Detecting depression through language: from social media to clinical practice

Foto de Anxo

Date and time: 20/11/25, 5pm CET

Speaker: Anxo Vila Pérez, Departamento de Ciencia de la Computación e Intelixencia Artificial, CITIC e Information Retrieval Lab (IRLab), Universidade da Coruña

Presenter: 

Abstract:

In recent years, the automatic analysis of language on social media has become an additional tool for understanding the signs of depression and other mental disorders. However, this task poses significant methodological, ethical, and clinical challenges: how can we detect emotional signals in everyday, noisy, and changing language? How can we guarantee interpretability and trust in the models?

In this talk, we will explore the most recent methods and trends in detecting depression using NLP, from the informal language of social media to the texts and questionnaires of clinical practice. We will also discuss open challenges: biases, clinical validation, differences between symptoms, and the cost of annotation (including LLMs as support in human-in-the-loop systems). We will close with the recent trend of using "LLM personas": simulating profiles and conversations for screening and support. Their potential is high (case coverage, interactive assessment), but it demands actionable explainability, traceable evidence, and control of hallucinations, with a clinician-in-the-loop.

Bio: Miguel Anxo Pérez Vila is an Assistant Professor in the Department of Computer Science and Artificial Intelligence at the Universidade da Coruña (UDC) and a researcher at CITIC and the Information Retrieval Lab (IRLab). He previously completed a predoctoral research visit at the UKP Lab (TU Darmstadt) under the supervision of Prof. Iryna Gurevych; this work led to a paper at EMNLP. In January 2024, he was awarded a PhD in Computer Science from the UDC. His research lies at the intersection of Information Retrieval, Natural Language Processing, and Machine Learning, with a particular focus on digital mental health, specifically on symptom modelling and explainable methods for early risk detection on social media and in conversational environments. Since 2025, he has co-organised the eRisk Lab (CLEF eRisk) alongside Javier Parapar, Xi Wang, and Fabio Crestani, contributing resources and evaluation protocols that are widely used in the community. Anxo has published at conferences such as EMNLP, SIGIR, CIKM, and ECIR, and in journals such as *Artificial Intelligence in Medicine* and the *Journal of Healthcare Informatics Research*. He has also worked in technology transfer and intellectual property, notably the registration of ANGUS (a music similarity algorithm based on onsets). In 2025, his doctoral thesis received the SEPLN Best Doctoral Thesis Award. His current research interests include interpretable classification and the study of LLMs to provide clinically meaningful solutions for mental health applications.

Registration (mandatory): https://zoom.us/webinar/register/WN_bIZyUU2cRhSlMR8DA8IBjA

Link to the talk