Joseph A. Konstan, Sean M. McNee, Cai-Nicolas Ziegler, Roberto Torres, Nishikant Kapoor, John T. Riedl
Automated recommender systems predict user preferences by applying machine learning techniques to data on products, users, and past user preferences for products. Such systems have become increasingly popular in entertainment and e-commerce domains, but have thus far had little success in information-seeking domains such as identifying published research of interest. We report on several recent publications that show how recommenders can be extended to more effectively address information-seeking tasks by expanding the focus from accurate prediction of user preferences to identifying a useful set of items to recommend in response to the user's specific information need. Specific research demonstrates the value of diversity in recommendation lists, shows how users value lists of recommendations as something different from the sum of the individual recommendations within, and presents an analytic model for customizing a recommender to match user information-seeking needs.
Subjects: 1.10 Information Retrieval; 12. Machine Learning and Discovery