Task-driven sampling of attributed networks

Œis paper introduces new techniques for sampling aŠributed networks to support standard Data Mining tasks. Œe problem is important for two reasons. First, it is commonplace to perform data mining tasks such as clustering and classi€cation of network aŠributes (aŠributes of the nodes, including social media posts). Furthermore, the extraordinarily large size of real-world networks necessitates that we work with a smaller graph sample. Second, while random sampling will provide an unbiased estimate of content, random access is o‰en unavailable for many networks. Hence, network samplers such as Snowball sampling, Forest Fire, Random Walk, Metropolis-Hastings Random Walk are widely used; however, these aŠribute-agnostic samplers were designed to capture salient properties of network structure, not node content. Œe laŠer is critical for clustering and classi€cation tasks. Œere are three contributions of this paper. First, we introduce several aŠribute-aware samplers based on Information Œeoretic principles. Second, we prove that these samplers have a bias towards capturing new content, and are equivalent to uniform sampling in the limit. Finally, our experimental results over large real-world datasets and synthetic benchmarks are insightful: aŠribute-aware samplers outperform both random sampling and baseline aŠribute-agnostic samplers by a wide margin in clustering and classi€cation tasks.