Aggregation trade offs in family based recommendations

Shlomo Berkovsky, Jill Freyne, Mac Coombe

Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review

9 Citations (Scopus)

Abstract

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.
Original languageEnglish
Title of host publicationAI 2009, Advances in Artificial Intelligence
Subtitle of host publication22nd Australasian Joint Conference. Proceedings
EditorsAnn Nicholson, Xiaodong Li
Place of PublicationBerlin
PublisherSpringer, Springer Nature
Pages646-655
Number of pages10
ISBN (Electronic)9783642104398
ISBN (Print)9783642104381
DOIs
Publication statusPublished - 2009
Externally publishedYes
Event22nd Australasian Joint Conference on Artificial Intelligence, AI 2009 - Melbourne, VIC, Australia
Duration: 1 Dec 20094 Dec 2009

Publication series

NameLecture Notes in Computer Science
PublisherSpringer
Volume5866
ISSN (Print)0302-9743

Other

Other22nd Australasian Joint Conference on Artificial Intelligence, AI 2009
CountryAustralia
CityMelbourne, VIC
Period1/12/094/12/09

Fingerprint Dive into the research topics of 'Aggregation trade offs in family based recommendations'. Together they form a unique fingerprint.

Cite this