MindWell: A Conversational Agent for Professional Depression Screening on Social Media

Eliseo Bao, Anxo Pérez, Javier Parapar

Abstract: Depression is among the most prevalent mental health conditions, with an early and accurate diagnosis being essential for mitigating its effects. Yet, stigma often prevents individuals from seeking professional help. In this context, social media offers a unique resource for depression screening, as users frequently share, comment, and disclose their daily struggles, providing key insights into their mental health through online activity. However, the immense volume of data generated on these platforms presents a significant challenge, requiring substantial time and effort for mental health professionals to analyze. This demo paper introduces MindWell, an open-source conversational agent designed to support clinicians in identifying symptoms and emotions relevant to clinical assessments. MindWell uses a Retrieval-Augmented Generation (RAG) framework, incorporating a Large Language Model (LLM) based on Llama 3.1 and fine-tuned specifically for depression screening based on clinical symptom criteria, particularly the Beck Depression Inventory-II (BDI-II). By leveraging users’ social media history as informed and reliable context, MindWell is designed to answer questions formulated by clinicians, facilitating the review process. We collaborated with a professional psychologist to assess MindWell’s responses in a clinical setting, finding that the system effectively captures users’ depressive signs and shows promise for mental health support applications.

External IDs:dblp:conf/ecir/BaoPP25