Abstract

Network analysis is an important and emerging field, where many challenges remain. One important challenge is the monitoring of large dynamic networks for identifying changes in the behavior of nodes in the network. These methodologies have a wide application area from social networks, to finance and biology. Recently many methods have been proposed for monitoring dynamic networks, however, most of these methods have strict assumptions and/or can be applied to relatively small networks only (hundreds of nodes). In this work, we propose a new methodology that is scalable to networks of 10.000+ nodes. In addition, the methodology allows for time-varying network structure by including an autoregressive model. Thereby allowing the network to slowly change over time. We implement our proposed method on a case study with Reddit data where we detect changes in the subreddit WallStreetBets before the stock price frenzy with GameStop in January 2021. We also evaluate the performance of our proposed method on a variety of simulated network structures and show that it is scalable.