Mixtapes webinar

Marianne Lumeau, Université de Rennes

Within the growth of music consumption through playlists and recommendation algorithms, concerns has raised among professionals and certain governments in small economies, including France, about the discoverability of local content. In this context, the objective of this study is to examine whether the recommendation algorithm of a dominant music streaming platform generates geographical and linguistic biases against local music from a small economy (France) compared to local content from a large economy (the United States). Based on an experiment involving the creation of fictitious user accounts consuming local music tracks in France and the United States during five weeks, we show that there is no systematic geographical bias against music produced in France. However, we identify a linguistic bias: French-language songs are recommended less frequently than English-language songs. Furthermore, playing recommended songs amplifies this linguistic bias and generates a geographical bias.

In collaboration with Samuel Coavoux (ENSAE, CREST), François Moreau (Paris Sorbonne Nord Univ., ACT), and Jordana Viotto (Univ. Edinburgh Business School)

Laurie Hanquinet, Université Libre de Bruxelles

Massimo Airoldi, Università degli Studi di Milano Statale

Mads Meier Jæger, Københavns Universitet

Gabriel P. Oliveira, Universidade Federal de Minas Gerais

Social media and streaming platforms have reshaped music consumption, enabling songs to go viral and achieve commercial success. In this talk, I explore a simple yet powerful question: can the spread of music popularity be understood in the same way we understand the spread of diseases? Inspired by epidemiology, in my doctoral research, I investigated this question by applying epidemic models to analyze how songs gain traction in the digital landscape. Insights from studies of both the Brazilian and global music markets reveal that classic epidemic models are effective at describing viral explosions, but are less capable of explaining why some songs endure. This approach not only improves our ability to describe and forecast viral behavior but also provides interpretable parameters that help us reason about the underlying diffusion processes.

See also:
–  GP Oliveira, L Vassio, APC da Silva & MM Moro (2025) Modeling music popularity as an epidemic: insights from the Brazilian market. Brazilian Workshop on Social Network Analysis and Mining (BraSNAM’25), 79-92.
– GP Oliveira, L Vassio, APC da Silva, MM Moro (2025) Contagious Rhythms: A Wave-Based Epidemic Approach for Music Virality on Social Platforms. International Conference on Advances in Social Networks Analysis and Mining (ASONAM’25), 182-196.

Listen to Gabriel’s mixtape on Deezer

Lysandre Champagne, Observatoire de la culture et des communications du Québec

Cette présentation fait état de la situation de l’écoute de musique locale et francophone au Québec à partir d’une analyse comparative de deux sources de données : l’Enquête québécoise sur les loisirs culturels et le divertissement (EQLCD) 2024, réalisée par l’Institut de la statistique du Québec (ISQ), et les données de consommation issues de la plateforme Luminate, traitées et analysées par l’ISQ. Dans le contexte actuel, le rôle des technologies numériques et des algorithmes dans la consommation et la découvrabilité de musique locale et francophone suscite un intérêt croissant, tant au Québec qu’en France. Nos résultats, publiés en août 2025, mettent en lumière les liens entre l’utilisation des différents modes de diffusion, les profils sociodémographiques et l’écoute de musique locale et en français, et révèlent des portraits d’écoute contrastés.

Lien vers la publication : Optique culture – Numéro 107. Août 2025 – Les pratiques d’écoute et la consommation d’enregistrements musicaux au Québec en 2024

Florian Keusch, Universität Mannheim

For decades, social scientists relied primarily on self-reported data collected through surveys and interviews. These remain essential for understanding attitudes, motivations, and social contexts. The rise of smartphones and online platforms, however, has made digital behavioral data a powerful complementary source, offering detailed traces of actual practices but also raising questions about data quality, selection mechanisms, and ethics. In this talk, I bring together results from several recent methodological projects showing how these two data sources function as “solo artists” with distinct strengths and limitations, and how they produce their best performance when combined. I discuss the methodological and practical challenges involved, especially concerning data access, linkage, and evaluating data quality using examples from various digital behaviors.

