Designing Machine Learning Models for Graph Analytics

University of Technology Sydney. Faculty of Engineering and Information Technology.With growing popularity of the machine learning methods, there have been a great number of machine learning methods proposed for graph analytics. In this thesis, we design three machine learning based models for the popular graph analysis tasks such as node classification, graph interaction prediction and subgraph matching. Firstly, we design a binarized graph neural network to efficiently obtain the vector representations for vertices and graphs. Recently, there have been some breakthroughs in graph analysis by applying the Graph Neural Networks (GNNs). However, the parameters of the network and the embedding of nodes are represented in real-valued matrices in existing GNN-based approaches which may limit the efficiency and scalability of these models. This motivates us to develop a binarized graph neural network to learn the binary representations of the nodes with binary network parameters following the GNN-based paradigm. Our proposed method can be seamlessly integrated into the existing GNN-based embedding approaches to binarize the model parameters and learn the compact embedding. Secondly, we design a graph of graphs neural network for entity interaction prediction, and then extend the model to support the graph classification task with more expressive representations. Entity interaction prediction is essential in many important applications, which can be quite challenging when there are two types of graphs are involved: local graphs for structured entities and a global graph for the interactions between structured entities. We observe that existing works cannot properly exploit the unique graph of graphs structure. In this thesis, we propose a Graph of Graphs Neural Network, namely GoGNN, which extracts the features of the given graph in a hierarchical way. Based on GoGNN, we further propose a Powerful Graph Of graphs neural Network, namely PGON, which has 3-Weisfeiler-Lehman expressive power and can be used to handle the graph classification task. Thirdly, we design a reinforcement learning based query vertex ordering model for subgraph matching. Subgraph matching is a fundamental problem in graph analytics. Instead generating the matching order with heuristics, our model could capture and make full use of the graph information, and thus determine the query vertex order with the adaptive learning-based rule that could significantly reduce the number of redundant enumerations. With the help of the reinforcement learning framework, our model could consider the long-term benefits during order generation. Extensive experiments on real-life datasets indicate the efficiency and effectiveness of our proposed models in the corresponding graph analytic tasks