Heterogeneous Graph Neural Networks: Techniques and Applications
Heterogeneous Graph Neural Networks: Techniques and Applications KDD2021 Tutorial
Abstract

Heterogeneous graphs (HGs) a.k.a heterogeneous information networks (HINs), which are able to consist of different types of entities and relations, have become ubiquitous in real-world scenarios such as bibliographic networks, social networks, and recommendation systems. Due to the ubiquity, HG representation learning which aims to learn nodes representations in a low-dimensional space while preserving the heterogeneous structural and semantic information is an emerging research field in data mining and machine learning. To solve the HG representation learning problem, leveraging advances of graph neural networks (GNNs), the research on heterogeneous graph neural networks (HGNNs) has attracted considerable attention in recent years. Different from traditional GNNs, due to the heterogeneity, HGNNs face some unique challenges: (1) the structures in HGs are usually semantic dependent; (2) different types of nodes and edges have different attributes encoded in different feature spaces; (3) HG representation learning is usually application/domain dependent.

In this tutorial, we will give a survey on recently developed HGNN techniques and their applications. More specifically, the tutorial will include following aspects: (1) basic concepts in HGNNs; (2) research challenges of HGNNs; (3) state-of-the-art HGNN techniques; (4) extensive applications of HGNNs; and (5) frontiers of research in this field. This tutorial will enable researchers and practitioners to share new techniques of HGNNs and give illustrative guidance for real-world applications with broader impacts.

Tutorial Materials(TBD) Tutors

Chuan Shi, Beijing University of Posts and Telecommunications
 E-mail:shichuan at bupt.edu.cn
 Web-page: www.shichuan.org
Yanfang Ye, Case Western Reserve University
 E-mail:yanfang.ye at case.edu
 Web-page: http://community.wvu.edu/~yaye/
Jiawei Zhang, Florida State University
 E-mail:jiawei@ifmlab.org
 Web-page: http://jiaweizhang.net/
Philip S. Yu, University of Illinois at Chicago
 E-mail:psyu@cs.uic.edu
 Web-page: https://www.cs.uic.edu/~psyu/

Target audience and prerequisites.

The topics of this tutorial cover main research directions of heterogeneous graphs, graph mining, graph neural network and deep learning; and the target audiences are those who are interesting in heterogeneous network mining, representation learning and deep learning from both academia and industry. The expected audiences have background on data mining and machine learning. It is better for them to know information network analysis and graph mining.

Outline of the tutorial

In this tutorial, we will give a survey on recently developed HGNN techniques and their applications. More specifically, the tutorial will include following aspects: (1) basic concepts in HGNNs; (2) research challenges of HGNNs; (3) state-of-the-art HGNN techniques; (4) extensive applications of HGNNs; and (5) frontiers of research in this field. This tutorial will enable researchers and practitioners to share new techniques of HGNNs and give illustrative guidance for real-world applications with broader impacts.

The tutorial will cover the state-of-the-art and key work in this field. We have collected most of work in my webpage: http://www.shichuan.org/HIN_topic.html. We will select some important and representative work to present in the tutorial.