WalDis: Mining Discriminative Patterns within Dynamic Graphs

Abstract

Real-world networks typically evolve through time, which means there are various events occurring, such as edge additions or attribute changes. In order to understand the events, one must be able to discriminate between different events. Existing approaches typically discriminate whole graphs, which are, in addition, mostly static. We propose a new algorithm WalDis for mining discriminate patterns of events in dynamic graphs. This algorithm uses sampling by random walks and greedy approaches in order to keep the performance high. Furthermore, it does not require the time to be discretized as other algorithms commonly do. We have evaluated the algorithm on three real-world graph datasets.