Global Security and Intelligence Studies Volume 5, Number 1, Spring / Summer 2020 | Page 57
Global Security and Intelligence Studies • Volume 5, Number 1 • Spring / Summer 2020
Discovering Influence Operations on Twitch.tv:
A Preliminary Coding Framework
Alexander Sferrella and Joseph Z. Conger
Abstract
Bots are an important tool for influence actors, and greatly contribute
to the complexity and breadth of influence operations (IFOs) across
many platforms. Twitch.tv—the second-most popular streaming
site—is one such platform. Recognizing that influence actors may
expand operations within Twitch, the following study develops a
framework that mines data from the Twitch platform to identify potential
bots running IFOs. Stream comments from 14 Twitch channels
were run through a custom Python script. We identified 69 of
128 streams, from 12 channels, as having an anomalous comment
count OR comment speed. Of those streams, we identified 7,332 users
as having an anomalous comment count AND comment speed.
However, we could not distinguish 100 randomly selected anomalous
users as bots or humans after a manual analysis. Overall, our research
provides future researchers with a modular method to collect
and isolate Twitch data containing bots.
Keywords: influence operation, influence actor, social media,
streaming, Twitch, bot, psychological domain, sixth domain
Descubriendo las operaciones de influencia en Twitch.tv:
un marco preliminar de coding
Abstract
Los bots son una herramienta importante para los actores de influencia
y contribuyen en gran medida a la complejidad y amplitud
de las operaciones de influencia en muchas plataformas. Twitch.tv,
el segundo sitio de transmisión más popular, es una de esas plataformas.
Reconociendo que los actores de influencia pueden expandir
las operaciones dentro de Twitch, el siguiente estudio desarrolla un
marco que extrae datos de la plataforma Twitch para identificar posibles
bots que ejecutan operaciones de influencia. Los comentarios
de flujo de 14 canales de Twitch se ejecutaron a través de un script
Python personalizado. Identificamos 69 de 128 transmisiones, de 12
43
doi: 10.18278/gsis.5.1.4