Use this link to cite:
https://hdl.handle.net/2183/46458 Análisis de sostenibilidad de sistemas de recomendación
Loading...
Identifiers
Publication date
Authors
Patiño Martínez, Xesús
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen]: Los sistemas de recomendación se han consolidado en nuestra sociedad como herramientas fundamentales en numerosos sectores como el entretenimiento, el turismo, los servicios financieros, etc. Estos sistemas proporcionan una buena personalización de contenidos, mejorando así la experiencia de usuario en plataformas digitales. Sin embargo, su desarrollo y entrenamiento implican un elevado consumo de recursos computacionales, es decir, un impacto medioambiental alto debido al consumo energético y las emisiones de carbono producidas. En este trabajo se presenta un análisis de la sostenibilidad de distintos sistemas de recomendación, evaluando tanto su rendimiento predictivo como su huella de carbono bajo diferentes escenarios experimentales. Se utilizan métricas de sostenibilidad como emisiones de CO2, uso de CPU/GPU y memoria RAM, junto con métricas de rendimiento como el error cuadrático media, la precisión o el recall, para comparar de forma analítica varias combinaciones de modelos, métodos de entrenamiento y conjuntos de datos. Los resultados obtenidos permiten identificar las configuraciones más eficientes y sostenibles, proporcionando así una ayuda práctica para el desarrollo de sistemas de recomendación más sostenibles y ecológicos.
[Abstract]: Recommender systems have become established in our society as fundamental tools in numerous sectors such as entertainment, tourism, financial services, etc. These systems offer effective content personalization, thereby enhancing the user experience on digital platforms. However, the development and training of these systems involve a high consumption of computational resources, that is, a high environmental impact due to energy consumption and the carbon emissions produced. This work presents an analysis of the sustainability of various recommender systems, evaluating both their predictive performance and carbon footprint under different experimental scenarios. Sustainability metrics such as CO2 emissions, CPU/GPU, and RAM usage are used, together with performance metrics such as mean squared error, precision, or recall, to analytically compare various combinations of models, training methods, and datasets. The results obtained allow for the identification of the most efficient and sustainable configurations, thus providing practical guidance for the development of more sustainable and eco-friendly recommender systems.
[Abstract]: Recommender systems have become established in our society as fundamental tools in numerous sectors such as entertainment, tourism, financial services, etc. These systems offer effective content personalization, thereby enhancing the user experience on digital platforms. However, the development and training of these systems involve a high consumption of computational resources, that is, a high environmental impact due to energy consumption and the carbon emissions produced. This work presents an analysis of the sustainability of various recommender systems, evaluating both their predictive performance and carbon footprint under different experimental scenarios. Sustainability metrics such as CO2 emissions, CPU/GPU, and RAM usage are used, together with performance metrics such as mean squared error, precision, or recall, to analytically compare various combinations of models, training methods, and datasets. The results obtained allow for the identification of the most efficient and sustainable configurations, thus providing practical guidance for the development of more sustainable and eco-friendly recommender systems.
Description
Keywords
Sistemas de recomendación Sostenibilidad Huella de carbono Impacto ambiental Conjuntos de datos: MovieLens, Netflix, GDSC, CTRPv2, Kiva Microloans, IMF DOTS Factorización matricial Redes neuronales profundas Modelos de redes de grafos Aprendizaje automático Aprendizaje profundo CodeCarbon Recommender Systems Sustainability Carbon Footprint Environmental Impact Datasets Matrix Factorization Deep Neural Networks Graph-based Models Machine Learning Deep Learning
Editor version
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International








