Application of Deep Walk Based on Hyperbolic Coordinates on Unsupervised Clustering

摘要

In the real world, various information can be represented by graph structure data. For example, interpersonal relationships and protein structure. In recent years, with the development of artificial intelligence, graph embedding has become a popular method of network analysis. It can reduce the dimension of network structure data, so that network structure data can be applied to various machine learning and deep learning tasks. At the same time, many studies of network geometry show that the hidden metric of many complex networks is hyperbolic. After hyperbolic space mapping, nodes in the original network data structure can be represented by hyperbolic coordinates. Hyperbolic coordinates contain information about the popularity and similarity of nodes which is very important for unsupervised clustering tasks. However, the random walk strategy in the native Deep Walk algorithm cannot effectively extract this information. So we propose an improvement of the Deep Walk algorithm based on hyperbolic coordinates and achieved good results on many datasets.