Inference in heterogeneous networks

Public Deposited

Last two decades have seen a surge of interests in approaches that leverage network structure in machine learning models. For many networks, not only the connections of the network but also the network attributes, such as node attributes and dyadic attributes, are observed. This heterogeneity in networks raises new challenges for the inference problem in networks. This dissertation discusses how to handle the heterogeneous networks for different ma- chine learning applications, namely community detection, node classification, and node representation learning. For community detection in network with node attributes, we introduce a mathematical approach that combines topology information and nodes at- tributes. The algorithm explores the correlation between node attributes and community assignment, and uses the diversity of dyadic attributes induced by different types of nodes to improve performance as well. We also study node classification problem in a transaction network, where rich information of node and edge is available, within Markov random field framework. We present a novel algorithm that automatically learns node prior and edge potential in the Markov random field, hence results in better classification. Finally, we generalize deepwalk to incorporate the dyadic attributes in network representation learning by biasing the random walk sampling procedure in deepwalk. The algorithm learns the sampling weights in a data driven manner and constructs a proper proximity measure based on the dyadic attributes.

Last modified
  • 05/06/2019
Date created
Resource type
Rights statement