Description
Date depot: 28 mars 2022
Titre: Polarization in large socio-informational networks: models and measures
Directeur de thèse:
Lionel TABOURIER (LIP6)
Encadrant :
Pedro RAMACIOTTI MORALES (medialab)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Données et connaissances
Resumé: The prevalence of AI-based algorithmic recommendations in social and media platforms has raised public concern about undesired societal effects. A central threat is the risk of polarization, despite being difficult to conceptualize and to measure, making it difficult to assess the role of Recommender Systems in this phenomenon. These concerns have inspired a wealth of results, but mostly limited to two types: 1) purely topological approaches that study how recommenders connect or isolate types of nodes in a graph, and 2) spatial opinion approaches that study how recommenders change the distribution of users on some opinion scale. The former prove inadequate in settings where users cannot classified into categorical types (e.g., in two-party systems with binary social divides, which excludes French, European, and most settings), while the latter rely on synthetic data and simulations due to the unobservability of opinions, offering little empirical or actionable results.
To overcome both difficulties this doctoral research project builds on two recent advances: 1) empirical geometrical opinion spaces achieved using graph embedding methods and political survey data [1], and 2) mathematical frameworks for the measurement of properties of recommendations in complex networks [2]. This project first focuses on Recommender Systems of the state of the art to propose a much needed general conceptualization of polarization in graphs models of socio-informational ecosystems and large social systems. It then leverages large social network and information consumption datasets (including Twitter and Facebook data), mathematical modeling of polarization as a property of complex systems in geometrical opinion spaces, and numerical experiments with Recommender Systems (including matrix factorization and GNN recommender systems) trained on large datasets to understand and measure polarization and polarizing dynamics. This highly interdisciplinary project is unique in that it requires –besides research in graphs and algorithmic recommendation– mathematical modeling, and constant dialogue with modals of social dynamics from the emerging field of computational social sciences.
[1] Ramaciotti Morales, Pedro, Jean-Philippe Cointet, Gabriel Muñoz Zolotoochin, Antonio Fernández Peralta, Gerardo Iñiguez, and Armin Pournaki. "Inferring Attitudinal Spaces in Social Networks." (2022). In revision in Social Network Analysis and Mining.
[2] Ramaciotti Morales, Pedro, Robin Lamarche-Perrin, Raphaël Fournier-S'Niehotta, Rémy Poulain, Lionel Tabourier, and Fabien Tarissan. "Measuring diversity in heterogeneous information networks." Theoretical Computer Science 859 (2021): 80-115.
Doctorant.e: Cassells Duncan