Opinion dynamics : from data to models and back
Measuring the Complexity of Interactions in the Language System, Quentin Feltgen
What is the structure of language interactions? We already know that language is heavily structured by Zifp’s law (Zipf 1935), be it at the general vocabulary level (Condon 1928), for higher n-grams (Ha et al. 2009), or over the syntactic dependency network (Ferrer i Cancho et al. 2004). This behavior is also found at a more local level: for lexical niches like natural entities or numbers (Piantadosi 2014), for individual semi-schematic constructions (Feltgen 2020), for argument structure constructions (Ellis 2012), etc. The emergence of Zipf’s law at the general level is believed to reflect a sharing of the coding and decoding efforts between speaker and hearer (Ferrer i Cancho & Solé 2003), while Zipf’s law at the individual level has been related to learning mechanisms (Goldberg et al. 2004) and to the scale-free structure of semantic networks (Ellis et al. 2014). However, these works do not address the structure of the interaction between two such Zipfian paradigms, like the noun and adjective ones in the epithet construction, even though combination is crucial to generate a meaningful, creative, and diverse language output.
In this contribution, we aim to address general properties of the syntactic interactions between two paradigmatic slots. To do so, we shall focus on the epithet construction in contemporary French (1980-2024), based on data extracted from the Frantext corpus associated with that period (ATILF 1998-2025). To reduce the volume of occurrences and ensure a better homogeneity in the output, we focus on indefinite contexts (e.g. une joie sincère), yielding 171,000 interactions between 12,000 nouns and 9,000 adjectives, for a total of 111,000 different combinations.
To assess the richness of the interaction structure, we may consider computing the mutual information between the two slots. This mutual information varies between 0 (all nouns combine with every adjective in equal measure) and the minimum of the two slots’ entropies (each noun combines with a unique and therefore entirely predictable adjective). None of these extreme values correspond to interesting structures in a communication perspective, which is why, following Santamaría-Bonfil et al. (2016), we rather consider the complexity coefficient, striking a balance between these extremes, and defined as C = 4I(1-I), where I is the mutual information. Computing this complexity score requires normalizing the mutual information, so that it varies between 0 and 1. In this paper, we define precise bounds for the mutual information to return a min-max transform of it, ensuring that the complexity score is robust with sample size variations.
We find a complexity score of 0.84 for the epithet construction system. Interestingly, if we redraw the links of the system, therefore only keeping the Zipfian structure of each of the two slots, the complexity drops to 0.51, and the number of interactions increases to 156,000. If we add the information of the degrees of each noun and each adjective to match the observed number of combinations, such that the tokens are randomly re-drawn to reinforce existing links based on the respective Zipfian frequencies of the nodes, the complexity increases to 0.62, which is still far from the observed complexity score.
This discrepancy reveals that the interactions are highly structured beyond the random associations of the two slots, even when accounting for their frequency and degree distributions. An examination of the noun-adjective constructs’ paradigm shows that the observed Zipf’s law over these constructs is far steeper than the Zipf’s law for the randomly redrawn system, whichever structural constraints are taken into account. These findings highlight the over-abundance of formulaic elements in language use, and echo with the observation that language production is highly reliant on prefabricated contents (Erman & Warren 2000); more surprisingly, our results evidence that this formulaic character is precisely the property that guarantees the high degree of complexity of language’s syntactic interactions.
References
ATILF. (1998-2025). Base textuelle Frantext (Online). ATILF-CNRS & Université de Lorraine. https://www.frantext.fr/
Condon, E. U. (1928). Statistics of vocabulary. Science, 67(1733), 300-300.
Ellis, N. C. (2012). Formulaic language and second language acquisition: Zipf and the phrasal teddy bear. Annual review of applied linguistics, 32, 17-44.
Ellis, N. C., O’Donnell, M. B., & Römer, U. (2014). Does Language Zipf Right Along? In J. Connor-Linton & L. Wander Amoroso (Eds.), Measured language: Quantitative studies of acquisition, assessment, and variation (pp. 33–50). Georgetown University Press.
Erman, B., & Warren, B. (2000). The idiom principle and the open choice principle. Text & Talk, 20(1), 29-62.
