,
11 tweets,
6 min read
Read on Twitter
Just published: case study of some heavy-handed attempts to game Twitter in India in February.
Bots on both sides, but the attempt to boost the pro-Modi traffic was the craziest I've seen.
medium.com/dfrlab/electio…
Bots on both sides, but the attempt to boost the pro-Modi traffic was the craziest I've seen.
medium.com/dfrlab/electio…
We measured two hashtags using the Coefficient of Traffic Manipulation (CTM), and compared them with known traffic flows.
#GoBackModi was among the more heavily gamed hashtags.
#TNwelcomesModi was off the charts.
For the CTM, see comprop.oii.ox.ac.uk/wp-content/upl…
cc @polbots
#GoBackModi was among the more heavily gamed hashtags.
#TNwelcomesModi was off the charts.
For the CTM, see comprop.oii.ox.ac.uk/wp-content/upl…
cc @polbots
The CTM measures attempts by small groups to give traffic a disproportionate boost. Caveats:
1. It's comparative not absolute, so doesn't tell you "gamed / not gamed";
2. It's not just about spotting bots;
3. It shows attempts to manipulate traffic, not if they succeeded.
1. It's comparative not absolute, so doesn't tell you "gamed / not gamed";
2. It's not just about spotting bots;
3. It shows attempts to manipulate traffic, not if they succeeded.
It looks at three indicators:
R = % of retweets in the scan;
F = % of traffic generated by the 50 most active accounts;
U = average number of posts per user.
The point is to see if posts mainly came from many low-volume posters, a few high-volume ones, or simple retweets.
R = % of retweets in the scan;
F = % of traffic generated by the 50 most active accounts;
U = average number of posts per user.
The point is to see if posts mainly came from many low-volume posters, a few high-volume ones, or simple retweets.
The traffic on #TNwelcomesModi was crazy.
The first 50,000 tweets on February 9 came from just 891 accounts.
Of those, the top 50 generated over 30,000 posts in under 8 hours.
That's attempted manipulation on an industrial scale.
The first 50,000 tweets on February 9 came from just 891 accounts.
Of those, the top 50 generated over 30,000 posts in under 8 hours.
That's attempted manipulation on an industrial scale.
The top posters were massively active. This one posted on #TNwelcomesModi 1,803 times in the scan.
That's about one tweet every 15 seconds for almost 8 hours.
That's about one tweet every 15 seconds for almost 8 hours.
Important to note, though: these mega-tweeters didn't have many followers. They were in the double digits.
This was an attempt at massive amplification, but that's not the same as actually reaching people or having an impact.
This was an attempt at massive amplification, but that's not the same as actually reaching people or having an impact.
On the other side, #GoBackModi, the figures were relatively less wild, but still characteristic of attempted manipulation.
50,000 tweets from 7,394 accounts.
Over 16,000 tweets from the 50 most active posters.
This was more subtle, but not that much.
50,000 tweets from 7,394 accounts.
Over 16,000 tweets from the 50 most active posters.
This was more subtle, but not that much.
Interestingly, the most active posters on #GoBackModi were even busier than the pro-Modi ones individually, but there were fewer mega-tweeters in the mix.
This one posted 2,453 times on February 10. About one post every 40 seconds from midnight to midnight.
h/t @TwitonomyApp
This one posted 2,453 times on February 10. About one post every 40 seconds from midnight to midnight.
h/t @TwitonomyApp
This one posted 1,899 times in about 3 hours, or roughly one tweet every 6 seconds.
It had rather more followers, but still a low number.
It had rather more followers, but still a low number.
Upsum, these manipulation attempts ranged from large to extreme.
They were too clumsy to have much impact, but the sheer scale of the attempts on both sides is worrying, ahead of the election.
//
Thread ends.
They were too clumsy to have much impact, but the sheer scale of the attempts on both sides is worrying, ahead of the election.
//
Thread ends.
