[Air-L] Twitter Chatbots

Fri Sep 27 15:40:55 PDT 2019

I would second everything Dr Bolsover just wrote and add this. We are
currently testing new methods for detecting automation, bots, and trolls in
the Canadian election. This is a time-sensitive project. If you are fluent
in Canadian politics, interested in bot detection, and worried about the
future of democracy, please drop me a line. We have a promising new
approach, but our goal is to merge many approaches into a more unified
framework.

~stu

On Fri, Sep 27, 2019 at 3:55 PM Gillian Bolsover <G.Bolsover at leeds.ac.uk>
wrote:

> Hi Natalie (and others),
>
> As a knee-jerk reaction, I'd say this was too much for a student project
> unless there was significant prior knowledge and training in relevant
> quantitative research methods or time to acquire those or an existing
> dataset in mind that could be worked with. But a few ideas briefly.
>
> 1 - A chatbots and bots are not the same thing. I haven't seem much about
> chatbots used in politics apart from, if I remember correctly, a Le Keqiang
> WeChat chatbot developed around the 2017 National Congress, which I think
> was more humorous/cute than influencing.
>
> 2 - Bot identification: The easiest thing to do is to plug stuff into the
> BotorNot (Botometer) api (check out any of the publications by the group
> who developed this too) but also being aware that people use lots of
> different methods and no method is perfect. I've seen things as simple as
> just per day post volume used to decide "likely automation." I always check
> the post source in metadata (if you are working directly with API derived
> posts) as this way that produces no false positives but likely lots of
> false negatives. In the recent stuff over pro-China automation surrounding
> Hong Kong on Twitter, a sudden switch in language and content from English
> content about sports to Chinese content about Hong Kong was seen as a
> predictor. BotorNot is probably sufficient for a student project unless the
> main point is to develop new means of bot identification.
>
> 3 - I'd think carefully about using network analysis to assess the
> influence of bots. Bots tend to work in networks reposting each others
> content having mutual friendship connections. Indeed, this is one of the
> best ways of really identifying them is similarity and similarity in
> practice across a large group of accounts. Thus, there is the potential for
> circularity in using network analysis metrics of influence to assess bots
> that are programmed to work in a networked way. This isn't to say that it
> isn't a useful analysis but there is an extra level of complication to
> interpreting the metrics when working with bots as opposed to just human
> networks because bots work in networks.
>
>
> Dr Gillian Bolsover
> Lecturer in Politics and Media<
> https://essl.leeds.ac.uk/politics/staff/693/dr-gillian-bolsover>
> University of Leeds
>
> Support the technologies of the future you want; end-to-end encrypt your
> emails.
> PGP Key: 17EC60B3<
> https://pgp.mit.edu/pks/lookup?op=vindex&search=0x15E8761217EC60B3>
>
> Latest Article:
> C. Goron and G. Bolsover. Engagement or control? The impact of the Chinese
> environmental protection bureaus’ burgeoning online presence in local
> environmental governance.
> <https://www.tandfonline.com/doi/full/10.1080/09640568.2019.1628716>
> <https://www.tandfonline.com/doi/full/10.1080/09640568.2019.1628716>
> On 27/09/2019 19:57, Marcela Canavarro wrote:
>
> This one is not about bots but it may bring good clues on network strength,
> using a few metrics available on gephi:
>
> https://www.tandfonline.com/doi/abs/10.1080/1369118X.2015.1043315
>
>
>
> On Friday, September 27, 2019, Mark Chen <markchen at u.washington.edu
> ><mailto:markchen at u.washington.edu> wrote:
>
>
>
> Hi,
>
> It's been in the news a bit over the last couple of years so that might be
> a good place to start.
> https://www.buzzfeed.com/alexspence/nigel-farages-
> brexit-party-twitter-following
>
> https://journals.sagepub.com/doi/abs/10.1177/0894439317734157?journalCode=
> ssce
>
> mark
>
> On Fri, Sep 27, 2019 at 6:54 AM Natalie Rock <drnatalierock at gmail.com
> ><mailto:drnatalierock at gmail.com>
> wrote:
>
>
>
> This one is out of my wheelhouse but maybe you know!I have a student who
> wants to investigate the impact Twitter chatbots have on framing
>
>
> political
>
>
> conversations, driving specific discussions, and determining voter
> intentions. His focus is the Brexit referendum.
>
> He wants to first classify tweets as coming from bots and then measure
>
>
> the
>
>
> impact of the tweets using three elements: strength of network
>
>
> connection:
>
>
> how important the influencing group of people or their status are to you;
> Immediacy of network connection: how close the group are to you (physical
> distance and time) at the time of the influence attempt; number: How many
> people are present in the environment.
>
> I can foresee some challenges but like I say, this type of analysis is
> very much outside of my specialism. This is an ambitious student so
>
>
> before
>
>
> I recommend an alternative approach: Is there a strategy anyone can
> recommend that will allow him to correctly classify tweets as coming from
> bots? How can he measure strength and immediacy of network connection?
>
>
> Can
>
>
> anyone recommend any publications for him?
>
> Thank y’all!
>
> Natalie
>
> Sent from my iPhone
> _______________________________________________
> The Air-L at listserv.aoir.org<mailto:Air-L at listserv.aoir.org> mailing list
> is provided by the Association of Internet Researchers http://aoir.org
> Subscribe, change options or unsubscribe at:
> http://listserv.aoir.org/listinfo.cgi/air-l-aoir.org
>
> Join the Association of Internet Researchers:
> http://www.aoir.org/
>
>
>
>
>
> --
> You see before you *Mark Chen, PhD*.
> Above his head appears a label that changes every time you look at it
> between "*Hoodie-Wearing Games Scholar Thug*," "*PT Lecturer at UW
> Bothell*," and
> "*A very happy young girl looking forward to a bright and wonderful
> future.*
> "
> Do you send him a tweet (*@mcdanger* <http://twitter.com/mcdanger><
> http://twitter.com/mcdanger>), check
> out his website (*markdangerchen.net* <http://markdangerchen.net/><
> http://markdangerchen.net/>), or
> respond to this email?
> His desk and surroundings are on fire as he smiles and says, "*everything
> is fine*."
> _______________________________________________
> The Air-L at listserv.aoir.org<mailto:Air-L at listserv.aoir.org> mailing list
> is provided by the Association of Internet Researchers http://aoir.org
> Subscribe, change options or unsubscribe at: http://listserv.aoir.org/
> listinfo.cgi/air-l-aoir.org
>
> Join the Association of Internet Researchers:
> http://www.aoir.org/
>
>
> _______________________________________________
> The Air-L at listserv.aoir.org<mailto:Air-L at listserv.aoir.org> mailing list
> is provided by the Association of Internet Researchers http://aoir.org
> Subscribe, change options or unsubscribe at:
> http://listserv.aoir.org/listinfo.cgi/air-l-aoir.org
>
> Join the Association of Internet Researchers:
> http://www.aoir.org/
> _______________________________________________
> The Air-L at listserv.aoir.org mailing list
> is provided by the Association of Internet Researchers http://aoir.org
> Subscribe, change options or unsubscribe at:
> http://listserv.aoir.org/listinfo.cgi/air-l-aoir.org
>
> Join the Association of Internet Researchers:
> http://www.aoir.org/

-- 
Dr. Stuart W. Shulman
Founder and CEO, Texifter
Cell: 413-992-8513
LinkedIn: http://www.linkedin.com/in/stuartwshulman