Feltgen, Q. (2020). Diachronic emergence of Zipf-like patterns in construction-specific frequency distributions: A quantitative study of the way too construction. Lexis, 16.
Ferrer i Cancho, R., & Solé, R. V. (2003). Least effort and the origins of scaling in human language. Proceedings of the National Academy of Sciences, 100(3), 788-791.
Ferrer i Cancho, R., Solé, R. V., & Köhler, R. (2004). Patterns in syntactic dependency networks. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 69(5), 051915.
Goldberg, A. E., Casenhiser, D. M., & Sethuraman, N. (2004). Learning argument structure generalizations.
Ha, L. Q., Hanna, P., Ming, J., & Smith, F. J. (2009). Extending Zipf’s law to n-grams for large corpora. Artificial Intelligence Review, 32, 101-113.
Piantadosi, S. T. (2014). Zipf’s word frequency law in natural language: A critical review and future directions. Psychonomic bulletin & review, 21, 1112-1130.
Santamaría-Bonfil, G., Fernández, N., & Gershenson, C. (2016). Measuring the complexity of continuous distributions. Entropy, 18(3), 72.
Zipf, G. K. (1935). The Psycho-Biology of Language. Houghton Mifflin Company.
Synchronisation entre les partisan·es des médias et les sympathisant·es politiques lors d’un processus électoral : vers une étude en temps réel, Rémi Perrier, Laura Hernández, J. Ignacio Alvarez-Hamelin, Mariano G. Beiró and Dimitris Kotzinos
Les réseaux sociaux sont devenus un terrain d’étude courant de la diffusion des avis politiques. Les partis politiques et les médias traditionnels utilisent les discussions sur des plateformes spécialisées pour évaluer les tendances de l’opinion sociale, malgré les préoccupations concernant la représentativité des données des réseaux en ligne. La presse papier, les programmes de radio, ou encore les émissions de télévision discutent souvent de ces conversations en ligne, diffusant ainsi leur contenu au-delà des plateformes en ligne.
La plupart des travaux empiriques sur l’opinion sociale en ligne sont basés sur l’étude des usages d’un ensemble de mots-clés choisis a priori, en fonction du sujet étudié. Notre travail, basé sur les publications Twitter (désormais X) collectées entre septembre 2021 et juin 2022, suit une approche entièrement automatisée [1]. Nous construisons d’abord un réseau pondéré de hashtags avec le lien pondéré représentant le nombre d’utilisateur⋅ices uniques ayant utilisé deux hashtags dans le même tweet. Ensuite, nous déterminons les sujets de discussion sur la plateforme par détection de communautés sur ce réseau sémantique. Cette procédure a été appliquée avec succès pour étudier les discussions politiques pendant une période électorale en Argentine [2], ainsi que pour étudier la dynamique des interactions entre un média traditionnel, le journal New York Times, et ses abonnés sur Twitter [3], en utilisant un réseau sémantique statique basé sur les données collectées sur toute la période étudiée. En conséquence, la détermination des sujets discutés à un moment donné inclut des informations provenant du futur, ce qui n’est pas pertinent si l’on souhaite suivre les événements en temps réel. Dans ce travail, nous présentons une méthode qui permet de le faire. Nous traitons des réseaux sémantiques évolutifs, ce qui implique un compromis entre l’instantanéité des informations collectées et une quantité raisonnable de données requises pour les rendre robustes. Ici, nous comparons deux procédures différentes de construction du réseau sémantique. Dans un cas, nous cumulons les données du premier mois, puis nous ajoutons les nouvelles données chaque semaine, ce qui conserve la mémoire de toutes les discussions sur Twitter depuis le début de la capture. Dans l’autre, nous partons du même réseau qu’auparavant, et nous utilisons une fenêtre glissante d’une semaine pour intégrer les nouvelles données et supprimer les plus anciennes, ce qui entraîne une perte de mémoire chaque semaine. Dans les deux cas, pour chaque nouveau réseau, nous déterminons les communautés qui constituent les sujets de discussion de la semaine. Pour un groupe donné (partisan⋅es d’un média ou sympathisant⋅es d’un⋅e candidat⋅e), nous créons un vecteur où chaque composante relève le niveau de participation dans chaque sujet. La similarité entre deux groupes est alors calculée en comparant leur vecteur dans l’espace des sujets. Nous montrons qu’en général, les deux procédures donnent qualitativement le même comportement des courbes de similarité dynamique entre les partisan⋅es de différents candidat⋅es. Les rares exceptions concernent des situations atypiques que nous caractérisons.
