Abstract:We propose a local-to-global strategy for graph machine learning and network analysis by defining certain local features and vector representations of nodes and then using them to learn globally defined metrics and properties of the nodes by means of deep neural networks. By extending the notion of the degree of a node via Breath-First Search, a general family of {\bf parametric centrality functions} is defined which are able to reveal the importance of nodes. We introduce the {\bf neighbors degree frequency (NDF)}, as a locally defined embedding of nodes of undirected graphs into euclidean spaces. This gives rise to a vectorized labeling of nodes which encodes the structure of local neighborhoods of nodes and can be used for graph isomorphism testing. We add flexibility to our construction so that it can handle dynamic graphs as well. Afterwards, the Breadth-First Search is used to extend NDF vector representations into two different matrix representations of nodes which contain higher order information about the neighborhoods of nodes. Our matrix representations of nodes provide us with a new way of visualizing the shape of the neighborhood of a node. Furthermore, we use these matrix representations to obtain feature vectors, which are suitable for typical deep learning algorithms. To demonstrate these node embeddings actually contain some information about the nodes, in a series of examples, we show that PageRank and closeness centrality can be learned by applying deep learning to these local features. Our constructions are flexible enough to handle evolving graphs. Finally, we explain how to adapt our constructions for directed graphs.

Integrating Local & Global Features for Estimating Shortest-path Distance in Large-scale Graphs

Detecting User Community in Sparse Domain via Cross-Graph Pairwise Learning

Learn Low-dimensional Shortest-path Representation of Large-scale and Complex Graphs

Scalable Graph Embedding for Asymmetric Proximity.

A Learning Based Approach to Predict Shortest-Path Distances.

Exploiting Edge Features in Graphs with Fused Network Gromov-Wasserstein Distance

Asynchronous Collaborative Localization by Integrating Spatiotemporal Graph Learning with Model-Based Estimation

Long-range Meta-path Search on Large-scale Heterogeneous Graphs

Spatio-Temporal Dual Graph Neural Networks for Travel Time Estimation

Hub-Accelerator: Fast and Exact Shortest Path Computation in Large Social Networks

Synergistic Integration of Local and Global Information for Critical Edge Identification

Spatial Origin-Destination Flow Imputation Using Graph Convolutional Networks

Recovering Missing Node Features with Local Structure-based Embeddings

Enhancing the Network Embedding Quality with Structural Similarity

An Efficient Shortest Path Approach for Social Networks Based on Community Structure

L2G2G: a Scalable Local-to-Global Network Embedding with Graph Autoencoders

Preserving Local and Global Information for Network Embedding

Network Distance Prediction for Enabling Service-Oriented Applications over Large-Scale Networks

Location and time embedded feature representation for spatiotemporal traffic prediction

Learning to Match Features with Discriminative Sparse Graphneuralnetwork

Local Graph Embeddings Based on Neighbors Degree Frequency of Nodes