°µÍø½ûÇø

Detecting change points in the large-scale structure of evolving networks Ìý

,Ìý

Date and time:Ìý

Thursday, March 20, 2014 - 2:00pm

³¢´Ç³¦²¹³Ù¾±´Ç²Ô:Ìý

ECCR 257

´¡²ú²õ³Ù°ù²¹³¦³Ù:Ìý

Networks are an important tool for describing and quantifying data on interactions among objects or people, e.g., online social networks, offline friendship networks, and object-user interaction networks, among others. When interactions are dynamic, their evolving pattern can be represented as a sequence of networks each giving the interactions among a common set of vertices at consecutive points in time. An important task in analyzing such evolving networks, and for predicting their future evolution, is change-point detection, in which we identify moments in time across which the large-scale pattern of interactions changes fundamentally. Here, we formalize the network change point detection problem within a probabilistic framework and introduce a method that can reliably solve it in data. This method combines a generalized hierarchical random graph model with a generalized likelihood ratio test to quantitatively determine if, when, and precisely how a change point has occurred. Using synthetic data with known structure, we characterize the difficulty of detecting change points of different types, e.g., groups merging, splitting, forming or fragmenting, and show that this method is more accurate than several alternatives. Applied to two high-resolution evolving social networks, this method identifies a sequence of change points that align with known external shocks to these networks.