Mining Social Images
Written by Kay Ewbank   
Monday, 06 January 2014

Researchers have published analysis of a study of five million images from social media looking at how people view brands.

The researchers from Carnegie Mellon University analyzed five million images taken from photo-sharing sites such as Pinterest and Flickr to see whether images posted by people on social media can be used to find out what people feel about a particular brand. 

Traditional marketing uses questionnaires to gather this information, but the researchers wanted to see whether data mining and analysis could give more accurate information.

Eric Xing, associate professor of machine learning, computer science and language technologies, and Gunhee Kim, then a Ph.D. student in computer science, looked at images associated with brands in four categories — sports, luxury, beer and fast food. The analysis process was automated and looked for tags for 48 brand names on shared photos.



The researchers developed a method for analyzing the overall appearance of the photos and identified clusters of images based on core visual concepts associated with each brand. They also developed an algorithm that would then isolate the portion of the image associated with the brand, such as identifying a Burger King sign along a highway, or Adidas clothing or shoes worn by someone in a photo.

While some results were obvious - clusters of photos of watches for Rolex and tartan plaid for Burberry, other information was less obvious. For instance, clusters for Rolex included images of horse-riding and auto-racing events, which were sponsored by the watchmaker. Wedding image clusters were highly associated with the French fashion house of Louis Vuitton.


"Now, the question is whether we can leverage the billions of online photos that people have uploaded," said Kim, who considers the work to be the first attempt to perform photo-based association analysis. He added “We cannot completely replace text-based analysis, but already we have shown this method can provide information that complements existing brand associations."

While the work is at an early stage, Kim says it suggests some new directions and some additional applications of computer vision in electronic commerce.

For instance, it may be possible to generate keywords from images people have posted and use those keywords to direct relevant advertisements to that individual, in much the same way sponsored search now does with text queries.

Kim presented the research last month at the IEEE Workshop on Large Scale Visual Commerce in Sydney, Australia, where he received a Best Paper Award. He will also give a paper at WSDM 2014, an international conference on search and data mining on the Web, taking place in New York City, February 24-28.

More Information

Discovering Pictorial Brand Associations from Large-Scale Online Image Data

Related Articles

Deep Learning Finds Your Photos

Google Explains How AI Photo Search Works


To be informed about new articles on I Programmer, install the I Programmer Toolbar, subscribe to the RSS feed, follow us on, Twitter, Facebook, Google+ or Linkedin,  or sign up for our weekly newsletter.


C book



or email your comment to:



Google Gemini API Developer Competition

Google is running a Gemini API Developer Competition with prizes including a 1981 custom electric DeLorean. Entrants will use the Gemini API to tackle real-world challenges, and the organizers su [ ... ]

Is Rust Safe?

Rust is our great hope for the future. Even if you are not using it you probably have heard of it and believe it is a safer language than C or C++. But is it really?

More News



Last Updated ( Monday, 06 January 2014 )