Camille Roth’s research while affiliated with Centre d'Analyse et de Mathématique Sociales and other places

The impact of platforms’ algorithmic curation devices on users’ consumption diversity has long since been the subject of debate in popular culture and academia alike. In recent years, methods to concretely appraise this debate have taken a comparative turn, with emerging research strands seeking to assess the influence of algorithmic recommendations relative to users’ organic behaviour (i.e. those with a relative lack of algorithmic influence). Nevertheless, as the contribution of this work clearly shows, contradictory findings that algorithmic devices may both foster “bubbly” confinement dynamics and diverse exploration appear to stem, at least in part, from tensions surrounding information representation and scale. Taking into account various scales, affordances, and diversity measures, the contribution of this work conducts an empirical analysis of approximately 50,000 Deezer users’ artist consumption histories across different affordances offering various degrees of algorithmic assistance. Transitioning through three scales of increasing aggregation—intra-sessions, inter-sessions and inter-affordances—our work underscores how shifts in perspective, particularly in information representation, can yield markedly different conclusions concerning the role of algorithmic devices in fostering diversity. However, when considered collectively, these perspectives coalesce to provide a nuanced update to the seminal filter bubble narrative: algorithmic curation devices may introduce more novelty than what users’ achieve organically, yet concurrently, this novelty is more semantically confined. Consequently, our research significantly contributes towards an enriched understanding of the intricate dynamics surrounding on-platform algorithmic devices.

In the field of node representation learning the task of interpreting latent dimensions has become a prominent, well-studied research topic. The contribution of this work focuses on appraising the interpretability of another rarely-exploited feature of node embeddings increasingly utilised in recommendation and consumption diversity studies: inter-node embedded distances. Introducing a new method to measure how understandable the distances between nodes are, our work assesses how well the proximity weights derived from a network before embedding relate to the node closeness measurements after embedding. Testing several classical node embedding models, our findings reach a conclusion familiar to practitioners albeit rarely cited in literature—the matrix factorisation model SVD is the most interpretable through 1, 2 and even higher-order proximities.

We present a mixed methods research design that integrates individual-level traces of online content consumption with survey and interview data, all collected from the same sample of tens of thousands of people. We explain how the pairwise combination of these information sources solves methodological puzzles often encountered when measuring cultural practices and preferences. We provide a concrete illustration for the case of music listening on streaming platforms, and show that survey respondents and solicited users do not stream the same music. We illustrate how the mixed data collected make it feasible to infer the social properties of non-respondents, and hence to assess bias in studies based exclusively on self-reported survey data. We provide empirical evidence that unlimited access to all kinds of recorded music on platforms does not blur the social boundaries between repertoires across respondents, and we show that artists have distinct audiences whose differences are both related to generation, gender, and educational attainment. Finally, we describe how we used digital traces in individual interviews to foster spontaneous expressions of aesthetic judgments, which are known to be challenging to collect. We conclude by discussing the current limitations of the design, the potential applications of these results in the field of cultural sociology, and the feasibility of adopting similar experimental designs to investigate other social phenomena.

Since the 1990s, the Environmental Kuznets Curve (EKC) hypothesis posits an inverted U-shaped relationship between pollutants and economic development. The hypothesis has attracted a lot of research. We provide here a review of more than 2000 articles that have been published on the EKC. We aim at mapping the development of this specialized research, both in term of actors and of content, and to trace the transformation it has undergone from its beginning to the present. To that end, we combine traditional bibliometric analysis and semantic analysis with a novel method, that enables us to recover the type of pollutants that are studied and the empirical claims made on EKC (whether the hypothesis is invalidated or not). We principally exhibit the existence of a few epistemic communities that are related to distinct time periods, topics and, to some extent, proportion of positive results on EKC.

Since the 1990s, the Environmental Kuznets Curve (EKC) hypothesis posits an inverted U-shaped relationship between pollutants and economic development. The hypothesis has attracted a lot of research. We do here a review of more than 2000 papers that have been published on the EKC. To that end, we combine traditional bibliometric analysis and semantic analysis with a novel method, that enables us to recover the type of pollutants that are studied as well as the empirical claims made on EKC (whether the hypothesis is invalidated or not). We principally exhibit the existence of a few epistemic communities that are related to distinct time periods, topics and, to some extent, proportion of positive results on EKC.

Several recent works on recommender algorithms have called for shifting the focus away from the study of their effects, such as the emergence of prediction biases or filter bubbles, to look at how they are designed. We propose here to answer this call thanks to a qualitative study based on interviews with about thirty developers. We show that the conditions of production of these algorithms are very closely linked to their use. Deployed on platforms with a large number of users, thus allowing a permanent observation of their functioning, algorithmic code evolves in a hybrid way that continuously depends on the work of developers and the actions of users. Simply put, the use of algorithmic guidance guides its own evolution – whether it is introducing new variables, new algorithmic processes and, above all, choosing between numerous variants through tests that quantify user reactions in real time in the light of essentially commercial objectives. From this point of view, code development is to a large extent a semi-autonomous evolutionary process in which user testing is the main arbiter: developers introduce mutations, users implicitly produce performance calculations, expressed in standard business terms (audience, sales). By emphasizing the crucial importance of the choice of these metrics, once the choices concerning the architecture of a given platform are made, we call on future research to frame the question of algorithmic policy primarily in terms of the definition of these two dimensions –performance and platform design– rather than opening up further the black box of code and its design.

Interaction networks are generally much less homophilic than affiliation networks, accommodating for many more cross-cutting links. By statistically assigning a political valence to users from their follower ties, and by further contrasting interaction and affiliation on Twitter (quotes and retweets) within specific discursive events, namely quote trees, we describe a variety of cross-cutting patterns which significantly nuance the traditional “echo chamber” narrative.