How Do You Feel? Information Retrieval in Psychotherapy and Fair Ranking Assessment

Kumar V.;Medda G.;reforgiato recupero Diego.

;Riboni D.;Helaoui R.;Fenu G.

2023-01-01

---

Abstract

The recent pandemic Coronavirus Disease 2019 (COVID-19) led to an unexpectedly imposed social isolation, causing an enormous disruption of daily routines for the global community and posing a potential risk to the mental well-being of individuals. However, resources for supporting people with mental health issues remain extremely limited, raising the matter of providing trustworthy and relevant psychotherapeutic content publicly available. To bridge this gap, this paper investigates the application of information retrieval in the mental health domain to automatically filter therapeutical content by estimated quality. We have used AnnoMI, an expert annotated counseling dataset composed of high- and low-quality Motivational Interviewing therapy sessions. First, we applied state-of-the-art information retrieval models to evaluate their applicability in the psychological domain for ranking therapy sessions by estimated quality. Then, given the sensitive psychological information associated with each therapy session, we analyzed the potential risk of unfair outcomes across therapy topics, i.e., mental issues, under a common fairness definition. Our experimental results show that the employed ranking models are reliable for systematically ranking high-quality content above low-quality one, while unfair outcomes across topics are model-dependent and associated low-quality content distribution. Our findings provide preliminary insights for applying information retrieval in the psychological domain, laying the foundations for incorporating publicly available high-quality resources to support mental health. Source code available at https://github.com/jackmedda/BIAS-FairAnnoMI.