Personalized Music Recommendations Using Retrieval Augmented Generation: Evaluating Data Processing Strategies on Spotify Data

In recent years, music consumption has transitioned from physical formats to streaming platforms such as Spotify. Spotify, one of the leading platforms, has seen remarkable growth, surpassing 574 million active users globally by 2024. This shift has made music more accessible than ever before, but as the volume of available content increases, discovering personalized music recommendations has become increasingly complex. To address this challenge, the development of AI-powered generative recommendation systems has become essential. This research investigates the application of Retrieval-Augmented Generation (RAG) to provide personalized music recommendations by utilizing the Spotify dataset. We evaluate the impact of data preprocessing, converting numerical data into categorical data, on system performance. The RAG system, powered by OpenAI’s text-embedding-ada-002 model, processes the data, stores embeddings in Chroma DB, and generates recommendations using GPT-4o-mini. Experimental results show that the converted dataset improves efficiency, reducing both processing time and token usage. User feedback from 43 participants reveals an 80% satisfaction rate, demonstrating the effectiveness and user appeal of the proposed RAG-based music recommendation system.