Elena Epure & Bruno Massoni Sguerra, Deezer research

One particularly promising use case of Large Language Models (LLMs) for recommendation is the automatic generation of Natural Language (NL) user taste profiles from consumption data. These profiles offer interpretable and editable alternatives to opaque collaborative filtering representations, enabling greater transparency and user control. However, it remains unclear whether users consider these profiles to be an accurate representation of their taste, which is crucial for trust and usability. Moreover, because LLMs inherit societal and data-driven biases, profile quality may systematically vary across user and item characteristics. In this paper, we study this issue in the context of music streaming, where personalization is challenged by a large and culturally diverse catalog. We conduct a user study in which participants rate NL profiles generated from their own listening histories. We analyze whether identification with the profiles is biased by user attributes (e.g., mainstreamness, taste diversity) and item features (e.g., genre, country of origin). We also compare these patterns to those observed when using the profiles in a downstream recommendation task. Our findings highlight both the potential and limitations of scrutable, LLM-based profiling in personalized systems.

> Read the paper

Myriam Boualami, Centre Marc Bloch & Géographie-cités

Literature has shown how rap artists affiliate with their local scene in the hope of gaining recognition. At the same time, commercial success requires that artists capture industry professionals’ attention typically concentrated in major cities – coined as national scenes. This geographical interplay between proximity and scene hierarchy is also called into question by music «platformisation» which has the potential to foster large-scale success regardless of localisation and connections. In this context, does geographical proximity maintain its role in facilitating commercial success? Using consumption data from the Deezer streaming platform, this presentation will both define the object of ‘rap’ on-platform, adopting both inter- and intra-genre perspectives, and examine the relationship between local collaborations, spatial hierarchy and success. Together, these contributions exemply how digital consumption data offers new ways to engage in key debates in the Social Sciences of music production, ranging from the categorisation of musical objects to the role of geography in music networks, whilst also highlighting the challenges this empirical material pose.

Giulio Prevedello and Pietro Gravino, SONY CSL Paris

This talk explores the role of human behaviour in shaping media consumption, focusing on how attention drives dynamics in music and news. In the study about music, we investigate the factors influencing song popularity after release. In particular we investigated the opportunity of including lyrics features for popularity prediction, by assessing how they integrate with additional information such as audio characteristics and platform metadata. Shifting to the news domain, we analyse the interplay between demand, supply, and diffusion of information. This analysis reveals gaps between supply and demand, where disinformation flourishes and captures audience attention, highlighting vulnerabilities in the news ecosystem. Both studies underscore attention as a critical factor in these systems, illustrating how content characteristics and audience engagement shape outcomes, whether through music streams or the consumption of news.

Quentin Gilliotte, Université Panthéon-Assas & CERLIS

The productions on digital platforms are becoming an increasingly significant source of information. Within this context, a large number of actors on digital platforms offer analyses and commentary on political current events. Within the Francophone space, we find video creators such as HugoDécrypte, Tatiana Ventôse, Usul, and Gaspard G. These actors place themselves in opposition to traditional media (television, radio) : they advocate for an « alternative » way of handling information, whether in terms of the audience they target, the topics they cover, or the formats they use. How do these different actors address political current events? Considering both economic models and political positions, what topics are mobilized? The communication will proceed in two stages. Firstly, I will present the method for constructing the corpus of selected channels through data extraction from the YouTube API, utilizing network analysis tools. Secondly, I will present a mapping of the topics addressed by these different video creators based on the automated textual analysis of a corpus of 8740 transcribed videos (representing all videos published by the selected actors between January and July 2023) using a topic modeling method (Latent Dirichlet Allocation).

Harin Lee, Max Planck Institute for Human Cognitive and Brain Sciences

Studying large-scale music listening patterns can offer a unique perspective on societal sentiment and cultural preferences. In this talk, I will demonstrate how traces of comprehensive listening logs, with them linked to fine-grained demographic data, allows to causally test theories of cultural diversity and urban scaling laws in the real world. Furthermore, I will share insights from a worldwide study of music patterns in 1,423 cities showing how societal upheavals, such as war, induce measurable shifts in musical trends and reflect diverse pathways in cultural evolution. Specifically, using network analysis techniques and natural language processing, I will demonstrate how a once homogenous post-Soviet musical cluster fragmented into distinct national clusters following the recent and on-going Russia’s invasion of Ukraine. Together, these works provide insights into how societal dynamics and demography imprints on the evolution of human culture. The accompanying mixtape features key Ukrainian songs that rapidly gained popularity amidst the invasion, reflecting themes of war and resistance.

> Listen to Harin’s mixtape on Deezer