Engineers at the machine learning startup indico have been working to develop ad-targeting technology that let brands mine social media data and images in Facebook and Twitter, with Pinterest forthcoming, to gain greater insights about consumers.
While it's relatiely easy for marketers to analyze someone’s social media text and image history to quickly tell what activities a person likes, identifying his or her brand preferences, deep machine learning provides an entirely different level of sophistication, speed and scale.
Researchers at Indico use unstructured data, or "dark data," to interpret signals around text, images, sentiment and personality types, so marketers can gain a better understanding of consumers to refine ad budgets.
Slater Victoroff, indico CEO, found it possible to intuit emotions from text and images in Facebook to use as ad-targeting signals, and has been testing this concept for about two weeks. A similar project under way for about a month identifies and integrates a person's Myers-Briggs personality type gleaned from a "relatively small number of their posts." Victoroff said this creates a version of lookalike targeting.
The signals tie into another project based on photo recognition that's also been in the works with a couple of pilot customers for the past three months.This means working through an intermediary to target all Facebook and Twitter users who have posted a picture of a guitar or coffee mug, rather than a picture of a brand's logo.
"It will take between eight to 12 months of development to finish the fundamental technology," Victoroff said. "We're running initial tests before unveiling it to other companies."
Indico isn't the only company using deep-machine learning to target consumer ads. Other companies include Google, Facebook, and IBM Watson, as well as MetaMind and PredictionIO, two starts-ups recently acquired by Salesforce.