Personalized information access tools are frequently based on collaborative filtering recommendation algorithms. Collaborative filtering recommender systems typically suffer from a data sparsity problem, where systems do not have sufficient user data to generate accurate and reliable predictions. Prior research suggested using group-based user data in the collaborative filtering recommendation process to generate group-based predictions and partially resolve the sparsity problem. Although group recommendations are less accurate than personalized recommendations, they are more accurate than general non-personalized recommendations, which are the natural fall back when personalized recommendations cannot be generated. In this work we present initial results of a study that exploits the browsing logs of real families of users gathered in an eHealth portal. The browsing logs allowed us to experimentally compare the accuracy of two group-based recommendation strategies: aggregated group models and aggregated predictions. Our results showed that aggregating individual models into group models resulted in more accurate predictions than aggregating individual predictions into group predictions. © Springer-Verlag Berlin Heidelberg 2009.
CITATION STYLE
Berkovsky, S., Freyne, J., & Coombe, M. (2009). Aggregation trade offs in family based recommendations. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 5866 LNAI, pp. 646–655). https://doi.org/10.1007/978-3-642-10439-8_65
Mendeley helps you to discover research relevant for your work.