Classifying depression severity in online chats through human-coded psycholinguistic analysis using DSM-5 and Beck Depression Inventory

Despite advances in artificial intelligence, accurately detecting the severity of depression in online communications remains a challenge, underscoring the need for expert-led psycholinguistic analysis. This study employs such an approach to examine depression and other mental health issues in online chat data. Depression severity was classified using DSM-5 and Beck's Depression Inventory by five mental health professionals, with results tested for inter-rater reliability. Human-coded psycholinguistics adds clinical nuance to the classification. A random sample of 4,000 chat entries was analysed, with five professionals independently categorising each entry based on the DSM-5 and Beck Depression Inventory criteria into the categories of no depression, mild, moderate, severe, or unknown. The analysis showed a high average inter-rater reliability, indicating substantial agreement among raters. Results revealed that 7% of chats exhibited some level of depression (2% mild, 2% moderate, 3% severe), 19% indicated other mental health issues such as anxiety, and 58% were ambiguous. These findings suggest that psycholinguistic analysis of online communication has strong potential for early detection of mental health issues. Integrating such features into digital tools could enhance early identification on online platforms, enabling timely intervention and better mental health support within communities.