Motif Prediction with Graph Neural Networks

Explore a conference talk from the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD'22) that delves into motif prediction using graph neural networks. Discover how traditional link prediction methods fall short in effectively predicting motifs and learn about a novel approach to address this limitation. Gain insights into the general motif prediction problem and various heuristics developed to assess the likelihood of specific motifs appearing. Understand the importance of considering link correlations in motif prediction. Examine a cutting-edge graph neural network (GNN) architecture designed specifically for motif prediction, featuring vertex features and sampling schemes that capture complex structural properties. Compare the performance of the proposed GNN approach with existing methods, noting significant improvements in prediction accuracy for both dense and sparse motifs. Explore the potential applications of this architecture beyond motif analysis, including predicting clusters and communities in graph mining.

Introduction: Link Prediction
Problem Formulation: Motif Prediction
Score Function & Heuristics
Graph Neural Networks for Motif Prediction
The SEAM Architecture
Evaluation