Social Recommendation Terms: Probabilistic Explanation Optimization

Jie Liu, Lin Zhang, Victor S. Sheng, Yuanjun Laili

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

Abstract

The Probabilistic Matrix Factorization (PMF) model has been widely studied for recommender systems, which outperform previous models with a solid probabilistic explanation. To further improve its accuracy by using social information, researchers attempt to combine the PMF model with social network graphs by adding social terms. However, existing works on social terms do not provide theoretical explanations to make the models well understood. The lack of explanations limits further improvement of prediction accuracy. Hence, in this paper we provide our explanation and propose a unified covariance framework to solve this problem. Our explanation, including regularization terms, factorization terms and an ensemble of them, reveals how most social terms work from a probabilistic view. Our framework shows that those terms could be optimized in a direct way compatible to PMF. We find out that accuracy improvements for existing works on regularization terms rely more on personalized properties, and that social information for factorization terms is helpful but not always necessary.

Original languageEnglish
Title of host publicationChallenges and Opportunity with Big Data - 19th Monterey Workshop 2016, Revised Selected Papers
EditorsLin Zhang, Lei Ren, Fabrice Kordon
PublisherSpringer-Verlag
Pages155-167
Number of pages13
ISBN (Print)9783319619934
DOIs
StatePublished - 2017
Event19th Monterey Workshop on Challenges and Opportunity with Big Data, 2016 - Beijing, China
Duration: Oct 8 2016Oct 11 2016

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10228 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference19th Monterey Workshop on Challenges and Opportunity with Big Data, 2016
Country/TerritoryChina
CityBeijing
Period10/8/1610/11/16

Keywords

  • Factorization terms
  • Probabilistic matrix factorization
  • Regularization terms
  • Social networks

Fingerprint

Dive into the research topics of 'Social Recommendation Terms: Probabilistic Explanation Optimization'. Together they form a unique fingerprint.

Cite this