Inside the offices of Toutiao, it’s hard to tell you’re just outside of Beijing and not in the heart of Silicon Valley.
The walls are decorated with stylish graffiti, and there are all the usual perks: a game room, a gym, and a sumptuous cafeteria. But Toutiao doesn’t only look the part. The company makes an insanely popular Chinese news aggregation app called Toutiao (meaning “headlines”). Toutiao boasts some 700 million users in China, with more than 68 million active daily users. The company gets most of its revenue from advertising, which is also tailored to users’ likes and interests.
Toutiao’s parent company, Bytedance, has grown at a dramatic rate. In 2014 Bytedance sought investment that valued it at about $500 million. With its latest fund-raising round, the company is reportedly seeking a valuation in excess of $10 billion.
The success of Toutiao has been achieved, in large part, by using machine learning to figure out users’ interests and tastes, and tailoring its offerings accordingly to get more clicks. Facebook and Twitter also use machine learning to refine recommendations, but they rely more heavily on social connections.
When you open the app, you’ll see a stream of posts, ranging from serious news reports to cute videos. The more you click, the more Toutiao learns your tastes and interests, and refines its recommendations. Toutiao was created in 2012 by Zhang Yiming, a 33-year-old entrepreneur.
When I first opened the app, I was presented with a broad range of content. The top story was a post about coming economic reforms. This was followed by a piece on the level of pollution in Beijing that day, and then a video of a popular stand-up comedian talking about the smog. Further down was a salacious story about a famous actress having an affair, and, curiously, an article on how to discern the sex of newborn chickens (I’m told this was aimed at rural readers).
In the coming year, Toutiao aims to develop more AI-focused services, and to expand internationally. The ultimate goal is to become the world’s number-one online content destination, eclipsing the likes of Facebook and Twitter, as well as the likes of the New York Times and Buzzfeed in their own markets.
Indeed, signs of high-tech hustle were everywhere when I visited last month. Scores of coders, designers, and marketers were busy working away in a modern, open-plan space, with desks arranged around a futuristic-looking circular glass meeting room. In one corner, a pile of several hundred unpacked ThinkPad laptops sat ready for new employees to claim.
I met with Lei Li, who heads up Toutiao’s AI lab, having joined the company from another Chinese Internet company, Baidu, in March of last year. Besides developing better ways of determining what posts to recommend to users, Li’s team has created an artificial reporter, called Xiaoming, that’s capable of generating short news reports on European soccer games. Li’s team is also working on ways to identify users that might be able to answer specific questions, with a view to offering a service similar to Yahoo Answer or Quora. Toutiao opened this problem up to outside researchers with a contest held in December.
“We really want to use machine learning to understand question-and-answering and comments and news,” Li told me. “Toutiao Lab’s mission is to push the boundaries of machine learning and AI and to transfer that into products.”
Toutiao already has several ways of identifying fake news. In China this mostly consists of dubious health recommendations. It does this using a mix of human reviewers along with automated analysis of posts and comments related to them.
Li also argues that Toutiao’s way of surfacing content results is less of a filter bubble than is often seen on Facebook, where people are often exposed only to posts from those with whom they already agree. This involves designing the recommendation algorithms to offer some content even if a user hasn’t signaled a preference for it so far. “Our [machine learning] models will try to expand your horizons, your interests,” says Li.
Jie Tang, a professor at China’s Tsinghua University who studies information diffusion and social network dynamics, says as Toutiao becomes more popular it will become increasingly important for it to make recommendations that are accurate and compelling. He says the company may also need to generate more of its own content. “You need something new,” he says.
Toutiao is already working on this. Besides linking to outside content, a growing number of posts and videos come from those posting directly on Toutiao’s platform. Users include news outlets, government agencies, and celebrities.
To some degree, the emergence of Toutiao symbolizes China’s transition from a low-wage, low-tech manufacturing powerhouse to a more technologically sophisticated market. Appropriately enough, the company’s headquarters are located in what used to be an aerospace manufacturing factory.
Bytedance also offers an English-language app called TopBuzz. Its future strategy will include investing in and partnering with local media companies. Last year it invested $25 million in an Indian aggregator called Dailyhunt. The company is also looking for revenue opportunities outside of advertising. Last September, Toutiao signed a deal with JD.com, a Chinese e-commerce site, allowing users to buy products through the app. Current investors in Toutiao include Sequoia Capital China and the Russian billionaire Yuri Milner.
Ultimately, Li says, the company’s efforts to expand internationally will be informed by the work his team is doing. “Many of the lessons we have learned can be shared for our international products,” he says. “We have built the largest machine-learning platform for content. That’s our weapon.”
Editor’s note: This article was updated February 2, 2017, to reflect Toutiao’s latest user figures.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
DeepMind’s AI predicts almost exactly when and where it’s going to rain
The firm worked with UK weather forecasters to create a model that was better at making short term predictions than existing systems.
People are hiring out their faces to become deepfake-style marketing clones
AI-powered characters based on real people can star in thousands of videos and say anything, in any language.
What an octopus’s mind can teach us about AI’s ultimate mystery
Machine consciousness has been debated since Turing—and dismissed for being unscientific. Yet it still clouds our thinking about AIs like GPT-3.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.