Did you know that advertisers are mining your photos too?
Have you ever had that thing where after looking up something to buy it will somehow seem to follow you around the web?
Googling for a new phone one day turns into endless adverts for phones on every website you visit. It’s the same on Facebook – post about something one day, and a few days later you’ll start seeing adverts that are peculiarly specific to your circumstances.
As is well known, this isn’t a coincidence – it is a result of advertisers mining the information they have on you. Every time you post something online you’re potentially giving out more information that advertisers can use to better target you.
This is fairly well known – but did you know that they might be mining the actual contents of your photos too? For example – if you post a photo that has a Starbucks logo in the background then it is conceivable that Starbucks will know about it.
Ditto is a photo analytics service aimed at big companies to monitor what people are doing and saying about them online. The company claims that it can not only spot brand images in photos, but also faces – and even work out what mood the poster is in from their photo.
You can see this in action on the company’s Stream Ditto website.
This data can also be mashed up with other publicly accessible metadata that the company will find attached to tweets: It can also report on how prominent the brand mention was (ie: identifying how many followers people sharing images have), what time of day the photos were taken and so on. This means that a surprisingly sophisticated picture of who you are can be built up. Here’s the full list of data that Ditto reckons it can figure out:
Whilst certainly clever – it is pretty straightforward to figure out where all of the data comes from. Twitter and other social sites all offer “firehose” access to developers, meaning they can tap into the incoming tweet stream and download all of the incoming data. Then it is merely a case of picking through and analysing the data: Each tweet comes with a timestamp, and often a geo-location so the company can work out where you are and when you were there – mash this up with, say, a mapping dataset and you can figure out if the person is in, say, a bar or a library. Thanks to the rise of the selfie, everyone is sharing more photos of themselves with which facial recognition software can be trained – so such a system could conceivably figure out who you’re with and what brands you are interacting with, without you having needed to be tagged.
Heck, if Ditto wanted to, it could probably build the most evil app ever: Taking your drunken Instagram photos, running a facial analysis against LinkedIn profile pictures, and automatically emailing them to your boss.
So this raises the question… Is photo mining a step too far, or merely an extension of what we already see when we post online? Should we be creeped out or accepting? Let us know what you think in the comments.