Finding Early Bursting Cohesive Subgraphs in Large Temporal Networks

Big data with timestamps can be constructed as the temporal networks. The bursting cohesive subgraphs (BCS) in temporal networks satisfy to accumulate their cohesiveness at the fastest ratio and could be represented as the emergency events. Yet, existing methods demand BCS to last for at least one given time, which neglects the timeliness of BCS and avoids the events being discovered earlier. In this paper, we design the early bursting cohesive subgraphs (EBCS) model based on kappa-core to enable identifying the events as soon as possible. In particular, we first traverse the temporal network into a node-weighted graph by integrating the topological and historical temporal information. To find the EBCS, (1) we propose a global search algorithm, called GS-EBCS, which can find the exact EBCS by iteratively removing the nodes with the lowest weight and maintaining the topological constraints; (2) we further propose a local search algorithm, called LS-EBCS, which exploits the expand-refine idea. Specifically, it first expands the subgraph from the nodes with the largest weight, and then refines the discovered subgraph to satisfy the constraints of EBCS, whose runtime is proportional to the result subgraph. Extensive experiments are conducted on three real temporal networks, which demonstrate the efficiency and effectiveness of our proposed algorithms.