And with an accuracy of almost 90 percent.

Unfortunately, depression is a common problem. In that case you have a gloomy mood for a long time, you feel depressed and ‘disturbed’, when you no longer manage to live in resonance with yourself and the surrounding world. According to the World Health Organization, 300 million people worldwide (!) suffer from depression. Those are big numbers. But how do we detect in time whether someone around us is depressed? Perhaps thanks to a clever algorithm on Twitter.


Researchers have in a new study developed an algorithm that can recognize depressed Twitter users in 9 out of 10 cases. And that is an important step forward. Such a system could be able to identify a depressed Twitter user in time, enabling early diagnosis. The newly developed algorithm therefore paves the way for social media platforms to proactively detect mental health problems in users.

The numbers
Unfortunately, depression is also rampant in the Netherlands. It is estimated that one in five Dutch people will experience depression at some point in their lives. This applies to about five percent of adults every year. Women in the age group 25-34 are most likely to become depressed.

The algorithm works as follows. It determines a person’s mental state by analyzing 38 different factors from his or her public Twitter profile. It looks at the content of the messages and determines, for example, how many positive and negative words and emojis are used. The time at which a message is posted and the number of friends and followers are also taken into account. Based on that, the algorithm ultimately makes a decision about the user’s mental and emotional state.


Researchers trained the algorithm using two databases, which contain the Twitter histories of thousands of users, as well as additional information about their mental health. And as the results show, the algorithm has become quite adept at exposing depressed Twitter users. For example, it was correct in 88.39 percent of the cases. A particularly high score. “Anything above 90 percent is considered excellent,” said researcher Abdul Sadka. “So 88 percent is really fantastic. The closer you get to 90 percent, the better.”


Incidentally, the researchers underline that their proposed algorithm is platform-independent. “This means it can also be easily extended to other social media platforms, such as Facebook or WhatsApp,” said study researcher Huiyu Zhou. In addition, the algorithm does not only have to be used to detect depressed users in a timely manner. It may have different purposes in the future. For example, the researchers state that it may also be useful in criminal investigations.

While the algorithm is already partially proven, the research team plans to refine it further. “In the next phase of this research, we want to explore its validity in different environments and backgrounds,” Zhou outlines. “And, more importantly, the technology that emerged from this research can be further developed for other applications. Think of e-commerce, recruitment research or screening of candidates.”

Did you know…

…Facebook posts can predict some 21 health conditions? For example, the messages reveal no fewer than 21 different diseases and disorders, including diabetes, high blood pressure, psychosis and depression. Read more here!