DeepWalk

Abstract

We present DeepWalk, a novel approach for learning latent representations of\nvertices in a network. These latent representations encode social relations in\na continuous vector space, which is easily exploited by statistical models.\nDeepWalk generalizes recent advancements in language modeling and unsupervised\nfeature learning (or deep learning) from sequences of words to graphs. DeepWalk\nuses local information obtained from truncated random walks to learn latent\nrepresentations by treating walks as the equivalent of sentences. We\ndemonstrate DeepWalk's latent representations on several multi-label network\nclassification tasks for social networks such as BlogCatalog, Flickr, and\nYouTube. Our results show that DeepWalk outperforms challenging baselines which\nare allowed a global view of the network, especially in the presence of missing\ninformation. DeepWalk's representations can provide $F_1$ scores up to 10%\nhigher than competing methods when labeled data is sparse. In some experiments,\nDeepWalk's representations are able to outperform all baseline methods while\nusing 60% less training data. DeepWalk is also scalable. It is an online\nlearning algorithm which builds useful incremental results, and is trivially\nparallelizable. These qualities make it suitable for a broad class of real\nworld applications such as network classification, and anomaly detection.\n

Affiliated Institutions

• Stony Brook University US

Related Publications