WalkingTime: Dynamic Graph Embedding Using Temporal-Topological Flows

TL;DR

WalkingTime introduces a novel dynamic graph embedding method that models local temporal-topological flows without discretizing time, enabling more accurate representation of continuously evolving networks.

Contribution

It presents a fundamentally different approach to dynamic embedding by focusing on local flows and continuous time, avoiding the limitations of global time-step discretization.

Findings

Effective modeling of continuous-time network evolution

Improved embedding quality over traditional methods

Applicable to streaming and asynchronous networks

Abstract

Increased attention has been paid over the last four years to dynamic network embedding. Existing dynamic embedding methods, however, consider the problem as limited to the evolution of a topology over a sequence of global, discrete states. We propose a novel embedding algorithm, WalkingTime, based on a fundamentally different handling of time, allowing for the local consideration of continuously occurring phenomena; while others consider global time-steps to be first-order citizens of the dynamic environment, we hold flows comprised of temporally and topologically local interactions as our primitives, without any discretization or alignment of time-related attributes being necessary. Keywords: dynamic networks , representation learning , dynamic graph embedding , time-respecting paths , temporal-topological flows , temporal random walks , temporal networks , real-attributed knowledge graphs , streaming graphs , online networks , asynchronous graphs , asynchronous networks , graph algorithms , deep learning , network analysis , datamining , network science