De plus, puisque notre approche nous amène à mettre à jour le paysage sémantique, en reconstruisant le graphe de cooccurrence sur une base hebdomadaire, nous sommes en mesure de caractériser et de suivre l’évolution des sujets eux-mêmes. Nous utilisons une procédure dynamique pour suivre les communautés au fil du temps [4], et nous observons comment les sujets de discussion croissent, diminuent (voire disparaissent), se divisent ou fusionnent avec d’autres. Dans la Fig.1, nous montrons que pour des sujets controversés comme les politiques de vaccination, au fil du temps, les sujets intègrent de nouveaux hashtags très utilisés qui tendent vers des positions extrémistes.
[1] F. M. Cardoso, S. Meloni, A. Santanchè, and Y. Moreno, “Topical alignment in online social systems”, Frontiers in Physics 7, 58 (2019).
[2] T. Mussi Reyero, M. G. Beiró, J. I. Alvarez-Hamelin, L. Hernández, and D. Kotzinos, “Evolution of the political opinion landscape during electoral periods”, EPJ Data Science 10, 31 (2021).
[3] H. Schawe, M.G. Beiró, J.I. Alvarez-Hamelin et al. “Understanding who talks about what: comparison between the information treatment in traditional media and online discussions”. Sci Rep 13, 3809 (2023).
[4] D. Greene, D. Doyle, and P. Cunningham, “Tracking the evolution of communities in dynamic social networks”, in 2010 international conference on advances in social networks analysis and mining (Aug. 2010), pp. 176–183.
Theoretical models of opinion dynamics can accurately identify individual political preferences from online interaction data, Antoine Vendeville
Models of opinion dynamics describe how opinions are shaped in various environments. While these models are able to replicate macroscopical opinion distributions observed in real-world scenarios, their capacity to align with data at the microscopical level remains mostly untested. We evaluate the capacity of the celebrated voter model to capture individual opinions in a fine-grained Twitter dataset collected during the 2017 French Presidential elections. Our findings reveal a strong correspondence between individual opinion distributions in the equilibrium state of the model and ground-truth political leanings of the users. Additionally, we demonstrate that discord probabilities accurately identify pairs of like-minded users. These results emphasize the validity of the voter model in complex settings, and advocate for further empirical evaluations of opinion dynamics models at the microscopical level.
Inference of multi-dimensional political positions of online users and web domains: methodology and validation on large-scale French Twitter data, Antoine Vendeville et al.
The study of phenomena related to public opinion online and especially political polarization garners significant interest in Computational social sciences. The undertaking of several studies of political phenomena in social media mandates the operationalization of the notion of political stance of users and contents involved. Relevant examples include the study of segregation and polarization online, or the study of political diversity in content diets in social media. While many research designs rely on operationalizations best suited for the US setting, few allow addressing more general design, in which users and content might take stances on multiple ideology and issue dimensions, going beyond traditional Liberal-Conservative or Left-Right scales. To advance the study of more general online ecosystems, we present a methodology for the computation of multidimensional political positions of social media users and web domains. We perform a case study on a large-scale X/Twitter population of users in the French political Twittersphere and web domains, embedded in a political space spanned by dimensions measuring attitudes towards immigration, the EU, liberal values, elites and institutions, nationalism and the environment. We provide several benchmarks validating the positions of these entities (based on both LLM and human annotations), as well as a discussion of the case studies in which they can be used, including, e.g., AI explainability, political polarization and segregation, and media diets. To encourage reproducibility and further studies on the topic, we publicly release our anonymized data.
