From community detection to topical, interactive group detection in Online Social Networks

Online social networks are prevalent media, either for e-reputation management or for political debate. The presence of malicious actors is often signalled and usually detected by analysts, helped by monitoring tools. These tools still lack of a detection feature for groups of coordinated actors: accounts sharing topical interests, and interacting together, either for recruitment and planning. To tackle this problem, we propose a framework to detect such groups through a community detection approach. We introduce a measure of group topical focus to characterise the groups and distinguish the relevant ones. The value of these contributions is shown on three experiments, targeting datasets issued from three platforms: Reddit, Twitter and Discord, enabling an in-depth group analysis.