Yelp Collections uses machine learning to serve up recommendations

Yelp Collections, which launches on Android, iOS, and the web today, is all about highlighting the top restaurants, hair salons, hotels, and more in your area. The new feature uses a combination of machine learning, algorithmic sorting, and manual curation to put the local cream of the crop at your fingertips.

“This is something we’ve been working toward for the past six months,” Yelp product lead Jay Garg told VentureBeat in a phone interview. “The idea of curating content on Yelp makes a lot of sense.”

The feature is split into three categories: weekly user recommendations, Collections published by Yelp, and personal Collections users create themselves.

The weekly recommendations, which feed into a Collection called Recommended for You, are informed by machine learning. Yelp’s artificial intelligence engine bases recommendations on which business listings a user has viewed and the recommendations they received the previous week. “The emphasis is on fresh, dynamic content,” Garg said.

Collections created by Yelp, in contrast, are a combination of human-curated roundups and algorithmically generated lists.

Yelp’s social media and community managers wrap businesses — whether pubs, hotels, or pet groomers — into Collections and will eventually create themed Collections around events, holidays, and seasons.

Simultaneously, Yelp’s algorithms automatically publish “top 10 list” Collections for cities, and Garg said these are determined by a composite of star ratings and volume of ratings. “You’ll see a Collection with the top 10 cafes, for example, or top 10 brunch places,” he said.

The third pillar of Yelp’s new feature is user-created Collections. A “save” icon and Create New button in the Collections tab allows users to choose and add businesses to existing lists or create new ones. When they’re finished, they can add notes and descriptions to the collection or share it with friends.

Collections aren’t Yelp’s first foray into machine learning. In July 2016, the review platform rolled out deep learning-powered image analysis that automatically identifies the color, texture, and shape of objects in user-submitted photos, allowing it to predict attributes like “good with kids” and “ambiance is classy” with 83 percent accuracy.

Yelp also uses machine learning in its custom ads platform. Advertisers on Yelp can opt to have a “two-step” AI system recommend photos and review content to use in banner ads based on their “impactfulness” with users. The system has been shown to increase the rate that people click on ads by at least 15 percent, Yelp says.