Antoine Vendeville, Jimena Royo-Letelier, Duncan Cassells, Jean-Philippe Cointet, Maxime Crépel, Tim Faveron, Théophile Lenoir, Béatrice Mazoyer, Benjamin Ooghe-Tabanou, Armin Pournaki, Hiroki Yamashita and Pedro Ramaciotti
Uncovering the structure and dynamics of information flow on the Telegram network, Thomas Louf, Aurora Vindimian and Riccardo Gallotti
More than a messaging service, Telegram has emerged as a central online social network in recent years. The network has a particular organisation, as it is split into groups of users, or channels, in which either only a set of administrators can post content (broadcast channels), or any user who joins the group can participate in a discussion discussion channels). It is however not its particular ontology –which in
itself would warrant scientific attention– that earned it the attention of researchers, but rather as it was pointed out as a haven for the spread of hate speech and disinformation [1, 2]. But while other social media that witness the diffusion of similar problematic content have been extensively studied, little is known about the structure and dynamics of information diffusion on Telegram.
In this work, we aim to provide further understanding on the mechanisms at play behind the growth of the Telegram network. We first do so through an extensive characterisation, using the open Pushshift Telegram dataset [3] as a support for our study. It features 29,000 public channels in which more than 2 million users shared about 300 million messages between September 2015 and October 2019. From this dataset, we built a temporal network with more than 7.5 million edges, an edge from channel i to channel j appearing at any timestamp when j forwarded a message from i. This representation thus encodes the flow of information between channels that happens via message-forwarding.
We first aggregate these directed, temporal edges into static, weighted ones to perform a topological analysis of this aggregate network. We thus find features typical of more traditional social networks: scale-free distributions of the in- and out-strengths, high clustering relatively to randomised versions of the network, and node-feature assortativity, in particular in terms of channel’s language, as also revealed through community detection [4] (see Fig. 1(a-b)). The temporal aspect of this information flow was also characterised. The time between two forwarding events in channels follows a piecewise power law distribution, with two distinct regimes for times inferior or superior to a day, as shown in Fig. 1(c). Remarkably, this distribution holds when considering channels within different activity ranges, and rescaling the times within each activity-group with its average activity. The distribution of burst train sizes E [5] featured in Fig. 1(d) also uncovers the bursty nature of the phenomenon, a trait commonly found in human communication.
This extensive characterisation allowed us to uncover mechanisms which are central to the emergence of the structure and temporality of information flow in this Telegram network. We then propose a model of network growth by exploiting these insights, namely that the forwarding phenomenon is bursty, driven
my a memory function, and that its structure is driven by focal and triadic closure. We adapt and combine existing topological [6] and temporal [5] models into a complete one which faithfully fits our observations, as we partially show in Fig. 1(c-d).
This work may enable further works striving to understand the spread of information on such networks, and how different external interventions may actually impact this spread.
A model for French voters, Antoine Vendeville
Models of opinion dynamics describe how opinions are shaped in various environments. While these models are able to replicate macroscopical opinion distributions observed in real-world scenarios, their capacity to align with data at the microscopical level remains mostly untested. We evaluate the capacity of the celebrated Voter Model to capture individual opinions in online social networks. We leverage a directed, weighted network of retweets between Twitter (now X) users, collected during the campaign of the 2017 French Presidential Elections. We uncover a strong correspondence between individual opinions in the equilibrium state of the model, and ground-truth party affiliations explicitly stated by the users in their publications and self-descriptions. Users are well separated along party lines in the opinion space of the model, and the model correctly identifies ground-truth party affiliations in 92.5% of cases. We also show that discord probabilities allow us to deduce with high accuracy whether or not two users support the same party. Neither the undirected or unweighted counterparts of the retweet network, nor the follow and mention networks produce comparable results. Our findings highlight the necessity for a fine-grained modelling approach, and contribute to the growing literature on the empirical validity of opinion dynamics models.
Modeling the Emergence of Shared-Issue Networks in the Era of Fragmentation Using Digital Log and Survey Data, Choi Sujin
As signified in the phrase, ‘no issue, no public,’ attention to shared issues brings strangers together. In today’s increasingly fragmented issue landscape, establishing a common understanding of issues becomes particularly crucial for social cohesion. This study investigates what promotes issue overlap between individuals engaged in personalized news curation.
Through stochastic actor-oriented modeling (SAOM) with digital log data and survey data, we investigate underlying mechanisms leading to the formation of shared-issue networks during South Korea’s presidential election. We also compare network dynamics between individuals with low and high involvement in politics, examining how the election’s increased issue salience and meta-narrative catalyzed joint interest differently across involvement levels.
Our findings reveal that the likelihood of forming shared-issue relations increases over time, when individuals are less susceptible to political homophily, accumulate greater political knowledge, and practice manual filtering to increase exposure to diverse news genres. Notably, specialized issue interests tend to develop from general interests, rather than vice versa. Individuals became involved in specific issues based on their broader understanding of related contexts, highlighting the significance of cultivating genre-level news interests and ensuring diversified genre exposure in news consumption patterns.
This research extends scholarly discourse beyond personalized news consumption to issue sharing mechanisms by shifting the focus from the individual level to the network level—an approach rarely taken in public opinion formation literature. It offers insights into the evolving public discourse landscape shaped by both low-and-high involvement citizens. Our findings also contribute to a deeper understanding of the current information dynamics.
Coevolutionary Axelrod Model with Weighted Overlap and Features Competition, Chiara Giaquinta
As it is well known [1], the influence of media on social opinion does not come from the fact that they succeed in telling people what to think of a given subject but from their success in imposing what people should think about; a situation known as the Agenda Setting Problem. In this way, topics discussed in the public arena are in competition to attract limited people’s attention. In order to model this problem one needs to study two coupled dynamical processes that have comparable time-scales: the evolution of the opinion of the actors, and that of the attention got by the different topics under discussion. Here we propose a multi-dimensional opinion dynamics model inspired by the Axelrod model [2], where each dimension corresponds to a given topic under discussion. Unlike the original model, the contribution of the topics to the overlap that rules social influence among the agents, is neither uniform nor constant. Instead, their relative importance is dynamical, modulated by the attention they attract. The overlap is weighted based on topic popularity, therefore coupled to a process where topics gain or lose attention over time.
We tested the model on stylized networks (Barabási-Albert and Erdős-Rényi) and also on real-world retweet networks of comparable sizes, for various values of the number of features F (here representing the number of topics under discussion), and the number of traits for each feature q (the number of different options the agents can choose for each topic). Preliminary results reveal that the size of the largest opinion cluster and convergence times heavily depend on the choice of the parameters F and q, with lower q and higher F promoting consensus, aligning with previous findings [3].
Competition among topics intensifies with increasing F , making dominance less likely. Moreover, consensus often forms on key features while persistent disagreements on others slow the dynamics. Finally we observe that the network structure significantly impacts the dynamics, leading to distinct outcomes in stylized random and community-structured networks. This work constitutes a new step towards the possibility of comparing theoretical models with empirical studies where the evolution of the attention given to different topics has been measured [4,5].
Chiara; Hernandez, Laura; Chavalarias, David
When the heterogeneous Hegselmann–Krause model meets community structure, Lucas Andrés Sobehart
Sobehart, Lucas A.; Hernandez, Laura; Moreno, Yamir Sobehart,
Since the first appearance of the Hegselmann-Krause bounded confidence model, many efforts have been made to include properties usually present in real-world systems into it. In particular, the inclusion of heterogeneous agents on systems with a subjacent network structure have shown prominent advances in recent years. In this work we use these studies as a preliminary background to understand the effect of networks with community structure on the steady state of the heterogenous Hegselmann-Krause opinion model. To this extent, we propose a novel benchmark of random networks composed of sparsely interconnected communities that can be used to generate a uniform ensemble of networks with the desired properties to study real-world social systems. Using this ensemble to create networks with high clustering, power-law degree distributions and small world behavior, we show that, when agents are divided into communities, having a large amount of individuals with high confidence bounds will make the opinion on each community converge to a weak consensus. Nevertheless, we also show that given that communities are sparsely connected, each community will have a different mean opinion, preventing the system to reach consensus as a whole. Finally, we observed that the system can also reach a state of polarization where each community becomes polarized between two different opinions.
