Facebook for Marketing Research: Political Polls #auspol
Using Facebook for sentiment analysis of the coming election. A free marketing research tool. Beats focus groups!
Want to know how 11 million Australians will vote? What if instead of checking a thousand people, instead of ringing those poor souls when they get home from work and are trying to get dinner on the table, we could just check in on what they were thinking, facebooking, tweeting about? What if instead of a thousand people we could check 11 million Australians or even 1 billion people worldwide? What if we could go on Facebook and have a look at 11.5 million Australian Monthly ACTIVE Users and determine how they were feeling about Julia Gillard, Tony Abbott, the Labor Party or the Liberal Part?
Well actually, we can. Sort of.
I read an article on Wendy Harmer’s Hoopla Polls Fake Story by @gabriellechan that got me thinking about Nielsen polls, sentiment analysis and focus groups. And how totally outdated, outmoded and antique they seem compared to Big Data of social media. Not maybe where it is today, but where it’s going…
UPDATE: #Spill 26 June – interesting to see where this shifts to in a week or two.
- Julia Gillard 220,000
- Kevin Rudd 198,000
- Tony Abbott 120,000
- Malcolm Turnbull 24,000
Polls suck Big Data rocks
Nielsen, Gallop, Newspoll, I don’t care, you all suck. Worked for a while (maybe) but now? No. Polls are about declared, conscious statements: Big Data is about undeclared analysis. Ask someone what they read and they will say a worthy tome like The Australian. Watch what they read and it’s some trashy mag. Expecting 1400 people to represent 20 million is deranged. Especially when most of the people contacted are a tainted database – “people on our business list” is I think Nielsens database. That wouldn’t be a drug user out of rehab then would it?
Does Facebook do Sentiment Analysis?
Yes. How well, we don’t know, but we suspect billions of dollars and newly minted PHDs are being thrown at it. We already know that Facebook does sentiment analysis of billions of status updates around particular subjects such as “Thanksgiving”. This article of mine from 4 years ago is about Facebook analysing sentiment to asses Gross National Happiness. Checking the collective temperature of a billion members, as it were. Behavioural Statistics is THE job of the future and of Big Data.
Facebook Advertising and Sentiment
It’s not huge stretch to see that companies will want (most of the time) to advertise to people who are positive and beaming about their products, or neutral and enquiring about their products. Trying to bring people to your Facebook Business Page that actively hate you is an option not often taken up by wussy marketers. 😛
So it seems unlikely that when a record company advertises Britney Spears new album on Facebook, and puts in Britney Spears as the keyword to find people, that they would put up with attracting a bunch of people who are all, like “I’m so over Britney Spears”, or “Britney Spears sucks”. Just because we mention a keyword doesn’t mean we are in the target marketing segment. Trust me, Facebook are working hard to remove the haterz. Howevery, irony stings and in Facebook Advertising: Irony Is The Algorithm’s Achilles Heel, (Forbes) we can see that Facebook fell over, big time, in sponsored stories to those having a laugh. Lube anyone?
So given that we know Facebook does sentiment analysis and that it is improving all the time, can we see how Julia Gillard vs Tony Abbott, Australian Labor Party vs Liberal Party of Australia are faring in discussions online? Does Uncle Facebook really know all about us, including how we will vote?
Facebook Advertising as a Free Marketing Tool
One of the things I point out in my classes is that you don’t have to advertise to get a good grasp of your market segments keywords ie. what they are talking about. Just use Facebook Ads to check number of “prospects” chatting on that topic in the last 30 days.
For members it works this way: Facebook looks at what you are putting in your status updates (say, “surfing”), what links you click on (“surfing video”), Business Pages you Like (“Ripcurl”) and what your friends also do (social graph behaviours). Surfing then become your keyword and you get ads targeted at the surfing psychographic. Yay! At least it’s more useful than moisturiser ads. Unless you got sunburned surfing…?
Question:
How many Australians are over 18 and active on Facebook?
Oh about 10,548,920 (Facebook rounds to nearest 20).
How to do Facebook Marketing Research
Go to Facebook.com/ads
In the first box put any old website (not Facebook Page) into the address bar.
In the second one, try these keywords – Julia Gillard, Tony Abbot, Australian Labor Party or Liberal Party of Australia.
Facebook will then assume you want to target a product to supporters of those “keywords”. And return a “prospect” number.
Twitter sucks at this sort of thing and will pay the price. Google on the other hand will probably come in even stronger in the next year or so on Big Data marketing research.
This is what I got – note that it is people who “LIKE #JULIA GILLARD”. The Like implies “like” not “talking about”. The #hashtag tells us that Facebook is or will soon be using, hashtags for keyword social graph searches like Twitter does. And more!
Here’s Tony Abbott
Here’s Liberal
What do you think? can Facebook predict an election outcome? or is it just more big numbers?
Where will it all end up, hmmm?
hmmm… the path looking forward, I suppose – still underdone at the moment, but I reckon FB will fine tune that – who’s got time nowadays for a 1 hour Nielson or Morgan poll? probably still valid but outdated, to say the least.
Your comment about the major polls ‘Nielsen, Gallop, Newspoll, I don’t care, you all suck. Worked for a while (maybe) but now? …. Expecting 1400 people to represent 20 million is deranged.’ shows your complete lack of understanding about how science is conducted as a discipline. You fail to note that these methods are still able to accurately predict an election within known sampling error. Your suggested methods have no hope in hell of doing this. Central Limit Theorem proves how you can ask 1,400 to represent 20 million as long as a probability sample is used (which is becoming more and more of an issue). Lets not throw science out the door just because its easy to do things on Facebook.
You may be correct, but to then suggest the numbers Laurel shows above are false would be just as wrong. The polls leading up to the recent US election varied wildly. A small sample can predict a large population as long as that sample is representative and the poll questions not leading.
You may be correct, but to then suggest the numbers Laurel shows above are false would be just as wrong. The polls leading up to the recent US election varied wildly. A small sample can predict a large population as long as that sample is representative and the poll questions not leading.