Fecha
2026-06-01
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial
Resumen
Los sistemas de recomendación se han consolidado como una herramienta fundamental para gestionar la sobrecarga de información a la que se enfrentan los usuarios en plataformas digitales con grandes catálogos. En el dominio de las apuestas deportivas, esta funcionalidad cobra particular relevancia debido a la cantidad y diversidad de eventos disponibles y a sus particularidades específicas: la ventana de validez corta de los eventos recomendables, la renovación continua del catálogo a lo largo del calendario deportivo, la baja densidad de interacción por evento y la naturaleza implícita de la señal de preferencia.
Este trabajo aborda el problema de recomendar eventos deportivos sobre los que apostar a los usuarios de un casino o casa de apuestas online. Se han seleccionado, implementado y evaluado ocho métodos representativos de las principales familias de la literatura, desde los recomendadores de referencia hasta una arquitectura híbrida que combina factorización matricial y aprendizaje profundo, pasando por los métodos basados en contenido y los basados en filtrado colaborativo. Como punto de partida se ha construido un dataset a partir del conocimiento obtenido de la actividad real de una plataforma de juego online y se ha derivado una valoración implícita comparable entre usuarios a partir de la actividad económica de cada apuesta. La evaluación se ha realizado mediante una estrategia offline sobre dos conjuntos de test diseñados para representar perfiles de usuario diferenciados, utilizando un conjunto de doce métricas que cubren tanto la calidad predictiva como el comportamiento del recomendador.
Los resultados del análisis comparativo identifican los métodos de factorización matricial y el enfoque híbrido como las opciones más equilibradas para este dominio, tanto por su rendimiento de ranking como por su robustez frente a poblaciones de usuarios heterogéneas.
Además, se han identificado varios compromisos estructurales entre objetivos (precisión frente a exploración del catálogo, diversidad colaborativa frente a semántica, sofisticación arquitectónica frente a eficacia) que ningún método consigue resolver simultáneamente. A partir de los resultados se formula una recomendación práctica orientada a una posible implementación en producción y se identifican las direcciones de profundización que esa implementación requeriría.
Recommender systems have become an essential tool for managing the information overload faced by users on digital platforms with large catalogs. In the domain of sports betting on online casinos, this functionality is particularly relevant due to the quantity and diversity of available events and to the domain’s specific particularities: the short validity window of recommendable events, the continuous renewal of the catalog throughout the sports calendar, the low interaction density per event, and the implicit nature of the preference signal. This Master’s Thesis addresses the problem of recommending sports events to bet on for the users of an online casino or sportsbook. Eight recommender methods representative of the main families in the literature have been selected, implemented and evaluated, ranging from baseline recommenders to a hybrid architecture combining matrix factorization and deep learning, including content-based and collaborative filtering methods. As a starting point, a dataset has been built from the knowledge gained through real activity on an online gaming platform, and an implicit rating comparable across users has been derived from the economic activity associated with each bet. The evaluation has been carried out following an offline strategy on two test sets designed to represent different user profiles, using a set of twelve metrics covering both predictive quality and recommender behavior. The results of the comparative analysis identify matrix factorization methods and the hybrid approach as the most balanced options for this domain, both in terms of ranking performance and robustness against heterogeneous user populations. Furthermore, several structural trade-offs between objectives have been identified (precision versus catalog exploration, collaborative versus semantic diversity, architectural sophistication versus effectiveness) that no method manages to resolve simultaneously. Based on these results, a practical recommendation is formulated towards a potential production implementation, and the directions of further refinement that such implementation would require are identified.
Recommender systems have become an essential tool for managing the information overload faced by users on digital platforms with large catalogs. In the domain of sports betting on online casinos, this functionality is particularly relevant due to the quantity and diversity of available events and to the domain’s specific particularities: the short validity window of recommendable events, the continuous renewal of the catalog throughout the sports calendar, the low interaction density per event, and the implicit nature of the preference signal. This Master’s Thesis addresses the problem of recommending sports events to bet on for the users of an online casino or sportsbook. Eight recommender methods representative of the main families in the literature have been selected, implemented and evaluated, ranging from baseline recommenders to a hybrid architecture combining matrix factorization and deep learning, including content-based and collaborative filtering methods. As a starting point, a dataset has been built from the knowledge gained through real activity on an online gaming platform, and an implicit rating comparable across users has been derived from the economic activity associated with each bet. The evaluation has been carried out following an offline strategy on two test sets designed to represent different user profiles, using a set of twelve metrics covering both predictive quality and recommender behavior. The results of the comparative analysis identify matrix factorization methods and the hybrid approach as the most balanced options for this domain, both in terms of ranking performance and robustness against heterogeneous user populations. Furthermore, several structural trade-offs between objectives have been identified (precision versus catalog exploration, collaborative versus semantic diversity, architectural sophistication versus effectiveness) that no method manages to resolve simultaneously. Based on these results, a practical recommendation is formulated towards a potential production implementation, and the directions of further refinement that such implementation would require are identified.
Descripción
Categorías UNESCO
Palabras clave
Sistemas de recomendación, Apuestas deportivas, Ítems efímeros, Catálogo dinámico, Valoración implícita, Evaluación offline, Recommender systems, Sports betting, Ephemeral items, Dynamic catalog, Implicit rating, Offline evaluation
Citación
Rasero Jiménez, Manuel Ángel. Trabajo Fin de Máster: Sistemas de recomendación de apuestas deportivas en casinos online: análisis de métodos y enfoques. Universidad Nacional de Educación a Distancia (España) 2026.
Centro
E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial



