Detection and Analysis of Depression-Related Language in an Online Community: Machine Learning, Topic Modeling, and Corpus-Linguistic Approaches

Abstract

Depression is one of the most common mental disorders worldwide and may affect an individual’s ability to perform essential life activities. Still, “less than 25 % of individuals with depression receive adequate treatment” (American Association of Suicidology 2009). Instead of requesting support from mental health professionals, individuals often seek help through online communities. The main objectives of this study are to analyze the linguistic characteristics in a corpus of posts written by members of the depression community on Reddit and to investigate the major topics discussed in this community. To meet these objectives, we employed machine learning, topic modeling, and linguistic analysis. Results show that depression-related posts can be detected at an accuracy rate of 95.74 % using a Support Vector Machine with n-grams, Term Frequency-Inverse Document Frequency, and the Valence Aware Dictionary and sEntiment Reasoner as features. Additionally, BERTopic revealed the most frequently discussed topics, which included substance use and depressive symptoms. Further linguistic interpretation of these results highlighted patterns of self-reference and negative self-evaluation. The study provides new insights into the linguistic expressions used by members of the online depression community, which facilitates a better understanding of affected individuals and could help in developing strategies for detecting and treating depression.