IRLab

Happy to share that the @IRLab_UDC team will present the following works at #SIGIR2026 🎉 Great work by everyone involved, very proud of these contributions! Abstracts below👇

👉 In “Hybrid Pooling with LLMs via Relevance Context Learning” (with @davidoterof), we turn examples into explicit, topic-aware relevance narratives, rethinking how LLMs judge relevance. #SIGIR2026

👉In “Beyond Top-e: Simulation-Based Interactive Evaluation for Query Suggestions” (with @JorgeGabin_ & @wangxieric ), we move from static assumptions to interactive simulation, using LLMs to model how users actually select query suggestions. #SIGIR2026

👉In “Lost in the Evidence? Reproducing Document Position and Context Size Effects in RAG” (with @JorgeGabin_ & @anxopvila ), we revisit position and context effects in RAG, highlighting the role of evaluation design and reproducibility. #SIGIR2026

👉In “Learning Evidence of Depression Symptoms via Prompt Induction”, (with Eliseo Bao, @anxopvila and @davidoterof ) we teach LLMs to detect symptoms by turning examples into clear, reusable guidelines. #SIGIR2026

👉 Finally, in “Faithful LLM-Assisted Question Answering for Technical and Academic Search” @jlfilgueiras explores how to design LLM-based search systems that provide grounded, verifiable, and up-to-date answers. #SIGIR2026

Today we will be presenting our work "PartisanLens: A Multilingual Dataset of Hyperpartisan and Conspiratorial Immigration Narratives in European Media". 📍 Poster Board #41 🕓 16:30–18:00 📌 Poster Hall 🤝 If you're around, come say hi! #EACL2026 #NLP

Una reflexión personal: ¿y si la IA pudiera detectar a tiempo señales de malestar emocional en lo que escribimos? 👇 🔗theconversation.com/cuando-la-ia-e…

Paper accepted at #LREC2026 🎉 Mute Cods: A Multilingual Telegram Dataset with Benchmark Models for Conspiracy Theory Detection 🌐 w/ Katarina Laken, Erik Marino, Davide Bassi, Søren Fomsgaard, Michele Maggini, Renata Vieira, Marcos Garcia & Sara Tonelli See you in Palma! ☀️📚

🥳 @itspalomap paper "Can LLMs Evaluate What They Cannot Annotate? Revisiting LLM Reliability in Hate Speech Detection" has been accepted at #LREC2026 w/ @davidoterof & @patrimrodilla See you in Palma 🇪🇸✈️ 🔗arxiv.org/abs/2512.09662 📀github.com/palomapiot/hat…

Hoy, en el #SaferInternetDay, reafirmamos nuestro compromiso: investigamos para detectar y frenar mensajes de odio y conductas abusivas, y para identificar señales de riesgo en redes sociales, con rigor y responsabilidad. Por un internet más seguro.

🎉 Our paper “PartisanLens: A Multilingual Dataset of Hyperpartisan and Conspiratorial Immigration Narratives in European Media” accepted at #EACL2026 🇲🇦 main conference. Work done with Michele Maggini @itspalomap @anxopvila and the @HYBRIDS_MSCA team! See you in Rabat! ✈️📍

Happy to see this work with @davidoterof accepted at #ECIR2026. A deceptively simple yet effective idea: improve PRF through selective relevance feedback using small LLMs, denoising the feedback set without polluting or overcomplicating the underlying PRF model.

Encantada de participar na xornada “Onde queda a ética cando falamos de IA?” que tivo lugar o pasado luns 1 de decembro no CIE Mans. Foi un pracer compartir reflexións e aprendizaxes nun ambiente tan enriquecedor ✨ 📸 Fotos: ESTUDIOS MANS

O luns 1 de decembro no CIE Mans terá lugar a xornada “Onde queda a ética cando falamos de IA?”. @itspalomap achegará unha visión investigadora sobre o impacto positivo e negativo da IA na sociedade. Máis info e inscricións 👉 manspaideia.com/xornada-inteli…

📢 ¡Nueva charla invitada del ciclo #CSEPLN de la #SEPLN! 🗓️ 20/11/25, Anxo Vila Pérez (U. A Coruña) con la ponencia: 🗣️ Detectando la depresión a través del lenguaje: de las redes sociales a la práctica clínica. Registro 👉 zoom.us/webinar/regist… #NLProc #LLMs #eRISK

Mañana @anxopvila participa en los webminars de la @sepln con una charla donde explicará los retos de la detección de depresión en redes sociales a través del lenguaje. Más información y registro aquí 👉 sepln.org/actualidad/eve…

🎓 Happy to share our latest work: “Beyond questions: Leveraging ColBERT for keyphrase search”. A study on improving dense retrieval for keyphrase-style queries using ColBERTKP. 🔗sciencedirect.com/science/articl… With @jparapar & @craig_macdonald

This week @wangxieric & @eliseobao presented our works at CIKM 2025. We unveiled: TalkDep: Clinically Grounded LLM Personas for Conversation-Centric Depression Screening and ReDSM5: A Reddit Dataset for DSM-5 Depression Detection. Great work! #CIKM2025 w/ @anxopvila

Just published @itspalomap's agentic software for supporting hate speech moderation, WATCHED is a Web Agent that combines LLMs, BERT-based classification, and explainable reasoning to assist moderators in identifying and understanding harmful content. 🔗 authors.elsevier.com/sd/article/S23…