Transforming social media text into predictive tools for depression through AI: A test-case study on the Beck Depression Inventory-II.

Abstract

The characterization of mental states through assessment tools is a fundamental aspect in psychiatric and psychological clinical practice. In this context, standardized questionnaires based on Likert scales are often used for the assessment of emotions, attitudes, and perceptions. These tools enable clinicians and researchers to quantify subjective experiences, providing valuable data that elucidate the intricate nature of human emotions and beliefs. Despite their utility, administering and completing these questionnaires presents significant challenges. The process requires substantial time and resources from both clinicians and participants, which can create barriers to efficient data collection and analysis. Consequently, we aim to streamline this process without compromising the quality and reliability of the gathered data. This study was designed to develop a tool (aka EnsemBERT) that leveraging the power of Pre-trained Language Models (PLMs) could reliably predict the scores associated with each item of the Beck Depression Inventory (BDI-II) on the basis of users' generated social media posts. The results confirm that such AI-based approach is feasible and that the specific tool, i.e. EnsemBERT, can accurately predict questionnaire scores at various levels of granularity, i.e. individual item scores as well as overall questionnaire scores.