A Neural Method of Finding Frequent Subgraph Patterns

Abstract

MS Defense:

Frequent subgraph pattern mining is an important problem in network analysis. Counting subgraph patterns is computational expensive and often depends on approximate algorithms. Recently, approaches based on machine learning are proposed to solve this problem. This work introduces a new approach that leverages the power of graph neural network to find frequent subgraph patterns. Using embeddings of graph structures in the linear space, our approach converts the discrete pattern-searching problem to a continuous optimization problem, which can find frequent subgraph patterns efficiently. Furthermore, its running time is only quadratic in the size of graph patterns. Experiment results show promising result of this new approach.