Baidu, the equivalent of Google in China, will roll out an update to its group-buying ecommerce app, Nuomi, later this year. This will significantly change the way people review food — in addition to accepting and posting users’ reviews of restaurants, the app will start asking for people’s impressions of individual dishes.
Instead of having users simply snap photos and type in a few words later — and maybe upload photos — the app will also let people use voice dictation to quickly jot down their thoughts about the dishes they photograph. That way, users can receive more useful information about the specific foods they can order from restaurants.
Not that the app isn’t already intelligent: It’s collected so many food photos that when you upload a new dish, it identifies the dish and the restaurant. The company has data from hundreds of thousands of restaurants.
“I would say 90 percent of the dishes I think we are very well able to recognize,” Yuanqing Lin, director of Baidu’s Institute of Deep Learning in Beijing, told VentureBeat in an interview this week at Baidu’s Silicon Valley A.I. Lab in Sunnyvale, California.
AI Weekly
The must-read newsletter for AI and Big Data industry written by Khari Johnson, Kyle Wiggers, and Seth Colaner.
Included with VentureBeat Insider and VentureBeat VIP memberships.
But the dish-level information could distinguish the app even more from other online-to-offline (O2O) services with which people can order car washes, laundry services, personal chefs, food delivery, and even massages on demand.
Baidu, whose mobile search app has more than 663 million monthly active users, doesn’t have legions of people looking at photos to figure out what’s what in the photos added to the restaurant part of Nuomi, and there aren’t people in call centers who will be asking people what they think of the food they eat. The company is doing these things with deep learning, a type of artificial intelligence that involves training artificial neural networks on lots of data, such as photos of food, and then getting them to make inferences about new data.
Apple, Facebook, Google, Microsoft, and Twitter have all invested in deep learning, as well, and have released new features of their apps that take advantage of the technology. For instance, the Smart Reply feature of Google’s Inbox email app suggests one-line responses to incoming emails, and the Facebook app now offers machine-generated photo captions for the News Feed through the screen reader on iOS to help blind users understand what their Facebook friends are sharing.
Baidu uses deep learning for speech recognition, image recognition, and natural language processing for many of its applications. One example is the DuLight app that’s designed to help the blind. Based on what’s in front of the camera on a mobile device, the app can estimate the age and gender of a person, tell you the denomination of a piece of currency, turn printed text into speech that a blind person can listen to, and recognize objects.
The underlying A.I. systems get smarter as they sop up more data. So encouraging users to adopt a new behavior — talking into your phone at the table — might seem like something researchers thought up in a lab purely as a means to make their software smarter. But it’s more than that, because dish-level reviews in Nuomi (literally translated as “sticky rice” in Mandarin) could actually be useful to end users.
Over time, the app could become so smart about food that it will be able to tell users the best place to get a certain dish.
“I’m visiting the Bay Area,” Lin said. “If I open the app [looking for a] ranking of which restaurant in the area cooks the best Kung Pao chicken right now, there’s no such thing. So it could be extremely powerful for food recommendation.”
The idea is that the new version of Nuomi — which Baidu first acquired in 2014 from Renren — will catch people’s attention by virtue of its food knowledge and thereby get people to share more information. For example, if it sees a user uploading five photos, it could tell the user that it knows that the first dish is Kung Pao chicken, the second dish is sauteed eggplant, and so on. From there, the app will direct the user to push a photo in order to talk to the app and record their observations about the dish. The app will dutifully transcribe the user’s words and turn that into the text of the dish review.
Baidu will talk more about Nuomi’s dish-reviewing capability at its Baidu World conference coming up later this year, Lin said.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn More