Modeling networks with auxiliary information
- Networks provide a powerful tool for representing social, technological, and biological systems. The study of networks has focused on developing models that either analytically show the emergence of global structural properties in networks or statistically perform inference tasks, such as link prediction, on given network data. This thesis presents a model of networks that not only gives rise to realistic networks with the global structural properties but also permits statistical inference for given network data. Under the proposed model, we prove many structural properties commonly found in real-world networks, such as heavy-tailed degree distributions and small diameters. We also develop a statistical inference algorithm that fits the model to given network data and represents common linking patterns through our model. A network is often associated with auxiliary information, such as node features or temporal information about node and link creations and deletions. This thesis also proposes models of networks with such auxiliary information so that the models capture the relationships between given network links and the auxiliary information. In the second part of the thesis, we propose a model that allows for relationships between given node features and network links. We develop a fitting algorithm that identifies which node features are relevant to network links and how the node features affect the formation of the links. By fitting our model to given network links and node features, our model improves performance for various prediction tasks compared to baseline models. Finally, we develop a model for dynamic networks by considering two notions of dynamics: the birth and death of each group of nodes, and each individual node's behavior of joining or leaving some node groups over time. By embedding these two dynamics into the model, we can achieve the interpretable representation of network dynamics as well as the predictive power of inferring missing links or forecasting future networks.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Stanford University, Department of Electrical Engineering.
|Owen, Art B
|Owen, Art B
|Statement of responsibility
|Submitted to the Department of Electrical Engineering.
|Thesis (Ph.D.)--Stanford University, 2014.
- © 2014 by Myunghwan Kim
Also listed in
Loading usage metrics...