{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"id":[{"id":"https:\/\/ror.org\/03mb6wj31","id-type":"ROR","asserted-by":"publisher"},{"id":"https:\/\/www.isni.org\/000000041937028X","id-type":"ISNI","asserted-by":"publisher"},{"id":"https:\/\/www.wikidata.org\/entity\/Q1640731","id-type":"wikidata","asserted-by":"publisher"}],"name":"Universitat Polit\u00e8cnica de Catalunya","acronym":["UPC"]}],"indexed":{"date-parts":[[2026,1,30]],"date-time":"2026-01-30T18:34:17Z","timestamp":1769798057024,"version":"3.49.0"},"reference-count":0,"publisher":"Universitat Polit\u00e8cnica de Catalunya","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"abstract":"<jats:p>Current recommender systems aim mainly to generate accurate item recommendations, without properly evaluating the multiple dimensions of the recommendation problem. However, in many domains, like in music, where items are rarely consumed in isolation, users would rather need a set of items, designed to work well together, while having some cognitive properties as a whole, related to their perception of quality and satisfaction.\r\nIn this thesis, a hybrid case-based recommendation approach for item collections is proposed. In particular, an application to automatic playlist continuation, addressing similar cognitive concepts, rather than similar users, is presented. Playlists, that are sets of music items designed to be consumed as a sequence, with a specific purpose and within a specific context, are treated as cases. The proposed recommender system is based on a meta-level hybridization. First, Latent Dirichlet Allocation is applied to the set of past playlists, described as distributions over music styles, to identify their underlying concepts. Then, for a started playlist, its semantic characteristics, like its latent concept and the styles of the included items, are inferred, and Case-Based Reasoning is applied to the set of past playlists addressing the same concept, to construct and recommend a relevant playlist continuation. A graph-based item model is used to overcome the semantic gap between songs\u2019 signal-based descriptions and users\u2019 high-level preferences, efficiently capture the playlists\u2019 structures and the similarity of the music items in those. As the proposed method bases its reasoning on previous playlists, it does not require the construction of complex user profiles to generate accurate recommendations. Furthermore, apart from relevance, support to parameters beyond accuracy, like increased coherence or support to diverse items is provided to deliver a more complete user experience.\r\nExperiments on real music datasets have revealed improved results, compared to other state of the art techniques, while achieving a \u201cgood trade-off\u201d between recommendations\u2019 relevance, diversity and coherence. Finally, although actually focusing on playlist continuations, the designed approach could be easily adapted to serve other recommendation domains with similar characteristics.<\/jats:p>\n                <jats:p>Los sistemas de recomendaci\u00f3n actuales tienen como objetivo principal generar recomendaciones precisas de art\u00edculos, sin evaluar propiamente las m\u00faltiples dimensiones del problema de recomendaci\u00f3n. Sin embargo, en dominios como la m\u00fasica, donde los art\u00edculos rara vez se consumen en forma aislada, los usuarios m\u00e1s bien necesitar\u00edan recibir recomendaciones de conjuntos de elementos, dise\u00f1ados para que se complementaran bien juntos, mientras se cubran algunas propiedades cognitivas, relacionadas con su percepci\u00f3n de calidad y satisfacci\u00f3n. En esta tesis, se propone un sistema h\u00edbrido de recomendaci\u00f3n meta-nivel, que genera recomendaciones de colecciones de art\u00edculos. En particular, el sistema se centra en la generaci\u00f3n autom\u00e1tica de continuaciones de listas de m\u00fasica, tratando conceptos cognitivos similares, en lugar de usuarios similares. Las listas de reproducci\u00f3n son conjuntos de elementos musicales dise\u00f1ados para ser consumidos en secuencia, con un prop\u00f3sito espec\u00edfico y dentro de un contexto espec\u00edfico. El sistema propuesto primero aplica el m\u00e9todo de Latent Dirichlet Allocation a las listas de reproducci\u00f3n, que se describen como distribuciones sobre estilos musicales, para identificar sus conceptos. Cuando se ha iniciado una nueva lista, se deducen sus caracter\u00edsticas sem\u00e1nticas, como su concepto y los estilos de los elementos incluidos en ella. A continuaci\u00f3n, el sistema aplica razonamiento basado en casos, utilizando las listas del mismo concepto, para construir y recomendar una continuaci\u00f3n relevante. Se utiliza un grafo que modeliza las relaciones de los elementos, para superar el ?salto sem\u00e1ntico? existente entre las descripciones de las canciones, normalmente basadas en caracter\u00edsticas sonoras, y las preferencias de los usuarios, expresadas en caracter\u00edsticas de alto nivel. Tambi\u00e9n se utiliza para calcular la similitud de los elementos musicales y para capturar la estructura de las listas de dichos elementos. Como el m\u00e9todo propuesto basa su razonamiento en las listas de reproducci\u00f3n y no en usuarios que las construyeron, no se requiere la construcci\u00f3n de perfiles de usuarios complejos para poder generar recomendaciones precisas. Aparte de la relevancia de las recomendaciones, el sistema tiene en cuenta par\u00e1metros m\u00e1s all\u00e1 de la precisi\u00f3n, como mayor coherencia o soporte a la diversidad de los elementos para enriquecer la experiencia del usuario. Los experimentos realizados en bases de datos reales, han revelado mejores resultados, en comparaci\u00f3n con las t\u00e9cnicas utilizadas normalmente. Al mismo tiempo, el algoritmo propuesto logra un \"buen equilibrio\" entre la relevancia, la diversidad y la coherencia de las recomendaciones generadas. Finalmente, aunque la metodolog\u00eda presentada se centra en la recomendaci\u00f3n de continuaciones de listas de reproducci\u00f3n musical, el sistema se puede adaptar f\u00e1cilmente a otros dominios con caracter\u00edsticas similares.<\/jats:p>","DOI":"10.5821\/dissertation-2117-125310","type":"dissertation","created":{"date-parts":[[2023,9,13]],"date-time":"2023-09-13T01:25:13Z","timestamp":1694568313000},"approved":{"date-parts":[[2018,11,23]]},"source":"Crossref","is-referenced-by-count":0,"title":["A hybrid approach for item collection recommendations : an application to automatic playlist continuation"],"prefix":"10.5821","author":[{"sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Anna","family":"Gkatzioura","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"3865","container-title":[],"original-title":[],"deposited":{"date-parts":[[2026,1,30]],"date-time":"2026-01-30T06:40:33Z","timestamp":1769755233000},"score":1,"resource":{"primary":{"URL":"https:\/\/hdl.handle.net\/2117\/125310"}},"subtitle":[],"editor":[{"given":"Miquel","family":"S\u00e0nchez-Marr\u00e8","sequence":"first","affiliation":[],"role":[{"role":"editor","vocabulary":"crossref"}]}],"short-title":[],"issued":{"date-parts":[[null]]},"references-count":0,"URL":"https:\/\/doi.org\/10.5821\/dissertation-2117-125310","relation":{},"subject":[]}}