TDAE: An Approach for Predicting Communities on Dynamic Network Based on Deep Auto-Encoder

Community testing is an important research content in network analysis. But most of the work usually deals with information in static networks and does not reflect the dynamic evolution of the network over time. This article focuses on the issue of combining network links, node attributes, and time information for community detection and prediction in an evolving network. We propose a TDAE (Temporal Deep Auto-encoder) based on deep auto-encoder. The method considers the network structure and node attributes simultaneously, embeds the continuity of time on the basis of the deep auto-encoder framework to extract the potential representation of the network nodes, and then uses the clustering method to predict the future community in the potential representation space of the node. The experimental results show that the TDAE method is superior to the current mainstream methods in the similarity coefficient and purity coefficient.