Coauthor
  • CARDON Dominique (9)
  • RAMACIOTTI MORALES Pedro (6)
  • PARASIE Sylvain (4)
  • CRÉPEL Maxime (3)
  • Show more
Document Type
  • Article (15)
  • Web site contribution (3)
  • Conference contribution (2)
  • Conference proceedings (2)
  • Show more
Ce site présente une partie des résultats de travaux réalisés dans le cadre du projet Algoglitch au médialab de Sciences Po. Il articule des analyses descriptives et des visualisations de données, portant sur un corpus d’articles de presse sur le thème de l’IA et des algorithmes couvrant une période de 5 années aux États-Unis et au Royaume-Uni. A partir de méthodes de traitement automatique du langage sur un corpus de presse, cette recherche vise à explorer les discours critiques sur l’IA dans la sphère médiatique. A partir de méthode de traitement automatique du langage sur un corpus de presse, nous montrons que le thème de l’IA occupe un espace de plus en plus important dans la presse depuis 5 ans. L’espace médiatique se structure de manière thématique autour de différentes technologies de calcul et domaines d’applications et peut être divisé en deux sous-ensembles sémantiques. L’analyse comparée de ces deux espaces sémantiques rend compte de deux régimes de critique dominants, mobilisant une variété d’entités techniques et humaines, ainsi que des temporalités et des enjeux différents. Le premier est fondé sur les injustices produites par les algorithmes qui façonnent nos environnements de calcul quotidiens et sont associés à un discours critique portant sur les biais, les discriminations, la surveillance et la censure dont des populations spécifiques sont victimes. Le second porte sur les peurs de l’autonomie de l’IA et des robots qui, en tant qu’entités techniques autonome et incarnées, sont associées à un discours prophétique, alertant sur notre capacité à contrôler ces agents simulant ou dépassant nos capacités physiques et cognitives, mettant ainsi en péril notre sécurité physique, notre modèle économique et menaçant ainsi l’humanité tout entière.

This website presents some of the results of research carried out under Sciences Po Médialab’s Algoglitch project. It combines descriptive analyses with data graphics on a corpus of press articles on the topic of AI and algorithms, spanning a 5-year period in the United States and United Kingdom. Using natural language processing applied to a press corpus, this research explores critical discourses on AI in the media sphere. Using natural language processing on a press corpus, we show that the subject of AI has been occupying an increasingly larger space in the press over the past five years. The media space is structured thematically around different calculation technologies and fields of application, and can be divided into two semantic subsets. A comparative analysis of these two semantic spaces reveals two dominant regimes of criticism involving a variety of technical and human entities, as well as different time scales and issues. The first is structured around the injustices produced by the algorithms that shape our everyday calculation environments, which are associated with criticism of the biases, discrimination, monitoring, and censorship of which specific populations are the victims. The second space is structured around fears of the autonomy of AI and robots which, as autonomous and embodied technical entities, are associated with a prophetic discourse drawing attention to our capacity to control these entities capable of simulating or surpassing our physical and cognitive abilities. The threat to our physical safety and economic model, and consequently to all of humanity, is thus highlighted.

