|
Abstract:
|
Community structure in a network is commonly influenced by homophily, the phenomenon that interactions are more likely to occur between nodes with similar attributes. Yet, generative models of community structure based only on the network topology may miss community structure determined by node attribute relations, while models that strictly enforce homophilous communities will be too restrictive. To model communities that may exhibit homophily in observed networks, we propose a generative model, the "Community and Node Attribute-Corrected Stochastic Blockmodel" (canacSBM), which preserves the joint structure of observed node attribute interactions and estimated community structure. We first introduce a flexible non-parametric method for inferring community structure, resulting in a data object we call the "stable structure matrix". Then, conditional on the estimated communities, we allow heterogeneity in community-level attribute dependencies. Finally, we apply canacSBM to a real-world network that represents a national security application and observe how the model appropriately recognizes the joint node attribute and community structure.
|
