Jorge Gabín, Anxo Pérez and Javier Parapar study in-context learning under a fixed token budget. Many short examples beat a few long ones, and filling the budget with concise demonstrations outperforms longer demonstrations and random sampling. The paper will be presented in Athens in March 2027.
Masterclass for psychologists on AI and early risk detection
Anxo Pérez taught "AI in Psychology and Mental Health, from the prompt to early risk detection" in the series run by the Galician and Portuguese psychology associations, online for up to 500 professionals.
Talk on AI and mental health for Healthy October
Eliseo Bao and Anxo Pérez gave the talk "Artificial Intelligence and Mental Health: What Can Social Media Tell Us?" at the Faculty of Computer Science, part of the 12th Healthy October programme at the Universidade da Coruña.
Two papers on hate speech at EMNLP 2026
Paloma Piot presented two papers in Budapest: one on whether persona-conditioned LLMs capture how different demographic groups perceive hateful content, and a bilingual English–German dataset of 16,000 posts for explicit and implicit dehumanisation, built with GESIS. She also co-organised the Workshop on Online Abuse and Harms (WOAH).
I will be presenting two papers on Tuesday, 27 October, from 09:00 to 10:30 (Poster Session F): From Self to Other: Evaluating Demographic Perspective-Taking in LLM Hate Speech Annotation w/ Javier Parapar Beyond the Explicit: A Bilingual Dataset for Dehumanization Detection in…
Mentors at DataForum Justicia 2026
Eliseo Bao and Javier Parapar mentored student teams from universities across Spain in the Datathon, whose three challenges were digital accessibility to the justice system, bringing justice closer to children and young people, and the needs of neurodivergent people.
Explainable text analysis for digital health at Pinchos&Datos
Javier Parapar spoke on large-scale text analysis with IR, NLP and LLMs for explainable conclusions in digital health, at the first Pinchos&Datos in A Coruña with DXC Technology and Hijos de Rivera.
Paloma Piot defended “Explainable and Efficient Language Models for Hate Speech Detection”, supervised by Javier Parapar and Patricia Martín Rodilla, before a committee of Paolo Rosso, Elisabetta Fersini and David E. Losada. She continues in the group as a postdoctoral researcher.
Two new projects start in 2026. XMED (Spanish State Research Agency) works on explainable and factual health search, and HERITWIN (Horizon Europe) on cultural heritage digital twins.
Main Conference: "From Self to Other: Evaluating Demographic Perspective-Taking in LLM Hate Speech Annotation" w/ Javier Parapar We analyse whether persona-conditioned LLMs can capture differences in how people from different demographic groups perceive the same hateful content,…
Five papers at SIGIR 2026
The group presented five papers at ACM SIGIR 2026, on depression symptom evidence, hybrid pooling with LLMs, query suggestion evaluation, RAG reproducibility and faithful question answering for academic search.
Retrieval Lab, Universidade da Coruña and CITIC Centro de Investigacion TIC - Universidade da Coruña, presenting two of our latest research papers! Here is a look at what we presented: Lost in the Evidence?
More reliable LLM reasoning for depression assessment
Eliseo Bao, Anxo Pérez and Javier Parapar published in Science Progress how structured generation and GRPO training make the reasoning of language models more clinically reliable when they assess depression symptoms.
We explored this idea for depression symptom detection, and the results were really promising! Yesterday, David Otero and I presented our work with Miguel Anxo Pérez Vila and Javier Parapar, titled "Learning Evidence of Depression Symptoms via Prompt Induction," at ACM SIGIR…
A master's thesis from the group at the Innovación e Talento Novo DemoDay
A master's thesis co-supervised by Paloma Piot and Anxo Pérez was selected for the DemoDay of the Innovación e Talento Novo programme.
I'll be presenting our paper: "Personalisation or Prejudice? Addressing Geographic Bias in Hate Speech Detection using Debias Tuning in Large Language Models" w/ Patricia Martín-Rodilla and Javier Parapar (IRLab, UDC - Information Retrieval Lab, Universidade da Coruña, HYBRIDS)…
Invited talk at UNIR on explainable depression detection
Eliseo Bao presented the group's work on explainable AI for detecting depression symptoms on social media, and the MindWell conversational agent, at the International University of La Rioja.
ColBERT for keyphrase search, in Information Processing & Management
Jorge Gabín, Javier Parapar and Craig Macdonald (University of Glasgow) show how to adapt ColBERT to the short keyphrase queries people actually type, beyond the natural-language questions it was trained on.
Explainable depression symptom detection at the Caen symposium
Eliseo Bao presented the group's work on explainable detection of depression symptoms in social media at the 2nd Symposium on Mental Health and Artificial Intelligence, Université de Caen Normandie.
2025
ReDSM5, a Reddit dataset for DSM-5 depression detection, at CIKM 2025
Eliseo Bao presented ReDSM5 in Seoul: Reddit posts annotated by a psychologist for the nine DSM-5 depression symptoms, with the clinical rationale for each label. The same month, Online Social Networks and Media published our work on modelling depression language with relevance-based language models.
Anxo Pérez wins the SEPLN award for the best PhD thesis of 2024
The Spanish Society for Natural Language Processing recognised his thesis on estimating depression severity on the Internet, supervised by Álvaro Barreiro and Javier Parapar.
Eliseo Bao spent October to December 2025 with the Natural Language Processing Group at the University of Sheffield, hosted by Xi Wang, working on simulating users with LLM personas.
SIGIR 2025 Doctoral Consortium
Eliseo Bao presented his PhD plan, "Towards Explainable and Safe Systems for Health Data", at the SIGIR 2025 Doctoral Consortium in Padua.
Five contributions at ECIR 2025
The group presented five contributions at the European Conference on Information Retrieval, including the MindWell demo for mental health monitoring.