Yambda Music Recommendation Dataset

Yambda-5B is a large-scale multimodal music analysis dataset released by the University of Amsterdam. It aims to provide training and evaluation resources for large language models (LLMs), such as music recommendation, information retrieval, and sorting. The relevant paper results are:Contrastive Learning of Musical Representations".

The data contains 4.79 billion interactions (including listening, liking, unliking, etc.), covering 1 million users and 9.39 million tracks. It is one of the largest public music recommendation data sets currently.

User interaction count graph