Researchers at Brunel University London and the University of Leicester have developed an algorithm capable of identifying depressed Twitter users with roughly 88% accuracy.

According to the researcher, the mental state of Twitter users can be determined from 38 data points derived from a Twitter user’s public profile. Such data points include information about the content of the user’s Twitter posts, the timing of such posts, the wording, both positive and negative, of those posts, the number of their followers and friends, and their emoji use — all of which reportedly make a statement about the user’s mindset.

The algorithm was trained and subsequently tested on two databases featuring the Twitter histories of thousands of users — the Tsinghua Twitter Depression Dataset and the Johns Hopkins University's CLPsych 2015 dataset.

"We tested the algorithm on two large databases and benchmarked our results against other depression detection techniques," said Professor Abdul Sadka, director of Brunel's Institute of Digital Futures. "In all cases, we've managed to outperform existing techniques in terms of their classification accuracy."

The researchers report achieving 88.39% accuracy with the Tsinghua Twitter Depression Dataset and an accuracy rate of 70.69% with the Johns Hopkins University's CLPsych 2015 dataset.

Although not 100%, the researchers suggest that the results are “fantastic” and enough to help Twitter and other social media platforms like Facebook to flag potentially concerning posts that might one day lead to early depression diagnostics and thus early medical interventions. Likewise, the researchers believe the algorithm, which can easily be applied to other social media platforms, could be used for police investigations and employment screening use cases.

The article, "Cost-sensitive Boosting Pruning Trees for depression detection on Twitter," was published in IEEE Transactions on Affective Computing.