Mining of Real-world Hypergraphs: Patterns, Tools, and Generators

Abstract

Groupwise interactions are prevalent in various complex systems (e.g., collaborations of researchers and group discussions on online Q&A sites), and they are commonly modeled as hypergraphs. Hyperedges, which compose a hypergraph, are non-empty subsets of any number of nodes, and thus each hyperedge naturally represents a groupwise interaction among entities. The higher-order nature of hypergraphs brings about unique structural properties that have not been considered in ordinary pairwise graphs. Recent studies have extensively investigated the structures and dynamics of real-world hypergraphs, using specialized tools, and discovered many interesting patterns pervasive in them. In addition, they demonstrated that many of such patterns can be reproduced and thus explained by simple hypergraph generative models that are based on intuitive mechanisms on individual nodes or hyperedges. In this tutorial, we offer a comprehensive overview of recent discoveries about real-world hypergraphs and data mining tools for them. Specifically, we first introduce a wide range of usage of hypergraphs in diverse fields. Then, we present recently revealed structural properties of real-world hypergraphs, including macroscopic and microscopic patterns, static and dynamic patterns, and connectivity and overlapping patterns. With the patterns, we also describe advanced data mining tools designed for such analyses. Lastly, we introduce simple yet realistic hypergraph generative models that successfully reproduce and thus explain the structural properties of real-world hypergraphs.

Presenter

Geon Lee, Jaemin Yoo, Kijung Shin

Graph Neural Networks for Tabular Data Learning

Abstract
The deep learning‐based approaches to Tabular Data Learning (TDL) have shown competing performance, compared to their conventional counterparts. However, the latent correlation among data instances and feature values is less modeled in deep neural TDL. Recently, graph neural networks (GNN), which can enable modeling relations and interactions between different data entities, has received tremendous attention across application domains including TDL. It turns out creating proper graph structures from the input tabular data, along with GNN learning, can improve the TDL performance. In this tutorial, we will systematically introduce the methodologies of designing and applying GNN to TDL. The topics to be covered include: (1) foundations and overview of GNN‐based TDL methods; (2) a comprehensive taxonomy of constructing graph structures and representation learning in GNN‐based TDL methods; (3) how to apply GNN to various TDL application scenarios and tasks; (4) limitations in current research and future directions.

Presenter

Cheng-Te Li, Yu-Che Tsai, Jay Chiehen Liao