SPIN at MentalRiskES 2023: Transformer-Based Model for Real-Life Depression Detection in Messaging Apps

Abstract

Depression is a prevalent and severe mental health condition that significantly impacts global population, causing personal suffering and reduced quality of life. Its symptoms are often visible on social media and digital platforms, making them valuable for detecting depression. This paper represents our submission for the MentalRiskEs task at IberLEF 2023. We present a novel hierarchical model for real-time chat applications, using natural language processing techniques to identify individuals at risk. Our approach combines similarity-based stance representation with a sentence-level transformer encoder block, reducing manual effort and time required for feature selection. Our focus includes binary classification of depressed and non-depressed users, as well as multi-class classification based on the user’s coping mechanisms.