What your photos reveal about you

rolexThere has been a lot in the media recently, and indeed on this blog, about the data mining possibilities when it comes to recruitment, with many employers harvesting our social profiles for clues and insights regarding our suitability for a position.

What is perhaps less well known however is how brands can find insights from even the most unlikely sources.  A Carnegie Mellon study published recently shows for instance the insights brands can glean from the photos we post online.

The researchers analysed five million images posted onto social networks to try and see if any brand association can be found from those images.  48 brands were selected across sports, luxury, beer and fast food industries, with the images themselves harvested from Pinterest and Flickr.

The analysis process produced clusters of certain types of photos that are common amongst particular brands.  For instance it would be easy to associate watch pictures with Rolex, and the researchers found that the Rolex cluster of images often included horse-riding or motor racing events, both of which the company sponsor.

Such clusters typically emerge in photos that form part of a large collection, such as at a wedding, which offer brands an invaluable opportunity to see the context in which their products are used.

It offers brands an interesting insight into how customers are using their products.  It’s the kind of insight that would only typically have emerged via things such as a survey or focus group.

“Now, the question is whether we can leverage the billions of online photos that people have uploaded,” says researcher Gunhee Kim.

“Our work is the first attempt to perform such photo-based association analysis,” Kim continues. “We cannot completely replace text-based analysis, but already we have shown this method can provide information that complements existing brand associations.”

Now it should be said, that for this study, brands were only identified if users had actually tagged the brand in the photo, which in itself must narrow down the sample size a fair amount.  Nevertheless, the researchers did manage to develop both a method for analysing and clustering photos together, whilst also building an algorithm that would isolate the part of the image associated with the brand.

The researchers accept that this is but the first step on the road to accurately mining social data for branding purposes, but it seems likely that both brands, and of course the social networks themselves, will be very keen to further develop this process of drawing out greater information from the photos we post online.

Kim will present the research December 7 at the IEEE Workshop on Large Scale Visual Commerce in Sydney, Australia, and at WSDM 2014, an international conference on search and data mining on the web, February 24-28 in New York City. The National Science Foundation and Google supported the work.

Related

Facebooktwitterredditpinterestlinkedinmail

4 thoughts on “What your photos reveal about you

Leave a Reply

Your email address will not be published. Required fields are marked *

Captcha loading...