Beyond Content: Behavioral Policies Reveal Actors in Information Operations

תקציר מקורי באנגליתarXiv:2602.02838v2 Announce Type: replace-cross Abstract: The detection of online influence operations -- coordinated campaigns by malicious actors to spread narratives -- has traditionally depended on content analysis or network features. These approaches are increasingly brittle as generative models produce convincing text, platforms restrict access to behavioral data, and actors migrate to less-regulated spaces. We introduce a platform-agnostic framework that identifies malicious actors from their behavioral policies by modeling user activity as sequential decision processes. We apply this approach to 12,064 Reddit users, including 99 accounts linked to the Russian Internet Research Agency in Reddit's 2017 transparency report, analyzing over 38 million activity steps from 2015-2018. Act