How AI Selects The Tastiest-Looking Food Photos For Yelp
Yelp implemented computer vision and machine learning techniques to bring the most appealing photos to the top of the page
Ever wonder what makes those pictures from the local sandwich shop so swoon-worthy (especially before lunch)? Yelp plays host to a whopping 100 million photos in reviews of businesses around the world, with an estimated 100,000 images uploaded by users every day. To manage this colossal library, the company relies on an artificial intelligence that selects the most appealing photos and automatically positions them in the top spots on the page.
Yelp engineers implemented a neural network with computer vision techniques that recognize depth of field, contrast and alignment—all elements of photo quality. After a few trials, they further trained the system to to filter out pictures of logos (i.e. an attractive image of a restaurant’s signage without food or setting, which is unhelpful to Yelp users) and low-resolution photos that fall below the site’s pixel standard.
Yelp’s Photo Understanding team continues to develop more comprehensive machine learning to improve featured images. This summer, the company redesigned its mobile app to display the AI-selected photos more prominently on business pages.