Networks provide a model of relationships between objects, which can represent numerous invisible services for humans every day. These range from connecting users on social media to controlling how power is distributed throughout a city. We are interested in extending the notion of centrality (mean, median, etc.) to a network. Multiple centrality measurements are introduced, including degree, closeness, betweenness, and eigenvector. Outlier nodes are useful for observing a network's structure. In a university Facebook network, for example, a user with a low centrality measurement may be disconnected from campus, a cause of concern for university faculty. This is a low outlier. On the other hand, users with a high centrality measurement will be influential and thus useful for spreading information to the greater student population. This is a high outlier. The nodes in the network data researched for this presentation have had their centrality measurements analyzed using the interquartile range (IQR) to determine if the node is an outlier.
Hamalis, Michael, "The Network Outlier Hunt" (2021). Symposium on Undergraduate Research and Creative Expression (SOURCE). 926.