Burstiness Analysis of Higher-Order Interactions in Document Streams

Abstract

There are various real systems in whichevent sequences display such bursty propertiesthat are characterized by intermittent switching between short periods of intense activity andlong periods of no or reduced activity.Examples includeoutgoing mobile phone calls of an individual, neuron spike trains, and earthquakes at a single location.Traditional analyses of temporal networks have mostly dealt with pairwise interactions.Recently, by exploring several temporal social networks approximating face-to-face communications of people,researchers have observedthe existence of burstiness in event sequencesfor higher-order human interactions as well as pairwise ones, andfound a temporal correlation between two consecutive events.In this paper,using Japanese newspaper data,we investigate the burstiness ofevent sequences for higher-order interactionsin a temporal network of topic wordsderived by their co-occurrences in a document stream.By analyzing the number of events in a train for each size of interaction,we show the existence of temporal correlation between two consecutive events.Moreover, we reveal the properties of the event trains for higher-order linksin terms of burstiness degree and memory coefficient.