TIDEM: Measuring Political Distance and Polarization through Retweet Networks in Spanish Regional Elections, Raul Broto
This study introduces TIDEM (Twitter Ideological Distance Estimation Method) a novel methodology for measuring ideological distances and evaluating political polarization using Twitter retweet networks. By using network-based analysis and spatial proximity within ForceAtlas2 layouts, the method captures ideological dynamics and provides a complementary perspective to traditional approaches such as self-placement surveys, the Chapel Hill Expert Survey (CHES), and Manifesto analysis. The methodology is applied to three Spanish regional elections (Catalonia and Madrid 2021, and Andalusia 2022) revealing consistent results through the three cases. A cross-election comparative demonstrates consistency in the relative positioning of left-right ideological blocks and the two major national parties (PP-PSOE). When evaluating TIDEM against traditional methods, the results indicate a strong correlation with self-ideological surveys across all three elections, except for the positioning of Cs in Madrid. However, comparisons with CHES and Manifesto data show mixed outcomes. Additionally, the analysis highlights the importance of regional context in shaping party positions, particularly in multi-dimensional ideological scenarios like Catalonia. Key findings indicate that TIDEM shows higher levels of polarization, likely due to the clustering effects inherent in retweet interactions. While traditional methods tend to position parties more centrally and show reduced distances between ideological blocks, our approach underscores the fluidity of public sentiment and the amplifying effects of online discourse. These results accentuate the potential of social media data as a valuable, scalable, and cost-effective source. TIDEM provides a relevant methodology for studying ideological distances and polarization. While it cannot replace traditional methods, it serves as a powerful complement.
Broto Cervera, Raul; Batlle, Albert; Pérez-Solà, Cristina
Ideological bias and information cascades on Twitter: evidence from French politicians, Shaden Shabayek
Shabayek, Shaden; Comola, Margherita Shabayek,
A social media analysis of the political interactions during the French 2022 presidential election, Ixandra Achitouv
Ixandra Achitouv and David Chavalarias
Beyond the Ideological Echo Chambers: Exploring the Dynamics of Diversity, and Demography in Digital Information Ecosystem, Burak Ozturan
Toggle Content goes here
Network pragmatic arenas: Analyzing a vaccine controversy on YouTube, Alexandre Doré
Beyond detection: disinformation and the amplification of toxic content in the age of social media
A Data-Driven Adaptive Approach to Supporting Fact-Checking and Mitigating Mis/Disinformation Through Domain Quality Evaluation, Kadkhoda Kaveh
Kadkhoda, Kaveh; Bertani, Anna; Louf, Thomas; Gallotti, Riccardo Kadkhoda,
The COVID-19 Infodemic on Twitter: Exploring Patterns and Dynamics across Countries, Anna Bertani
Bertani, Anna; Cortese, Alessandro; Pilati, Federico; Sacco, Pierluigi; Gallotti, Riccardo
A temporal-network perspective on the longitudinal analysis of online coordinated behaviour, Magnani Matteo
Rossi, Luca; Magnani, Matteo Magnani
Amplifying Extremism: Network Dynamics of Conspiratorial and Toxic Content in the Canadian Freedom Convoy Movement, Abul-Fottouh, Abul-Fottouh Deena
Abul-Fottouh, Deena; Eckardt, Jan; McLay, Rachel; Turgeon, Mathieu Abul-Fottouh
The Rise of the Right in the UK from Brexit to Tommy Robinson, Andrew Mackie
Cognitive Warfare on Social Networks, André Carvalho
Carvalho, André; Mourad, Aimãn; Conejero, Maria Carvalho
Streamwork Makes the Dream Work! Cross-Platform Collaboration and Community-Building Among Far-Right and Conspiracy-Ideologist Actors on Telegram and YouTube, Harald Sick
Sick, Harald; Jost, Pablo; Schmidt, Michael; Donner, Christian
The resilience of conspiracy theory networks on social media: from COVID-19 to the Russian invasion of Ukraine, Gronow, Antti; Malkamäki, Arttu; Mullo, Pamela
Studying information segregation on YouTube: Structural differences in the recommendation graph, Marijn A.; Erhard, Lukas; Kharazian, Zarine; Lamba, Manika
The role of moral values in the social media debate, Pietro Gravino (Sony CSL)
Brugnoli, Emanuele; