PROJECT TITLE :

GloDyNE: Global Topology Preserving Dynamic Network Embedding

ABSTRACT:

Due to the time-evolving nature of many real-world networks, learning low-dimensional topological representations of networks in dynamic environments is attracting a lot of attention. This is because of the nature of the networks themselves. The primary and overarching goal of Dynamic Network Embedding (DNE) is to update node embeddings in an effective manner while simultaneously maintaining the topology of the network at each time step. The goal of the vast majority of the currently available DNE methods is to update the node embeddings in such a way as to take into account the topological changes that have occurred at or near the nodes that have been most significantly impacted. Unfortunately, although this type of approximation can improve efficiency, it is unable to effectively preserve the global topology of a dynamic network at each time step. This is because it does not take into consideration the inactive sub-networks that receive accumulated topological changes that are propagated via the high-order proximity. In order to overcome this obstacle, we have developed a novel node selecting strategy that can diversely select the representative nodes across a network. This strategy is coordinated with a new incremental learning paradigm that is based on an embedding approach that utilizes skip-grams. Extensive testing has shown that GloDyNE can already achieve performance that is superior to or comparable to that of the most advanced DNE methods in three different types of downstream tasks. This was determined by comparing the results of GloDyNE with those of the most advanced DNE methods. In particular, GloDyNE performs significantly better than other methods in the graph reconstruction task, which demonstrates its capacity for maintaining global topology.