Training recommender systems for next-item recommendation often requires unique embeddings to be learned for each item, which may take up most of the trainable parameters for a model. Shared embeddings, such as using content information, can reduce the number of distinct embeddings to be stored in memory. This allows for a more lightweight model; correspondingly, model complexity can be increased due to having fewer embeddings to store in memory. We show the benefit of using shared content-based features ('semantic IDs') in improving recommendation accuracy and diversity, while reducing model size, for two music recommendation datasets, including an online A/B test on a music streaming service.
@article{arxiv.2507.18800,
title = {Semantic IDs for Music Recommendation},
author = {M. Jeffrey Mei and Florian Henkel and Samuel E. Sandberg and Oliver Bembom and Andreas F. Ehmann},
journal= {arXiv preprint arXiv:2507.18800},
year = {2025}
}