relationship between user preference diversity and proper degree of diversity of the recommendation set.