Copula Mixed-Membership Stochastic Blockmodel

Xuhui Fan, Richard Yi Da Xu, Longbing Cao

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

11 Citations (Scopus)

Abstract

The Mixed-Membership Stochastic Blockmodels (MMSB) is a popular framework for modelling social relationships by fully exploiting each individual node's participation (or membership) in a social network. Despite its powerful representations, MMSB assumes that the membership indicators of each pair of nodes (i.e., people) are distributed independently. However, such an assumption often does not hold in real-life social networks, in which certain known groups of people may correlate with each other in terms of factors such as their membership categories. To expand MMSB's ability to model such dependent relationships, a new framework - a Copula Mixed-Membership Stochastic Blockmodel - is introduced in this paper for modeling intra-group correlations, namely an individual Copula function jointly models the membership pairs of those nodes within the group of interest. This framework enables various Copula functions to be used on demand, while maintaining the membership indicator's marginal distribution needed for modelling membership indicators with other nodes outside of the group of interest. Sampling algorithms for both the finite and infinite number of groups are also detailed. Our experimental results show its superior performance in capturing group interactions when compared with the baseline models on both synthetic and real world datasets.

Original languageEnglish
Title of host publicationProceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence
EditorsSubbarao Kambhampati
Place of PublicationCalifornia
PublisherInternational Joint Conferences on Artificial Intelligence
Pages1462-1468
Number of pages7
ISBN (Electronic)9781577357704, 9781577357711
Publication statusPublished - 2016
Externally publishedYes
Event25th International Joint Conference on Artificial Intelligence, IJCAI 2016 - New York, United States
Duration: 9 Jul 201615 Jul 2016

Conference

Conference25th International Joint Conference on Artificial Intelligence, IJCAI 2016
Country/TerritoryUnited States
CityNew York
Period9/07/1615/07/16

Fingerprint

Dive into the research topics of 'Copula Mixed-Membership Stochastic Blockmodel'. Together they form a unique fingerprint.

Cite this