Le refus de se positionner sur l’axe droite-gauche caractérise le mouvement des Gilets jaunes renvoyant sans cesse dos à dos les formations politiques plutôt que de prendre parti pour l’une d’entre elles. Pourtant les Gilets jaunes, lorsqu’ils font leur apparition en France, s’expriment dans un espace public déjà nourri de tensions et de structures idéologiques préexistantes. À ce titre, leur action est nécessairement située, elle s’inscrit dans cet espace et en hérite certaines propriétés. Il est dès lors légitime de s’intéresser à la place qu’occupe le mouvement, notamment dans sa déclinaison numérique sur Facebook. Comment les pratiques de citation en ligne trahissent-elles non pas la couleur politique du mouvement, mais l’espace politique dont ils se nourrissent et qu’ils alimentent ? Cet article répond à cette question en introduisant un cadre méthodologique original qui permet d’étendre un plongement idéologique d’utilisateurs sur Twitter vers des posts publiés sur Facebook. Nous faisons d’abord appel à une analyse de correspondance pour réduire la matrice d’adjacence qui lie les parlementaires français à leurs followers sur Twitter. Cette première étape nous permet d’identifier deux axes latents qui sont déterminants pour expliquer la structure du réseau. La première dimension distribue les individus selon leur positionnement sur l’axe droite-gauche de l’espace politique. Nous interprétons la seconde dimension comme une mesure de la distance au pouvoir. Ces deux dimensions sous-tendent un espace dans lequel nous positionnons successivement des centaines de milliers d’utilisateurs de Twitter, les URLs et les médias cités sur cette plateforme et, par extension, les publications de près de 1000 groupes Facebook parmi les plus actifs associés au mouvement des Gilets jaunes. Nous quantifions finalement l’évolution des publications de ces groupes dans l’espace idéologique latent pour donner à la fois un sens et une réponse à la question de l’inclinaison politique du mouvement. Les dynamiques observées renforcent l’interprétation d’un mouvement qui, d’abord positionné très à droite, a rapidement opéré un glissement vers la gauche tout en restant fidèle à une attitude contestataire. Cette description par l’usage que les Gilets jaunes font des médias sur Facebook illustre parfaitement l’idée d’un populisme polyvalent.

The prevalence of algorithmic recommendations has raised public concern about undesired societal effects. A central threat is the risk of polarization, which is difficult to conceptualize and to measure, making it difficult to assess the role of Recommender Systems in this phenomenon. These difficulties have yielded two types of analyses: 1) purely topological approaches that study how recommenders isolate or connect types of nodes in a graph, and 2) spatial opinion approaches that study how recommenders change the distribution of users on a given opinion scale. The former analyses prove inad- equate in settings where users are not classified into categorical types (e.g., in two-party systems with binary social divides), while the latter rely on synthetic data due to the unobservability of opin- ions. To overcome both difficulties we present the first analysis of friend recommendations acting on real-world sub-graphs of the Twitter network where users are embedded in multidimensional ideological spaces and in which dimensions are indicators of atti- tudes towards issues in the public debate. We present a polarization metric adapted to these dual topological and spatial states of social network, and use it to track both the evolution of polarization on Twitter networks where the graph evolves following well-known Recommender Systems, and opinions co-evolve following a De- Groot opinion model. We show that different recommendation principles can sometimes drive or mitigate polarization appearing in real social networks.

This study provides a large-scale mapping of the French media space using digital methods to estimate political polarization and to study information circuits. We collect data about the production and circulation of online news stories in France over the course of one year, adopting a multi-layer perspective on the media ecosystem. We source our data from websites, Twitter and Facebook. We also identify a certain number of important structural features. A stochastic block model of the hyperlinks structure shows the systematic rejection of counter-informational press in a separate cluster which hardly receives any attention from the mainstream media. Counter-informational sub-spaces are also peripheral on the consumption side. We measure their respective audiences on Twitter and Facebook and do not observe a large discrepancy between both social networks, with counter-information space, far right and far left media gathering limited audiences. Finally, we also measure the ideological distribution of news stories using Twitter data, which also suggests that the French media landscape is quite balanced. We therefore conclude that the French media ecosystem does not suffer from the same level of polarization as the US media ecosystem. The comparison with the American situation also allows us to consolidate a result from studies on disinformation: the polarization of the journalistic space and the circulation of fake news are phenomena that only become more widespread when dominant and influential actors in the political or journalistic space spread topics and dubious content originally circulating in the fringe of the information space.

Traditionally, the opinion of people on different issues of public debate has been studied through polls and surveys. Recent advancements in network ideological scaling methods, however, have shown that digital behavioral traces in social media platforms can be used to mine opinions at a massive scale. Yet this has been shown to work in the US for one-dimensional opinion scales, best suited for two-party systems and binary social divides. In this article, we use multidimensional ideological scaling together with referential attitudinal data for some nodes. We show that opinions can be mined in a multitude of issues from social networks, embedding them in ideological spaces where dimensions stand for indicators of positive and negative opinions towards issues of public debate. This method does not require text analysis and is thus language independent. We illustrate this approach on the Twitter follower network of French users leveraging political survey data.

