Provide a spam score (or rather a “good citizen” flag)
Dave-twiends
blocking spam on my site for a long time but this gives me an option
to report it now, and hopefully get these account suspended quicker so
that they don’t come back.
It would be really great if we could have a proactive api function as
well, where we could get the likelihood of a user being a spammer. I
know this is really difficult to do, and wrought with pitfalls, but
perhaps it could be structured in some way? Maybe you could provide a
“good citizen” flag for a user (i.e a user that hasn’t had any
complaints, and has a certain account age). That way you don’t
negatively impact any users, but we can then at least treat these
users differently when they sign up with our site.
I’m seeing more and more the need to provide different limits to users
based on private trust/karma score we develop for each user. This
would be a very valuable input for us to detect potential problem
users before they can cause damage.
Thanks
Dave
Justin
I've found sorting out bots and pure spammers to be very difficult.
Some folks tweet so much they resemble bots/spam.
Feels like a pipe dream but if they can I'd love a karma scoring
system directly from the API.
M. Edward (Ed) Borasky
and so does Klout. For that matter, Viralheat has an API and they can
get both Twitalyzer and Klout scores.
That said, I don't know that there's ever really going to be a "one
size fits all" Twitter user metric. But there are quite a few
"crowdsourcing" and "curation" tools starting to show up, some of them
open source. But personally, I think it's more fun to just collect raw
data via the API and roll your own. ;-)
--
M. Edward (Ed) Borasky
http://borasky-research.net http://twitter.com/znmeb
"A mathematician is a device for turning coffee into theorems." - Paul Erdos
Quoting Justin <justin....@gmail.com>:
> --
> Twitter developer documentation and resources: http://dev.twitter.com/doc
> API updates via Twitter: http://twitter.com/twitterapi
> Issues/Enhancements Tracker: http://code.google.com/p/twitter-api/issues/list
> Change your membership to this group:
> http://groups.google.com/group/twitter-development-talk
>
Dave-twiends
to provide "good citizen" flag for us.
So, they shouldn't provide any kind of score to the outside world via
the web or API, but they could use it to provide an "indicator" for a
clean account.
In other words, we would assume all accounts to be potential spammers
until we see the "good citizen" flag on the account, then we can be a
little more trusting of that account. Not completely trusting, but
just use it as an indicator..
On Oct 17, 6:43 am, "M. Edward (Ed) Borasky" <zn...@borasky-
> I don't know about a "karma" score, but Twitalyzer does have an API
> and so does Klout. For that matter, Viralheat has an API and they can
> get both Twitalyzer and Klout scores.
>
> That said, I don't know that there's ever really going to be a "one
> size fits all" Twitter user metric. But there are quite a few
> "crowdsourcing" and "curation" tools starting to show up, some of them
> open source. But personally, I think it's more fun to just collect raw
> data via the API and roll your own. ;-)
> --
> "A mathematician is a device for turning coffee into theorems." - Paul Erdos
>
Adam Green
your goals in ranking users? Are you looking for good people to follow
or engage with? In that case it doesn't really matter if they are a
bot or a human. What matters is whether others find them interesting.
Even more important is how interesting they are within a specific
domain. The best sports account, whether their tweets come from a
script or a human typing, means nothing to someone who doesn't follow
sports. Even if Twitter HQ maintained an absolute rank, it wouldn't
apply to all readers of an account. What I do for clients is rank
users within their specific area of interest based on a collection of
keywords. I ignore follower count, since that is so easy to game with
brute force methods. Mentions work as a better measurement of
popularity.
I have a detailed tutorial on mention tracking on my site:
http://140dev.com/twitter-api-programming-tutorials/identifying-influential-twitter-users/
Here is a summary of the algorithm. What I do is aggregate tweets for
keywords, and record @mentions. If @fred mentions @sally, I record
@fred as the source and @sally as the target. Then I can rank users by
the number of mentions they receive as the target, or even more useful
is ranking users by the number of different sources mention them. That
is hardest to game. The real key is that this is based on mentions in
tweets that contain specific keywords.
M. Edward (Ed) Borasky
> Terms like "spammer" or "good citizen" are kind of vague. What are
> your goals in ranking users? Are you looking for good people to follow
> or engage with? In that case it doesn't really matter if they are a
> bot or a human. What matters is whether others find them interesting.
Obviously, I don't speak for all Twitter "consumers of the feed" but I
have so far not found a single bot that I find interesting, and there
are some people bot-like enough that I won't follow them either.
--
M. Edward (Ed) Borasky
http://borasky-research.net http://twitter.com/znmeb
M. Edward (Ed) Borasky
> I think one solution is if twitter could use their hidden karma score
> to provide "good citizen" flag for us.
I wasn't aware that Twitter had a "hidden karma score." They've hidden
it well. ;-)
But seriously, there are two broad classes of data Twitter has that
are not available to outsiders via the APIs:
1. Twitter users' page reading behavior, and
2. Twitter users' search behavior.
And what Twitter has announced publicly or hinted at is so far only
the mystical "resonance" score for *tweets*, although I suppose you
could extend that easily to users by counting either total resonant
tweets for each user or fraction of tweets that resonated for each user.
I think in the end, though, a good Twitter "user score" is going to
look a lot like Twitalyzer's "Impact" score, which is based on how
other people interact with the user - how often they retweet the user,
how often they reference the user, how often they follow the user, how
often they add the user to lists, and so on. Twitalyzer can also
collect data from bit.ly and Google Analytics, but I don't think
they're integrated into the scoring mechanisms, since not all users
have GA accounts or use bit.ly link shortening / tracking.
--
M. Edward (Ed) Borasky
http://borasky-research.net http://twitter.com/znmeb