Saturday, August 10, 2013

Tony Abbott is paying people to like him.

Not too long ago Tony Abbott's staff uploaded this image to their Facebook page - 


100,000 likes is very impressive!  It's a pity that they are clearly being made by bots.  

Some enterprising souls noticed that Mr Abbott's likes had been growing very quickly of late.  Growing out of line with any expectation.  

This is what Mr Abbott's increase in likes looks like - https://www.facebook.com/TonyAbbottMP/likes

For comparison this is Kevin Rudd's - https://www.facebook.com/KevinRuddMP/likes

And the Liberal Party of Australia's - https://www.facebook.com/LiberalPartyAustralia/likes
Edit - Added the Liberal Party page for another comparison.

That alone looks suspicious.  Notice the sharp increase in the grey line - new likes - and the almost direct correlation between new likes and "people talking about this".  

Something is off.  

Luckily, as I said, some enterprising souls noticed this yesterday, and have been tracking the growth in likes ever since - http://pastebin.com/AhTATCb4


22 hours of data collected, once per minute. Sampling likes and "mentions" (mentions are cached and only update once every 24 hours).
Raw data is here: http://puu.sh/3YjTi.csv
This is what Abbott's graph looks like: 
http://puu.sh/3YkGP.png

This is Rudd's:

http://puu.sh/3YkIQ.png

There are four distinct periods: 
"Normal": 8am-11pm EST
"Rampdown": 11pm-12:15pm EST
"Low": 12:15pm-6:45am EST
"Rampup": 6:45am-8am EST

During these periods, the differences were pretty stark:

First Normal
Rudd: Mean 1, Stdev 1.18
Abbott: Mean 18.22, Stdev 4.31

Rampdown
Rudd: Mean 1.19, Stdev 1.07
Abbott: Mean 12.32, Stdev 4.43

Low
Rudd: Mean 0.22, Stdev 0.53
Abbott: Mean 2.89, Stdev 2.38

Rampup
Rudd: Mean 0.39, Stdev 0.69
Abbott: Mean 10.15, Stdev 3.06

Second Normal
Rudd: Mean 0.6, Stdev 0.82
Abbott: Mean 18.47, Stdev 4.96

Some notes:
Not only are the total numbers of likes way too high to account merely for likebait saturation, Abbott's variation is far too low for to be entirely human-based. There are also no major spikes that we'd tend to see during periods of policy releases (for instance, his release of indigenous policy this morning). The means of each time period tend to line up too perfectly and again, lack in variation. Overnight, the deviation returns to what we'd tend to expect from this kind of data - indicating that the bot/net are probably turned off overnight after they ramp down/up (so as to not see an immediate jump from a mean of ~16-20 straight to 2-3). In short, this is exactly how I'd code a bot to be difficult to detect (if I were actually ridiculous enough to do so).
There's some other notes: the number of mentions cached line up almost exactly with the number of likes added in the same time period - 6727 likes from collection start to cache refresh, 8010 new mentions since previous cache refresh. There are several hours unaccounted after the previous cache refresh, which would likely make up the missing number there. Note that in a similar time period, Rudd gained approximately 350 likes and 0 new mentions. As people tend not to reference people they don't know on facebook except through page likes (not mentions), you can pretty safely assume that each bogus account is also mentioning the page 0-1 times.
Since collection, Rudd has gained 582 likes and 0 mentions, and Abbott has gained 15867 likes and 8010 mentions. It's worth noting that Abbott averages to almost exactly 1000 likes per hour yesterday, and 1200 today. Rudd's is all over the place, by comparison - 20-60 yesterday to 35-50 today.
tl;dr it's a bot, but written exactly how I would do so if I were to write a bot to spam likes. I'd make a couple of modifications - randomised ramping start/end times (within a tolerance of 2 hours), and much greater variation in the number of likes per minute.
Another kind soul decided to compare the increase in likes to people outside of the Australian political spectrum, and found that Tony Abbott is gaining popularity at a greater rate than One Direction, Justin Bieber, and even Facebook itself.  He's around 50% more popular than Game of Thrones, even.


 It's just too perfect an increase in likeability, particularly when combined with the people talking about it, and the curve over time.

What it adds up to is somebody paying a bot to mass-like Mr Abbott's page.  Alas for the image at the top of this post, and for Mr Abbott's ego, the likes are illusory.

And against Facebook's terms of service.

Update August 11 (Morning):
Tony A-bot-t update: 
Rampdown / Rampup occurred at exactly the same time today, and at extremely similar rates. Normal operation has returned to very similar means as yesterday.  
For comparison:
First rampdown: Mean 12.32, Stdev 4.43Last night's: Mean 13.21, Stdev 3.6
 
First rampup: Mean 10.15, Stdev 3.06This morning's: Mean 10.26, Stdev 3.41 
And now we're back to normal operation of ~16-20 likes per minute.
From the last two days worth of data (sans most normal operation)(from normal -> rampdown -> low -> rampup -> normal)1094, 833, 510, 250, 128, 88, 85, 142, 356, 681, 1092 1123, 847, 595, 260, 168, 90, 112, 131, 401, 724, 1011

Update August 11 (Evening):
The graph for the two days of activity that were monitored.  Almost perfectly mirroring.  



Edit - 10th of August: Clarity.
Edit - 11th of August: Added the Liberal Party of Australia Likes graph. Updated with last night's data.