A Survey of Dynamic Network Embedding based on Discrete Snapshots

The real-world is composed of various complex systems, and the systems can be represented by networks that evolve over time, which are called dynamic networks by scholars. Earlier research focused on static networks, but in recent years, greater progress has been made in the analysis of dynamic networks, especially with respect to dynamic network embedding. In this paper, a systematic study of dynamic network embedding based on discrete snapshot method is conducted. We first introduce some basic concepts related to dynamic network embedding, summarize the existing dynamic network embedding techniques and classify them, including random walk-based, matrix factorized-based and neural network-based methods, and introduce several neural network-based methods in detail. In addition, we summarize the commonly downstream tasks, present some challenges faced by existing models, and point out some directions for future research.