Bytedance, which was valued at US$75 billion after its latest funding round, has run afoul of regulators for vulgar content and on hit short-video sharing app TikTok, known as Douyin on the mainland.
Tracy Qu · · 2 min readBytedance upgrades AI tool to identify vulgar content on news app Jinri Toutiao
Bytedance, the world’s most valuable technology startup, has launched an updated AI tool to better identify violent and sexual content on its popular news aggregation app, Jinri Toutiao, amid Beijing’s efforts to clean up the internet in China.
The AI tool called Lingquan, which roughly translates to “dogs with good sense of smell,” has already been adopted by more than 3 million of Jinri Toutiao’s regular users and content producers, providing them with “health scores” on the stories, text, or videos they create for the app’s news feed.
Photo credit: Gavin Tracy
Sometimes, these creators are not aware that they have produced vulgar content, according to Wang Changhu, a director at Bytedance’s AI Lab, at a press conference in Beijing on Tuesday. His comments from the conference were confirmed by a company spokeswoman on Wednesday.
Known officially as Lingquan 3.0, the AI tool is also expected to help raise public awareness against content deemed as inappropriate by Chinese regulators.
Lingquan would also be used as a way to gather public opinion and suggestions about the regulation of vulgar content, according to Wang.
Bytedance’s use of AI to help sanitize content on its news app reflects the pressure that China’s cyberspace administration has put on major operators of popular social media platforms in the country.
Under Chinese President Xi Jinping, the ruling Communist Party has tightened its grip on the nation’s internet sector, which has the world’s largest online population with 829 million as of end of December, to guard against the proliferation of pornography, gambling, fake news, and political dissent.
That has led more Chinese social media platform operators to employ various AI tools, which handle the grunt work of labeling, rating, and sorting content online.
Its Lingquan tool has achieved a correction rate of 85%, up from 73% in the previous version, according to a post published by Jinri Toutiao. Apart from rating scores, the tool also offers suggestions if certain content “has some problems” or needed to be checked by a person.
At the press conference, Wang said the AI tool still required human check in identifying sexual content, for example, because an exposed part of the body is one of its general markers.
Visit SCMP.com for the latest China tech news.
Copyright (c) 2019. South China Morning Post Publishers Ltd. All rights reserved.
Stay updated on the go with our mobile app.
Get latest insights with smoother, more personalized experience through TIA mobile app.
