Skip to main content



AI bot finds suicidal voices on Chinese social media, helping save lives

Wang Le’s bedroom is dim and silent, the curtains tightly drawn. The only sounds come from mouse clicks and a clattering keyboard. Unable to live and work like a normal person for nearly a decade due to a social phobia, the internet has been his only connection to the outside world. It even saved his life.

AI bot finds suicidal voices on Chinese social media, helping save lives

BEIJING — Wang Le’s bedroom is dim and silent, the curtains tightly drawn. The only sounds come from mouse clicks and a clattering keyboard. Unable to live and work like a normal person for nearly a decade due to a social phobia, the internet has been his only connection to the outside world. It even saved his life.

Mr Wang had to rely on his relatives to leave food at his front gate every week or two because he could not even order takeout if it meant talking to people by phone. In the spring, he contemplated suicide but hesitated. Afraid of death, but also afraid of life, he shared his despair on Weibo, a popular Twitter-like social platform in China.

“Are you OK?”, a stranger said in a message he received soon after. “Would you like to talk to me?”

In his late 20s, Wang lives alone in a small town in northern China. His parents are migrant workers in one of the country's biggest cities. The message moved him, knowing that somewhere in the world there was someone who cared for him.

Wang has since befriended the stranger, who turned out to be a psychological consultant. She found him with the help of the Tree Hole bot, an AI program that spots suicidal intentions on Weibo and alerts a group of nearly 600 psychological scholars, consultants and volunteers who reach out to those in trouble.

Launched in July 2018, the team has prevented more than 1,000 suicides, said Mr Huang Zhisheng, creator of the program and a senior artificial intelligence researcher at Vrije University of Amsterdam.

As the team is small and charitable in nature, they can only cope with the most urgent cases and even some of those do not respond to their outreach, Mr Huang explained.

The bot was named because it scans so-called tree hole posts on Weibo, where people share and comment on emotional stories. It is the modern day equivalent of whispering secrets into a tree hole in the old days. 

One of the biggest Weibo tree holes dates back to 2012, posted by a depressed girl before she committed suicide. Users still add comments on a regular basis, with more than one million to date.

In China, at least 136,000 people committed suicide in 2016, accounting for 17 per cent of the world’s total that year, according to the latest data available from the World Health Organisation (WHO). Suicide is the second most primary cause of death among 15-29 year olds globally, according to WHO, which projected that 1.5 million people of all age groups will take their own lives over the next year.

While research has found that one of the best ways to prevent suicides is for the distressed person to hear from those who care about them, people like Mr Wang not only lack family support, they live in smaller cities that have limited or no access to professional psychological services. For these people, AI technology plays an essential role in connecting desperate souls with resources.

Outside China, Internet and social giants including Google, Facebook and Pinterest have used AI to assess suicide risk among users and identify suicidal or self-injury content. However, there are limitations in using AI for the follow up process of finding suicidal people in time and helping those with mental health problems, especially given privacy concerns.

The Tree Hole bot automatically scans Weibo every four hours, pulling up posts containing words and phrases like “death”, “release from life”, or “end of the world”.

The bot draws on a knowledge graph of suicide notions and concepts, applying semantic analysis programming so it understands that “not want to” and “live” in one sentence may indicate suicidal tendency.

In contrast, Facebook trains its AI suicide prevention algorithm by using millions of real world cases. From April to June, the social media platform handled more than 1.5 million cases of suicide and self-injury content, more than 95 per cent of which were detected before being reported by a user. For the 800,000 examples of such content on Instagram during the same period, 77 per cent were first flagged by the AI system first, according to Facebook, which owns both platforms.

The accuracy of Tree Hole, now in its sixth generation, has reached 82 per cent, Mr Huang said. Although Tree Hole only applies to Weibo, which is mostly used by mainland Chinese, some researchers from overseas universities have invited Huang to collaborate and expand it beyond China.

Tracking and storing actionable mental health data without user-consent has raised privacy concerns. In the case of Tree Hole, Mr Huang said because it only monitors Weibo, which is an open platform, saving lives is considered a priority over protecting privacy.

He classifies suicidal posts into 10 levels, with the highest requiring urgent action because they contain the most detail, such as time, location and method of the suicide. Only posts at level five and above are reported to the team.

The peak time for posts indicating suicidal intentions is between 10pm and 2am and three out of four are from women.

Mr Wang keeps in touch with the psychological consultants from the Tree Hole team to relieve his distress which comes back from time to time. Ms Li Jiayi, one of the consultants who keeps in touch with him, said she has at least 20 such cases and is also busy with her own full-time consulting job.

The charitable work at Tree Hole is demanding. Mr Huang is currently developing a conversational bot that he hopes can chat with people in the same way as a human psychologist. The bot will be programmed to react appropriately to those with a more fragile personality.

However, Li said that from a psychological evaluation perspective, online conversations are not as effective as a face-to-face consultation. “Words and voice do convey some information about the person, but their movements and expressions do a better and faster job,” she said.

With that in mind, scientists have developed AI that can detect depressed individuals by scanning their faces. A team at Stanford University led by AI researcher Fei-Fei Li designed a diagnostic system that combines 3D facial recognition and natural language processing technology, reaching an accuracy level of 83.3 per cent.

In China, a handful of start-ups are doing similar research. Beijing-based WonderLab and Shenzhen-based Huayuntong have developed programs that monitor the mental health of inmates at the country’s prisons by monitoring the movement of their eyes, their breathing, facial expressions and voices.

Mr Chris Wong, Huayuntong's deputy general manager, said AI is not able to replace psychologists in the foreseeable future because that would require at least 500,000 data samples. That would come up against privacy concerns, he added.

Meanwhile, life has improved somewhat for Mr Wang. After the consultants from Tree Hole convinced his mother to live with him, he now gets regular meals. SOUTH CHINA MORNING POST

Related topics

mental health suicide AI

Read more of the latest in




Stay in the know. Anytime. Anywhere.

Subscribe to get daily news updates, insights and must reads delivered straight to your inbox.

By clicking subscribe, I agree for my personal data to be used to send me TODAY newsletters, promotional offers and for research and analysis.