Random-walk–based methods are widely used for network embedding and downstream learning tasks, yet it remains unclear which aspects of network topology they capture and which they ignore. This is a ...