Improved collaborative filtering

Aviv Nisgav, Boaz Patt-Shamir

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


We consider the interactive model of collaborative filtering, where each member of a given set of users has a grade for each object in a given set of objects. The users do not know the grades at start, but a user can probe any object, thereby learning her grade for that object directly. We describe reconstruction algorithms which generate good estimates of all user grades ("preference vectors") using only few probes. To this end, the outcomes of probes are posted on some public "billboard", allowing users to adopt results of probes executed by others. We give two new algorithms for this task under very general assumptions on user preferences: both improve the best known query complexity for reconstruction, and one improving resilience in the presence of many users with esoteric taste.

Original languageEnglish
Title of host publicationAlgorithms and Computation - 22nd International Symposium, ISAAC 2011, Proceedings
Number of pages10
StatePublished - 2011
Event22nd International Symposium on Algorithms and Computation, ISAAC 2011 - Yokohama, Japan
Duration: 5 Dec 20118 Dec 2011

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume7074 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference22nd International Symposium on Algorithms and Computation, ISAAC 2011


Dive into the research topics of 'Improved collaborative filtering'. Together they form a unique fingerprint.

Cite this