查看文章

Multiple networks emerge in a wealth of high-impact applications. Network alignment, which aims to find the node correspondence across different networks, plays a fundamental role for many data mining tasks. Most of the existing methods can be divided into two categories: (1) consistency optimization based methods, which often explicitly assume the alignment to be consistent in terms of neighborhood topology and attribute across networks, and (2) network embedding based methods which learn low-dimensional node embedding vectors to infer alignment. In this paper, by analyzing representative methods of these two categories, we show that (1) the consistency optimization based methods are essentially specific random walk propagations from anchor links that might be too restrictive; (2) the embedding based methods no longer explicitly assume alignment consistency but inevitably suffer from the space …