Fashion taste is a fickle thing, but scientists have long pursued the neurological causes behind our aesthetic choices. We are still far from understanding the motives, but MIT Technology Review reports that a team at eBay Research Labs in San Jose is using neurology research and in-person opinions to help craft an algorithm that determines fashion combinations — ideally to be used to recommend clothing during shopping experiences.
In order to test the strength of recommendations, researcher Anurag Bhardwaj and his team developed two different kinds of algorithms to put in front of human subjects. The first algorithm analyzed the colors of clothing found in more than 13,000 different photographs, seeking out the best match for a given article of clothing (based on an undisclosed rating system) compared to the other possible matches in the database. The second algorithm operated on a much simpler premise: a patterned piece of clothing looks best with a solid piece of clothing.
The researchers surveyed 150 subjects from Amazon’s Mechanical Turk, asking them to rate the algorithms’ success based on the number of successful matches made.
While the researchers gained important insights — including the fact that many of the users noted more successful matches with simpler patterns than complex patterns — the report still shows the obstacles researchers face in finding a data-driven (if user-curated) algorithm for something as subjective as fashion. Particularly, one fashion result may work for one person and not the other, so an algorithm must be sophisticated enough to collect nuanced individual data about one user’s particular tastes. It’s a tough problem to solve, but unlocking that potential could be a major boon for ecommerce companies like eBay in the long run.