in 2020 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) Publication date 2020-12
87
views

0
downloads
Multidimensional scaling in networks allows for the discovery of latent information about their structure by embedding nodes in some feature space. Ideological scaling for users in social networks such as Twitter is an example, but similar settings can include diverse applications in other networks and even media platforms or e-commerce. A growing literature of ideology scaling methods in social networks restricts the scaling procedure to nodes that provide interpretability of the feature space: on Twitter, it is common to consider the sub-network of parliamentarians and their followers. This allows to interpret inferred latent features as indices for ideology-related concepts inspecting the position of members of parliament. While effective in inferring meaningful features, this is generally restrained to these sub-networks, limiting interesting applications such as country-wide measurement of polarization and its evolution. We propose two methods to propagate ideological features beyond these sub-networks: one based on homophily (linked users have similar ideology), and the other on structural similarity (nodes with similar neighborhoods have similar ideologies). In our methods, we leverage the concept of neighborhood ideological coherence as a parameter for propagation. Using Twitter data, we produce an ideological scaling for 370K users, and analyze the two families of propagation methods on a population of 6.5M users. We find that, when coherence is considered, the ideology of a user is better estimated from those with similar neighborhoods, than from their immediate neighbors.

in UiO: C-REX - Center for Research on Extremism Publication date 2020-03-30
126
views

0
downloads
First lines: Activists wearing Yellow Safety Vests started taking the streets in France since October 2018. Many commentators linked their grievances to radical right and “anti-establishment” politics. Why is it not so simple? Activists wearing yellow safety vests, or Gilets Jaunes, started taking the streets in France in October 2018. As these uprisings could not be connected to any political party or to any clear political agenda, some commentators linked their grievances with the ethnocentric and ‘anti-establishment’ discourse of the Rassemblement National (formerly Front National, RN). After more than a year of demonstrations, and some attempts by Marine Le Pen to latch on to the Yellow Vests (YVs) the RN has failed to capitalize on this discontent suggesting that the relationship of the YVs with the populist radical right is probably not that obvious. We argue that it is overly simplistic to associate the YVs with the populist radical right. While the ideology of the radical right is crucially informed by nativism, authoritarianism and populism, the YVs movement is not based on a single, accepted platform, and it talks very little about immigration and law and order issues. In addition, it is not just against the ‘establishment’ or democracy tout court but mostly concerned with institutional reforms (notably to improve the accountability of the executive). Our claim is supported by the findings of an ongoing research project at CEE & médialab of Sciences P

In this paper we introduce a novel approach for the computational analysis of research activities and their dynamics. Named SASHIMI (Symmetrical And Sequential analysis from Hierarchical Inference of Multidimensional Information), our approach provides a multi-level description of the structure of scientific activities that offers numerous advantages over traditional methods such as topic models or network analyses. Our method generates a dual description of corpora in terms of research domains (collections of documents) and topics (collections of words). It also extends this description to clusters of associated dimensions, such as time. SASHIMI only requires access to the textual content of individual documents, rather than specific metadata such as citations, authors, or keywords as is the case with other science-mapping approaches. We illustrate the analytical power of our method by applying it to the empirical analysis of an original dataset, namely the 1995-2017 collection of abstracts presented at ASCO, the largest annual oncology research conference. We show that SASHIMI is able to detect the presence of significant temporal patterns and to identify the major thematic transformations of oncology that underlie these patterns.

Le codage de texte est au coeur de la pratique des sociologues et renvoie à toute une variété de pratiques, de types de matériaux textuels et de corpus et plus largement de modalités de production de connaissance. Différentes options se présentent à l’analyste lorsqu’il souhaite coder avec une machine. Entre les méthodes inductives entièrement non-supervisées venant de l’informatique et la reconnaissance de motifs lexicaux assistés par ordinateur, nous proposons une troisième voie qui s’appuie sur les capacités d’inférence de l’apprentissage machine tout en garantissant un contrôle des catégories analytiques utilisées pour le codage. Une méthode de codage supervisé actif est ainsi appliquée à deux corpus textuels: un ensemble de commentaires collectés sur un corpus de commentaires publiés sur le web, et un corpus d’articles de presse.

Next