Exposing influence campaigns in the age of LLMs: a behavioral-based AI approach to detecting state-sponsored trolls

EPJ Data Science

Table 1 Number of trajectories of trolls and users for each trajectory length L

Input	L	Trolls trajectories	Users trajectories
State-Action	50	64,984	102,101
	65	49,975	78,424
	80	40,587	63,643
	100	32,457	50,831
	150	21,628	33,795
	200	16,230	25,299
Action	50	7709	93,721
	65	5927	72,091
	80	4802	58,593
	100	3819	46,863
	150	2547	31,267
	200	1920	23,476