Unsupervised Social Bot Detection via Structural Information Theory.

The article focuses on UnDBot, a novel unsupervised research framework for detecting social bots, designed to be interpretable and effective in identifying complex bot behaviors. Topics include developing social relationship metrics like posting influence, follow-to-follower ratio, and posting type distribution, constructing a weighted social multi-relational graph to capture user behavior similarities, and optimizing heterogeneous structural entropy to enhance bot detection accuracy.