Anomaly detection in dynamic social networks for identifying key events

Abstract

Finding the most relevant facts and the relations between each of them is not a trivial task due to vast amount of information in the Internet. Different significant events influence the World Wide Web and the blogosphere and because of its size and variety we are often not aware that such events take or took place. The identification of significant changes of the blogosphere may inform us about their occurrences. We define a state of social portal taking into consideration general network features, measures of key elements and distribution of these measures, neighbourhood distributions of nodes and existing communities, and analyse the changes of these factors in the subsequent network states to identify anomalies, possibly caused by significant events. Two portals (Polish Salon24 blog portal and Huffington Post) are used as cases in the evaluation part.