Show me your posts, and I’ll tell you if you’re burned out
Artificial intelligence helps to detect burnout. Researchers supported by the Swiss National Science Foundation have developed a promising technique based on automatic analysis of text data.
Burnout refers to a state of profound physical and mental exhaustion. It is difficult to detect because its symptoms can overlap with those of depression and anxiety. But artificial intelligence may hold the key to better recognising it. In an article recently published in Frontiers in Big Data (*), a team of researchers funded by the Swiss National Science Foundation (SNSF) report a new technique that uses natural language processing to detect burnout.
Burnout is typically diagnosed by means of psychological tests that take the form of a response rating scale. For example: “I’m exhausted at the end of the day: never/sometimes/every day”. However, these kinds of tests have significant limitations. For instance, some people are reluctant to check the answers “never” or “every day” or are tempted to lie to influence the results.
More comprehensive tests consisting of open-ended questions could also be used to detect burnout. These tests elicit more relevant information but require extensive analysis. Consequently, they are rarely used in practice.
A technique based on analysing texts
This is the problem that Mascha Kurpicz-Briki, professor of data engineering at the Bern University of Applied Sciences in Biel, and her team wished to address. To do so, she employed methods from natural language processing based on artificial intelligence to identify indicators of burnout. The method successfully identified 93% of burnout cases. Says Kurpicz-Briki: “Natural language processing effectively detects burnout and does it relatively efficiently, which is very promising”.
For this work, Kurpicz-Briki and her team analysed texts from Reddit, a social media platform that serves as a forum for discussions organised by topic. Kurpicz-Briki built up a database of more than 13,000 free-text samples. Some of them were pulled from discussions relating to burnout, whereas others were pulled from forums on a range of other topics.
Models trained on different data
Kurpicz-Briki then applied machine learning to develop a technique for determining whether a text contains indicators for burnout. Specifically, she first classified the text samples. The texts from discussion threads relating to burnout were classified manually to exclude those in which “burnout” referred to something else. Texts from other, non-mental health related discussion threads were labelled as not linked to burnout. Based on these examples, Kurpicz-Briki trained several models. Each used different approaches to determine whether a text never before seen by the model contained indicators of burnout or not. The models were then pooled to create the diagnostic method, which proved very effective.
The results are promising but need to be confirmed. As a next step, collaboration with medical experts is necessary to verify the conclusions of this preliminary investigation on real cases of burnout and on a representative sample of the population. Data collected on Reddit are anonymised.
Rapid testing of new ideas
This project was supported by the SNSF’s Spark scheme. The aim of Spark is to fund rapid testing or development of new scientific approaches, methods, theories, standards and ideas for application. It is designed for projects that show unconventional thinking and introduce a unique approach. The focus is on promising ideas of high originality, with minimal reliance on preliminary data. Spark was run as a pilot funding scheme by the SNSF in 2019–2020. It is currently being evaluated to define its future.
The text of this press release, and further information are available on the website of the Swiss National Science Foundation.
Haute école spécialisée bernoise
Phone: +41 32 321 63 13
G. Merhbene, S. Nath, A. Puttick, M. Kurpicz-Briki: BurnoutEnsemble: Augmented Intelligence to Detect Indications for Burnout in Clinical Psychology. Frontiers in Big Data (2022). https://doi.org/10.3389/fdata.2022.863100