In this paper, Kristina Matrosova (Géographie-cités, CNRS ; LIPN, USPN, France) et al. explore and confront the definitions of musical taste from sociology, psychology and music recommender systems, based on streamed and explicitly liked music items of a large set of users.

The notion of personal taste in general, and musical taste in particular, is pervasive in the literature on recommender systems, but also in cultural sociology and psychology. However, definitions and measurement methods strongly differ from one study to another.
Kristina Matrosova (Géographie-cités, CNRS ; LIPN, USPN, France), Manuel Moussallam (Deezer Research, France), Thomas Louail (Géographie-cités, CNRS ; PACTE, CNRS, Sciences Po Grenoble, France) and Olivier Bodini (LIPN, USPN, France) question two different views on taste that can be retrieved from the literature: either something that is distinctive of an individual, or something that essentially captures the extent and diversity of their preferences.

Relying upon a dataset that contains the complete list of musical items liked by individual users of a streaming service, as well as streaming logs, we propose two methods to compute fingerprints of their musical taste.
The first one explicitly targets a uniqueness property, aiming at selecting items that uniquely identify a user in the crowd. The second approach focuses on a representativeness task that is fundamental in recommendation, i.e. building a summary depiction of the user’s preferences that can be leveraged to propose other items of interest.The authors demonstrate that the two methods lead to conflicting solutions, hence highlighting the need to precisely acknowledge which point of view applies when addressing a computational question related to taste. They also raise the question of users’ identifiability through their online activity on music streaming platforms, and beyond.

This paper has been realized in the framework of the ‘RECORDS’ grant (ANR-2019-CE38-0013) funded by the ANR (French National Agency of Research).


The RECORDS project combines surveys and big data to study the diversity of consumption and listening practices on streaming platforms, and the measurable effects of recommendations, contexts, and locations on platform listening. Funded by the French National Research Agency (ANR), RECORDS is a collaborative research project involving researchers and engineers from three CNRS laboratories and the R&D departments of Deezer and Orange. The project is coordinated by Thomas Louail, CNRS research fellow at Géographie-cités, Camille Roth, CNRS research fellow at the Centre Marc Bloch (CMB) in Berlin, Philippe Coulangeon, CNRS research director at the Observatoire Sociologique du Changement, Jean-Samuel Beuscart, researcher at Orange R&D, and Manuel Moussallam, head of the Deezer R&D team.

Download Matrosova, K., Moussallam, M., Louail, T., and Bodini, O. (2024). Depict or Discern? Fingerprinting Musical Taste from Explicit Preferences. Transactions of the International Society for Music Information Retrieval, 7(1), 15–29. DOI: https://doi. org/10.5334/tismir.158

Read also

Quentin Villermet, Jérémie Poiroux, Manuel Moussallam, Thomas Louail, Camille Roth. Follow the guides: disentangling human and algorithmic curation in online music consumption. RecSys ’21: Proceedings of the 15th ACM Conference on Recommender Systems, September 2021, pages 380–389