A new method for computational cultural cartography: From neural word embeddings to transformers and Bayesian mixture models.

Recently, computational social scientists have proposed exciting new methods for 'mapping meaning space' and analysing the structure and evolution of complex cultural constructs from large text datasets. These emerging approaches to 'cultural cartography' are based on a foundation of neural network word embeddings that represent the meaning of words, in relation to one another, as vectors in a shared high-dimensional latent space. These new methods have the potential to revolutionize sociological analyses of culture, but in their current form, they are dually limited. First, by relying on traditional word embeddings they are limited to learning a single vector representation for each word, collapsing together the diverse semantic contexts that words are used in and which give them their heterogeneous meanings. Second, the vector operations that researchers use to construct larger 'cultural dimensions' from traditional embeddings can result in a complex vector soup that can propagate many small and difficult-to-detect errors throughout the cultural analysis, compromising validity. In this article, we discuss the strengths and limitations of computational 'cultural cartography' based on traditional word embeddings and propose an alternative approach that overcomes these limitations by pairing contextual representations learned by newly invented transformer models with Bayesian mixture models. We demonstrate our method of computational cultural cartography with an exploratory analysis of the structure and evolution of 120 years of scholarly discourse on democracy and autocracy.

Récemment, des spécialistes des sciences sociales informatiques ont proposé de nouvelles méthodes passionnantes pour "cartographier l'espace de sens" et analyser la structure et l'évolution de constructions culturelles complexes à partir de grands ensembles de données textuelles. Ces nouvelles approches de la "cartographie culturelle" reposent sur une base de réseaux neuronaux d'intégration de mots qui représentent la signification des mots, les uns par rapport aux autres, sous forme de vecteurs dans un espace latent partagé à haute dimension. Ces nouvelles méthodes ont le potentiel de révolutionner les analyses sociologiques de la culture, mais dans leur forme actuelle, elles sont doublement limitées. Premièrement, en s'appuyant sur les encastrements de mots traditionnels, elles se limitent à l'apprentissage d'une seule représentation vectorielle pour chaque mot, ce qui réduit les divers contextes sémantiques dans lesquels les mots sont utilisés et qui leur confèrent leurs significations hétérogènes. Deuxièmement, les opérations vectorielles que les chercheurs utilisent pour construire des "dimensions culturelles" plus importantes à partir des encastrements traditionnels peuvent donner lieu à une soupe vectorielle complexe susceptible de propager de nombreuses petites erreurs difficiles à détecter dans l'ensemble de l'analyse culturelle, ce qui en compromet la validité. Dans cet article, nous discutons des forces et des limites de la " cartographie culturelle " computationnelle basée sur les encastrements de mots traditionnels et nous proposons une approche alternative qui surmonte ces limites en associant des représentations contextuelles apprises par des modèles de transformation nouvellement inventés à des modèles de mélange bayésiens. Nous démontrons notre méthode de cartographie culturelle computationnelle par une analyse exploratoire de la structure et de l'évolution de 120 ans de discours savants sur la démocratie et l'autocratie.

Big data, computational social science, and other recent innovations in social network analysis.

While sociologists have studied social networks for about one hundred years, recent developments in data, technology, and methods of analysis provide opportunities for social network analysis (SNA) to play a prominent role in the new research world of big data and computational social science (CSS). In our review, we focus on four broad topics: (1) Collecting Social Network Data from the Web, (2) Non-traditional and Bipartite/Multi-mode Networks, including Discourse and Semantic Networks, and Social-Ecological Networks, (3) Recent Developments in Statistical Inference for Networks, and (4) Ethics in Computational Network Research.

Alors que les sociologues étudient les réseaux sociaux depuis une centaine d'années, les récents développements en matière de données, de technologie et de méthodes d'analyse offrent la possibilité à l'analyse des réseaux sociaux (ARS) de jouer un rôle de premier plan dans le nouveau monde de recherche du big data et des sciences sociales computationnelles (CSS). Dans notre revue, nous nous concentrons sur quatre grands sujets: (1) La collecte de données de réseaux sociaux sur le Web, (2) Les réseaux non traditionnels et bipartites/multimodes, y compris les réseaux discursifs et sémantiques, et les réseaux socio-écologiques, (3) Les développements récents de l'inférence statistique pour les réseaux, et (4) L'éthique dans la recherche informatique sur les réseaux.

