An approach of Bursty event detection in social networks based on topological features

User relations and information propagation on social networks can reflect events in real society. Online detection of bursty events is of great significance to studying the evolution of social networks and cyberspace security. Current research works focus on building an event recognition model based on text information and then utilizing clustering or topic model methods to extract features from the data stream and then detect events that have not existed before. The text-based method is designed for specific content, and it does not consider the features of network dynamic evolution. It is restricted by the type and quality of text in social networks, limiting its practical application scenarios. However, there exist remarkable correlations between the occurrence of events and the evolution of the network. In this paper, we consider mining the network structure changes to identify bursty events, the superiority which is that it is sensitive and widely used. We integrate snapshot network topology indexes to quantify its structural features. Then we can judge whether there is a burst event by investigating the change degree of the network structure features of the adjacent snapshots. The effectiveness and efficiency of our approach are further confirmed by experimental studies on four real social network data sets. In addition, we also discuss the salient features of the bursty events and compare the impact of the bursty events on the network structure with that of the scheduled events.

Automated summary

This paper investigates using changes in network structure to identify bursty events, which is more sensitive and widely applicable compared to text-based methods.