Diseño y desarrollo de una aplicación web para la recomendación y el descubrimiento de música.
Loading...
Official URL
Full text at PDC
Publication date
2026
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
License

Citation
Abstract
Este trabajo nace de la insatisfacción con los sistemas de recomendación musical en la actualidad. La mayoría de usuarios recibe propuestas de la misma aplicación de streaming que utiliza para escuchar música y esto les provoca a menudo una sensación de estancamiento, así como de que se priorizan los intereses comerciales de la industria discográfica frente a sus gustos reales. track1 es una aplicación full-stack desarrollada con Next.js que permite a los usuarios registrar y valorar los álbumes que escuchan, compartir sus gustos con otros y recibir recomendaciones basadas en ellos a través de una interfaz cuidada y atractiva.
El sistema de recomendaciones que se encarga de generar las propuestas se apoya en una base de datos orientada a grafos de Neo4j sobre la cual se aplica la técnica de Generación Aumentada por Recuperación (RAG), que ofrece al LLM (gpt-4o-mini de OpenAI) un contexto acotado sobre el que decidir la propuesta. De esta forma, selecciona y argumenta la propuesta final entre un conjunto limitado de candidatos, mitigando el riesgo de alucinaciones.
El desarrollo ha consistido en una fase previa de investigación y dos de desarrollo iterativo, cada una culminada con evaluaciones con usuarios para validar el proyecto. Así, se ha obtenido como resultado una aplicación funcional cuya arquitectura y hallazgos se documentan en la presente memoria.
This project grows out of the dissatisfaction with current music recommending systems. Most users receive suggestions from the same streaming application they use to listen to music, which often leads to a sense of standstil and the impression that the commercial interests of the record industry are prioritized over their actual preferences. track2 is a full-stack application developed with Next.js that enables users to log and rate the albums they listen to, share their tastes with others and receive recommendations based on those tastes through a refined and engaging interface. The recommendation system responsible for generating these proposals relies on a Neo4j graph database, combined with a Retrieval-Augmented Generation (RAG) approach that provides the LLM (OpenAI's gpt-4o-mini) with a constrained context from which to select the proposal. In this way, it makes a choice and justifies it from a limited set of candidates, mitigating the risk of hallucinations. The development process has consisted on an initial research phase followed by two iterative development cycles, each concluding with user evaluations to validate the project. Thus, the result is a functional application whose architecture and findings are documented in this dissertation.
This project grows out of the dissatisfaction with current music recommending systems. Most users receive suggestions from the same streaming application they use to listen to music, which often leads to a sense of standstil and the impression that the commercial interests of the record industry are prioritized over their actual preferences. track2 is a full-stack application developed with Next.js that enables users to log and rate the albums they listen to, share their tastes with others and receive recommendations based on those tastes through a refined and engaging interface. The recommendation system responsible for generating these proposals relies on a Neo4j graph database, combined with a Retrieval-Augmented Generation (RAG) approach that provides the LLM (OpenAI's gpt-4o-mini) with a constrained context from which to select the proposal. In this way, it makes a choice and justifies it from a limited set of candidates, mitigating the risk of hallucinations. The development process has consisted on an initial research phase followed by two iterative development cycles, each concluding with user evaluations to validate the project. Thus, the result is a functional application whose architecture and findings are documented in this dissertation.
Description
Link a la app desplegada: track-tfg.netlify.app










