New bot can spot nine out of 10 depressed Twitter users, scientists say

Scientists have developed a new algorithm that they say can spot depression in Twitter users with about 90 per cent accuracy, an advance that may lead to future early diagnosis methods.

The algorithm, described in the journal IEEE Transactions on Affective Computing, determines a Twitter user’s mental state by extracting and analysing 38 data points from their public profile, including the content of their posts, their posting times, and the other users in their social circle.

“We tested the algorithm on two large databases and benchmarked our results against other depression detection techniques. In all cases, we’ve managed to outperform existing techniques in terms of their classification accuracy,” study co-author Abdul Sadka, director of Brunel’s Institute of Digital Futures, said in a statement.

A large number of potential depression sufferers across the world do not seek professional help due to several factors, including social stigma or themselves being unaware of their mental condition, leading to “severe delay of diagnosis and treatment”, scientists said.

Previous research has shown that social media data can provide valuable clues about the physical and mental health status of individuals.

In the new study, researchers trained the algorithm using two databases that contain the Twitter history of thousands of users, alongside additional information about those users’ mental health.

“In this paper, we argue that it is feasible to identify depression at an early stage by mining online social behaviours,” the scientists said.

They used about 80 per cent of the information in each database to teach the bot, and the remaining data to test its accuracy.

After running through the database and excluding users with fewer than five tweets and correcting for misspellings, the bot then considers 38 distinct factors – such as a user’s use of positive and negative words, the number of friends and followers they have, and their use of emojis – to estimate a user’s mental and emotional state.

The team said it managed an accuracy of 89 per cent.

Researchers said they achieved an accuracy of about 71 per cent using John Hopkins University’s CLPsych 2015 dataset.

“It’s not 100% accurate, but I don’t think at this level any machine learning solution can achieve 100% reliability. However, the closer you get to the 90% figure, the better,” Dr Sadka said.

Scientists said the system could potentially flag a user’s depression one day before they post something into the public domain, and pave the way for social media platforms like Twitter and Facebook to proactively flag mental health concerns with users.

They said the bot may be further developed and used for a number of applications such as sentiment analysis and criminal investigations.

But the findings also raise further questions about data privacy, and the need for informed consent from users before their public data is used for analysis.

“The next stage of this research will be to examine its validity in different environments or backgrounds, and more importantly, the technology raised from this investigation may be further developed to other applications, such as e-commerce, recruitment examination or candidacy screening,” Huiyu Zhou, another co-author of the study from the University of Leicester in the UK, said.

“The proposed algorithm is platform-independent, so can also be easily extended to other social media systems such as Facebook or WhatsApp,” Dr Zhou added.