[Air-L] Toolkit for collecting instant messaging data and running controlled experiments
Gregory Mills
g.j.mills at rug.nl
Mon Nov 23 05:32:36 PST 2020
Dear AoIR colleagues,
We have just released a toolkit for collecting naturalistic instant
messaging data, as well as running controlled experiments on
algorithmically mediated conversation.
The toolkit runs as a bot on the Telegram network, which participants
connect to using the Telegram app on their mobile phones. Since all
messages between participants are mediated via the toolkit, data is
instantly available for analysis - there is no need for participants to
send in their data. The toolkit allows high levels of experimental
control: Participants can be dynamically assigned to different groups,
and experimental instructions (e.g. images, texts, questionnaires) can
be sent automatically by the toolkit to the participants in their chat
window.
The toolkit allows high-levels of experimental control over the content
and timing of participants’ turns. Turns can be selectively reordered,
or blocked, e.g. to investigate the effects of shadowbanning. The
toolkit also allows individual words, phrases, emoticons, etc. to be
filtered from participants' turns. This technique can also be used to
insert "spoof" turns into the dialogue, that appear, to participants to
originate from each other - allowing direct testing of hypotheses about
how participants respond to specific prompts.
In addition, the toolkit provides a set of customizable, (non-Telegram )
chat interfaces for investigating how different interfaces affect online
conversations.
The toolkit is open-source and is downloadable at:
http://cogsci.eecs.qmul.ac.uk/diet/
Best,
Gregory Mills
g.j.mills at rug.nl
More information about the Air-L
mailing list