Global Security and Intelligence Studies Volume 5, Number 1, Spring / Summer 2020 | Page 60

Global Security and Intelligence Studies
Figure 1: Example of Stream + Chat (Sample: Mibrtv, 3/14/2020).
To-date, no study has been conducted
to discover IFOs on Twitch. We
predicted that IFO actors would employs
bots to achieve their objectives,
so we built and executed a data-mining
script in Python to identify users who
post in a bot-like manner (defined in
the methods). We further analyzed approximately
100 users who met our bot/
bot-like criteria to determine whether
or not they were actual bots supporting
an IFO.
Methods
Previous researchers have built
bot-detection programs utilizing
multiple methods, such as decorate
classification (Lee et al. 2010), Naïve
Bayes (Wang 2010), Jrip (Ahmed and
Abulaish 2013), Random Forest (Chu
e. al. 2012), contrast patterns (Loyola-
González et al. 2019), and Botometer
(Yang et al. 2019). Because the Twitch
platform includes a separate commenting
interface from traditional social
media sites, such as Facebook and Twitter,
we coded a simple bot classification
tool to serve as a starting place for more
advanced bot researchers. But first, we
make a number of assumptions about
how an IFO might be conducted over
Twitch:
1. IFO actors prefer to automate their
operations.
2. Even if IFO actors have the resources
to target all twitch streams, to do
so would be overly conspicuous and
therefore counterproductive.
3. IFO actors selectively target the
streams they attempt to influence.
4. IFO actors do not target individuals
on the platform, and instead target
the largest number of users possible.
5. IFO actors do not target empty
streams.
46