You are now in the main content area

Seminar: Clustering Higher Order Relational Data

Date
December 19, 2023
Time
12:00 PM EST - 1:00 PM EST
Location
ENG-210 and virtually via zoom
Open To
All faculty, staff, students and guests are welcome to attend
Contact
Pawel Pralat (pralat@torontomu.ca)

Speaker: François Théberge

Title: Clustering Higher Order Relational Data

Abstract: Graphs are commonly used to represent binary relations between entities, and several tools exist to analyze such data. Many relations can involve more than two entities, such as co-authorship of papers, being part of the same sports team, or being in close physical contact for some period. Hypergraphs provide a convenient representation for relational data involving an arbitrary number of entities. However, tools and techniques to analyze such data are not as prevalent as with graph data, and for that reason, it is common practice to reduce a hypergraph to its two-section graph representation (replacing each hyperedge with a clique), and apply graph-based techniques, thus losing some information.

In this talk, we present some of the tools and techniques that we have been developing leading to hypergraph-based clustering algorithms. We first review \textbf{h-ABCD}, our benchmark model for hypergraphs with communities which allows us to compare clustering algorithms while controlling several aspects of the data. Modularity is a commonly used objective function when developing graph clustering algorithms; we present our generalized \textbf{hypergraph modularity} as well as some useful special cases. Finally, we present \textbf{h-Louvain}, our hypergraph clustering algorithm inspired by the popular Louvain algorithms for graphs and show some results on a few datasets.