IISD: Integrated Interaction Subgraph Detection for event mining

Abstract

Event detection on networks is an important research task in data mining. Most previous methods usually detect a particular type of event that satisfies the predefined rules in the model. However, few methods consider human expert’s interests during the detection process to discover unexplored events. In this paper, we regard the interactive event detection as the anomalous subgraph detection on attributed networks, named Integrated Interaction Subgraph Detection (IISD), where events are treated as anomalous connected subgraphs on the network. The core of our method is automatically identifying events by evaluating the abnormality of the subgraph and integrating the human expert’s interaction simultaneously. Specifically, we first define the human expert’s interaction and recommended interaction domain (i.e., the subgraph and neighbor vertices), which are used to conduct interactive operations based on the human expert’s interests. Afterwards, we propose an efficient subgraph detection algorithm that iteratively integrates human expert’s feedback and updates the recommended interaction domain. In this way, our method could retrieve the most anomalous subgraph on the network as the final event, which could contain the potential unexplored information due to the continuously optimized interaction of human experts. We have conducted extensive experiments on two real-world datasets and proved that our algorithm could achieve better performance compared with several competitive baselines. Moreover, the case study shows that our method could detect global abnormal events effectively.