• The project, developed by research staff at the ICT center of the Universidade da Coruña, has been awarded by the Spanish Research Network on Recommender Systems (ELIGE-IA) as the best scientific publication of the year in this field.
• Recommender systems suggest personalized content, but they are often opaque and energy-intensive. CITIC’s proposal uses images to make them more understandable and reduces the model size by 98% and CO₂ emissions by 75%

A Coruña, June 2, 2025.– Every day, millions of digital platforms present us with suggestions: that movie we might love, the song that fits our mood, or the perfect product for our next purchase. Behind these recommendations are intelligent systems that analyze our tastes and behaviors to anticipate what we might want. However, these systems operate like black boxes: they make predictions without explaining how or why, and they require vast computational resources that impact the environment.

To change this reality, a group of researchers from CITIC at the Universidade da Coruña—a center that is part of the CIGUS Network of the Xunta de Galicia, which certifies the quality and impact of its research—has developed BRIE, a model that uses intelligently selected images to explain the recommendations we receive. This helps users better understand suggestions while drastically reducing energy consumption.

The work was recently recognized with the Best Scientific Publication Award in Recommender Systems 2023–2024, granted by the Spanish Research Network on Recommender Systems (ELIGE-IA), which brings together the country’s leading experts in this field. This recognition highlights the most relevant national research work in the area for the 2023–2024 period.

The award-winning article, published in the journal Information Fusion, is titled “Sustainable transparency on recommender systems: Bayesian ranking of images for explainability” and was authored by CITIC researchers Jorge Paz Ruza, Amparo Alonso Betanzos, Bertha Guijarro Berdiñas, Brais Cancela Barizo, and Carlos Eiras-Franco.

What Are Digital Recommendations?

When a platform suggests a movie, a song, a product, or a post, it is using a recommender system. These technologies influence many of our online decisions: what we watch on Netflix, listen to on Spotify, buy on Amazon, or read on social media.

To generate these suggestions, they analyze our preferences and habits to predict what might interest us. Although they are useful tools, their inner workings are often opaque: they don’t explain the reasons behind their recommendations, which can lead to mistrust. Moreover, training these models requires large amounts of energy.

To offer a more understandable and efficient alternative, the CITIC team has developed BRIE, an innovative model that uses intelligently selected images to illustrate the reasoning behind each recommendation. For example, if a platform suggests a product, BRIE can display similar images of items the user previously liked, making the system easier to interpret.

What stands out most is BRIE’s unprecedented efficiency: it reduces the model size by 98% and the CO₂ emissions associated with training by 75%, without sacrificing accuracy or performance. This makes BRIE not only a more transparent tool but also a more sustainable one.

The model has been successfully validated on six real-world datasets and outperforms the most advanced current systems.

National Network ELIGE-IA

This achievement has been recognized by the Spanish Research Network on Recommender Systems (ELIGE-IA), which brings together the main scientific groups in the country in this field. The ELIGE-IA Network (RED2022-134302-T) is funded by the Ministry of Science and Innovation and the State Research Agency, as part of the research network grants within the Spanish State Plan for Scientific, Technical and Innovation Research 2021–2023.

Each year, the network awards a prize for the best scientific publication in the field, evaluating the innovation, social impact, and quality of the submitted work. In this edition, the committee selected CITIC’s work from among eight high-level candidates.

Ethical, Efficient, and Explainable AI

The award-winning project is part of two key CITIC initiatives: FrugalAI, which develops AI algorithms capable of operating in resource-limited environments such as the Internet of Things or robotics; and EthicDL, which seeks to combine the power of deep learning with more interpretable methods to build systems that are more understandable and trustworthy.

About CITIC

CITIC is a research center promoting advancement and excellence in applied R&D&I in ICTs. It was founded in 2008 by the Universidade da Coruña. The center’s scientific activity is structured into four main research areas: Artificial Intelligence; Data Science and Engineering; High-Performance Computing; and Intelligent Networks and Services, as well as a cross-disciplinary area: Cybersecurity.

CITIC is accredited as a Center of Excellence and a member of the CIGUS Network for the period 2024–2027, which certifies the quality and impact of its research. The accreditation, development, and enhancement of CITIC is co-financed by the Xunta de Galicia and 60% by the European Union through the FEDER Galicia 2021–2027 Operational Program, under the thematic objective of promoting “a smarter Europe: innovative and smart economic transformation” (ED431G 2023/01).