MentalWords
Stress and mental health are a major challenge of today’s society. This project investigates how Natural Language Processing (NLP) can be used to learn more about these diseases.
Factsheet
-
Schools involved
School of Health Professions
Bern Academy of the Arts
School of Engineering and Computer Science - Institute(s) Institute for Data Applications and Security (IDAS)
- Research unit(s) IDAS / Applied Machine Intelligence
- Strategic thematic field Thematic field "Caring Society"
- Funding organisation SNSF
- Duration (planned) 01.09.2025 - 31.08.2029
- Head of project Prof. Dr. Mascha Kurpicz-Briki
- Partner Universität Bern
- Keywords Depression, Anxiety, Burnout, Machine Learning, Data Collection, Artificial Intelligence, Natural Language Processing, Large Language Models
Situation
Different studies show that stress at the workplace or regarding care work has become a prevalent societal problem. Persistent situations of stress can lead to mental health problems, such as, e.g., depression, anxiety, or burnout. Detecting mental illness is difficult, as they are the consequence of many different factors and manifest in a variety of both physical and mental symptoms. Furthermore, the symptoms of different mental health problems can overlap, making it unclear which underlying condition should be treated. In clinical intervention and field research, mental health problems are typically detected by using so-called inventories, with scaled-response questions. However, it has been shown that such inventories can have limitations, such as social desirability bias, defensiveness/denial or extreme response bias.
Course of action
Whereas the use of free-text questions is promising, the evaluation of such questions generates a large manual effort and is thus not often used in practice. New technologies from the field of machine learning and natural language processing (NLP) provide ways of automatically processing text. However, there are three main challenges: (a) most research in the field is relying on social media, there is a scarcity of clinical data being available for NLP in mental health research, (b) the clinical distinction between burnout, depression and other related problems is challenging, as symptoms might be overlapping, (c) it can be difficult for the concerned groups of patients to express themselves in a written form.
Looking ahead
In this project, technical and medical researchers will closely collaborate with a clinical partner to collect transcribed data from different groups of mental health patients and from the working population without clinical evidence of a mental health condition. We aim to develop a data collection protocol that integrates as smoothly as possible into the clinical workflow to limit the overhead of the clinical partner to a minimum in the long term. We analyze the transcribed data to explore the differences and overlaps between the groups and develop a model to identify indication for burnout, depression and anxiety in German text data.
