A1363
Title: CITE-ME: Controlling for induced triangles in estimating network model evolution
Authors: Benjamin Leinwand - Stevens Institute of Technology (United States) [presenting]
Keith Levin - University of Wisconsin (United States)
Abstract: In the standard degree corrected block model (DCBM), once the parameters are fixed, edges are conditionally independent of one another. This ignores effects like triadic closure observed in real networks. Conversely, many network models that include conditional edge dependence do not explicitly incorporate both degree correction features for each node and community structure. The purpose is to discuss a network generative model whereby edges are sampled one at a time without replacement. The initial sampling weights of the edges depend on the underlying affinities between the incident nodes absent any triadic closure, which may follow the structure of the DCBM probability matrix. As the sampling continues, though, edge sampling weights are updated to promote triadic closure within the network. The details of this generative mechanism and its effects on the resulting network, including distorting community structures, are discussed.
