June 11, 2018
Is Yelp’s new user photo classification system a case of pattern recognition as a cultural form?
Some of the 100,000 pics snapped by Yelp users each day. Until recently, user-generated content like this has been made accessible as a database, timeline or data stream. Now, using a convolutional neural network, Yelp can identify and organise images much more accurately—sorting not just by image content but also by composition, tonal balance, color palette and overall aesthetic. It’s all about surfacing “beautiful” images, apparently—and burying the ones that look like Shitty Food Porn. Which kind of defeats the purpose of a review site, and leads to the depressively same-ish, faintly Leni Riefenstahl-ish effect above. Expect more aesthetics influenced by the power of deep learning to excavate similarity and analogy, though hopefully without the Stepfordy overtones.