A Simple and Flexible Modeling for Mental Disorder Detection by Learning from Clinical Questionnaires

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 12190–12206 July 9-14, 2023 ©2023 Association for Computational Linguistics A Simple and Flexible Modeling for Mental Disorder Detection by Learning from Clinical Questionnaires Hoyun Song Jisu Shin Huije Lee Jong C. Park∗ School of Computing Korea Advanced Institute of Science and Technology {hysong1991,jisu.shin,angiquer,jongpark}@kaist.ac.kr Abstract Social media is one of the most highly sought resources for analyzing characteristics of the language by its users. In particular, many re- searchers utilized various linguistic features of mental health problems from social media. However, existing approaches to detecting men- tal disorders face critical challenges, such as the scarcity of high-quality data or the trade-off be- tween addressing the complexity of models and presenting interpretable results grounded in ex- pert domain knowledge. To address these chal- lenges, we design a simple but flexible model that preserves domain-based interpretability. We propose a novel approach that captures the semantic meanings directly from the text and compares them to symptom-related descrip- tions. Experimental results demonstrate that our model outperforms relevant baselines on various mental disorder detection tasks. Our de- tailed analysis shows that the proposed model is effective at leveraging domain knowledge, transferable to other mental disorders, and pro- viding interpretable detection results. 1 Introduction Mental health problems, a significant challenge in public healthcare, are usually accompanied by distinct symptoms, such as loss of interest or ap- petite, depressed moods, or excessive anxiety. As these symptoms can often be expressed over social media, detecting mental health conditions using so- cial media text has been studied extensively (Yates et al., 2017; Coppersmith et al., 2018; Matero et al., 2019; Murarka et al., 2021; Harrigian et al., 2021; Jiang et al., 2021; Nguyen et al., 2022). Such ap- proaches could give rise to a monitoring system that provides clinical experts with information about possible mental crises. To automatically identify mental health prob- lems, traditional approaches focus on finding lin- guistic patterns and styles from the language of ∗ Corresponding author psychiatric patients. Utilizing these features, sta- tistical models can explain the correlation between linguistic factors and mental illnesses. However, these approaches suffer from increased complexity of models, necessitating pipelines of steps, from engineering features to producing results. By con- trast, more recent works have employed strong pre- trained models, which allow a direct use of raw data and simplify model development (Matero et al., 2019; Jiang et al., 2020). While such end-to-end approaches may be effective at achieving higher performance, they often lack domain-based inter- pretation, which is essential for decision-support systems (Mullenbach et al., 2018). Hence, there is a trade-off between providing interpretable predic- tions based on domain knowledge and the simplic- ity of the models. The lack of a sufficient sample size for high- quality data is another challenge in the clinical do- main (De Choudhury et al., 2017; Harrigian et al., 2020). Despite the availability of diverse datasets and methods for detecting mental disorders, most of them aim primarily at …

Content truncated to 500 words