Are You Really Okay? A Transfer Learning-based Approach for Identification of Underlying Mental Illnesses

Abstract

Evidence has demonstrated that there are similarities in language use across mental illnesses, and oftentimes underlying mental illnesses are also identified by healthcare providers following initial diagnosis. We review literature to identify semantic correlations and similarities among mental illnesses. We also present a novel transfer learning-based approach that learns from data associated with known mental illnesses and predicts the presence of those previously unknown by the model. Our model achieves a predictive accuracy of 85%, providing promising evidence that language models can harness learned patterns from known mental illnesses to aid in their prediction of others that may lie latent.

Tasks

Reproductions