Research
Projects
X-MED: eXplainable and Factual Models for Health Information Access: Clinically-grounded models, ...

X-MED: eXplainable and Factual Models for Health Information Access: Clinically-grounded models, factual & user-adaptive information access

X-MED aims to advance trustworthy, transparent, and evidence-based access to digital health information by integrating methods from Natural Language Processing, Information Retrieval, Clinical Informatics, and Computational Psychology. The project addresses a central societal challenge: citizens and health professionals increasingly rely on digital systems to obtain medical information, yet current tools often provide incomplete, poorly grounded, or misleading content. This problem is exacerbated by the prevalence of health misinformation, unequal levels of health literacy, and the limited transparency of modern AI models.

The project develops clinically-grounded and explainable language models capable of generating and interpreting health information with explicit links to validated medical evidence. It also designs retrieval and ranking algorithms that prioritise factual reliability over superficial relevance, ensuring that retrieved documents and generated answers are anchored in credible, verifiable sources. X-MED introduces new evaluation frameworks and benchmarks that measure reliability dimensions such as factuality, evidence sufficiency, and clinical validitybeyond traditional relevance-based metrics.

A distinctive component of the project is its focus on psychology-informed and user-adaptive information access. By modelling cognitive load, personality traits, and psychological vulnerability, X-MED will develop adaptive presentation strategies that enhance comprehension and reduce susceptibility to misinformation among diverse user groups.

Objectives

Advance the scientific foundations of trustworthy health-information access by integrating factuality evaluation, clinical grounding, explainability, and psychology-informed adaptation into unified NLP and IR methodologies.

Develop robust, transparent, and evidence-based language and retrieval models that can support safe, equitable, and reliable access to health information for both citizens and health professionals.

Establish evaluation methodologies, datasets, and benchmarks that capture reliability dimensions beyond relevance, enabling systematic assessment of factuality, interpretability, and source trustworthiness in health-related systems.

Mitigate the societal impact of health misinformation by designing models capable of detecting unreliable content, filtering harmful sources, and presenting verified evidence in an accessible and psychologically aligned manner.