Expert predictions of societal change: Insights from the world after COVID project.

How do experts in human behavior think the world might change after the coronavirus (COVID-19) pandemic? What advice do they have for the postpandemic world? Is there a consensus on the most significant psychological and societal changes ahead? To answer these questions, we analyzed interviews from the World After COVID Project-reflections of more than 50 of the world's top behavioral and social science experts, including fellows of National Academies and presidents of major scientific societies. These experts independently shared their thoughts on possible psychological changes in society in the aftermath of the COVID-19 pandemic and provided recommendations how to respond to the new challenges and opportunities these shifts may bring. We distilled these predictions and suggestions via human-coded analyses and natural language processing techniques. In general, experts showed little overlap in their predictions, except for convergence on a set of social/societal themes (e.g., greater appreciation for social connection, increasing political conflict). Half of the experts approached their post-COVID predictions dialectically, highlighting both positive and negative features of the same domain of change, and many expressed uncertainty in their predictions. The project offers a time capsule of experts' predictions for the effects of the pandemic on a wide range of outcomes. We discuss the implications of heterogeneity in these predictions, the value of uncertainty and dialecticism in forecasting, and the value of balancing explanation with predictions in expert psychological judgment. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

From Porter to Bourdieu: The Evolving Specialty Structure of English Canadian Sociology, 1966 to 2014.

How has English Canadian sociology changed from 1966 to 2014? Has it become more intellectually fragmented or cohesive over time? We answer these questions by analyzing cocitation networks extracted from 7,141 sociology articles published in 169 journals. We show how the most central early specialties developed largely in response to John Porter's The Vertical Mosaic. In later decades, the discipline diversified, fragmented, and then reorganized around a new set of specialties knit together by the work of Pierre Bourdieu. The discipline was most intellectually fragmented in periods where multiple specialties were emerging or declining concurrently (i.e., 1975 to 1984 and 1995 to 2004), and was more structurally cohesive from 2005 to 2014 than in any previous period. Comment est-ce que la sociologie canadienne-anglaise a-t-elle changé entre 1966 et 2014? Est-elle devenue plus intellectuellement fragmentée ou cohérente avec le temps? Nous répondons à ces questions en analysant des réseaux de co-citation qui ont été déduits de 7,141 articles publiés par 169 journaux. Nous démontrons les spécialités primordiales se sont développées en réponse de The Vertical Mosaic de John Porter. Durant les décennies suivantes, la discipline s'est diversifiée, fragmentée et puis s'est réorganisée autour d'une nouvelle série de spécialités liées ensemble par le travail de Pierre Bourdieu. La discipline était la plus intellectuellement fragmentée durant les périodes où plusieurs spécialités émergeaient ou déclinaient concurremment (par exemple de 1975 à 1985 et de 1995 à 2004). Par contre, elle était plus cohérente que tous les autres périodes entre 2005 et 2014.

Understanding policy research in liminal spaces: Think tank responses to diverging principles of legitimacy.

Research on scientific, social scientific, and technical knowledge is increasingly focused on changes in institutionalized fields, such as the commercialization of university-based knowledge. Much less is known about how organizations produce and promote knowledge in the 'thick boundaries' between fields. In this article, I draw on 53 semi-structured interviews with Canadian think-tank executives, researchers, research fellows, and communication officers to understand how think-tank knowledge work is linked to the liminal spaces between institutionalized fields. First, although think-tank knowledge work has a broadly utilitarian epistemic culture, there are important differences between organizations that see intellectual simplicity and political consistency as the most important marker of credibility, versus those that emphasize inconsistency. A second major difference is between think tanks that argue for the separation of research and communication strategies and those that conflate them from beginning to end, arguably subordinating research to demands from more powerful fields. Finally, think tanks display different degrees of instrumentalism toward the public sphere, with some seeking publicity as an end in itself and others using it as a means to influence elite or public opinion. Together, we can see these differences as responses to diverging principles of legitimacy.