Large-Scale Network Representation Learning Based on Improved Louvain Algorithm and Deep Autoencoder

In recent years, feature learning of nodes in network has become a research hot spot. However, with the growth of the network scale, network structure has become more and more complicated, which makes it extremely difficult for network representation learning in large and complex networks. This paper proposes a fast large-scale network representation learning method based on improved Louvain algorithm and deep autoencoder. First, it quickly folds large and complex network into corresponding small network kernel through effective improved Louvain strategy. Then based on network kernel, a deep autoencoder method is conducted to represent nodes in kernel. Finally, the representations of the original network nodes are obtained by a coarse-to-refining procedure. Extensive experiments show that the proposed method perform well on large and complex real networks and its performance is better than most network representation learning methods.