Recentprojectsin collaborativefiltering andinformationfiltering addressthetaskof inferring userpreferencerelationshipsfor productsor information.Thedataon which theseinferencesarebasedtypically consistsof pairsof peopleanditems.Theitemsmaybeinformationsources(suchaswebpagesor newspaper articles)or products(suchasbooks,software,movies or CDs). We areinterestedin makingrecommendationsor predictions.Traditionalapproachesto theproblemderive from classicalalgorithmsin statistical patternrecognitionandmachinelearning. The majority of theseapproachesassumea ”flat” datarepresentationfor eachobject,andfocuson a singledyadicrelationshipbetweentheobjects. In this paper , we examinea richermodelthatallows us to reasonaboutmany differentrelationsat thesametime. We build ontherecentwork onprobabilisticrelationalmodels(PRMs), anddescribehow PRMscanbeappliedto the taskof collaborati vefiltering. PRMsallow usto represent uncertaintyabouttheexistenceof relationshipsin themodelandallow thepropertiesof anobjectto dependprobabilisticallybothon otherpropertiesof that objectandon propertiesof relatedobjects.
[1]
Avi Pfeffer,et al.
Probabilistic Frame-Based Systems
,
1998,
AAAI/IAAI.
[2]
Wai Lam,et al.
LEARNING BAYESIAN BELIEF NETWORKS: AN APPROACH BASED ON THE MDL PRINCIPLE
,
1994,
Comput. Intell..
[3]
V. Rich.
Personal communication
,
1989,
Nature.
[4]
Thomas Hofmann,et al.
Latent Class Models for Collaborative Filtering
,
1999,
IJCAI.
[5]
David Heckerman,et al.
Empirical Analysis of Predictive Algorithms for Collaborative Filtering
,
1998,
UAI.
[6]
Dean P. Foster,et al.
A Formal Statistical Approach to Collaborative Filtering
,
1998
.
[7]
D. Rubin,et al.
Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper
,
1977
.
[8]
Lise Getoor,et al.
Learning Probabilistic Relational Models
,
1999,
IJCAI.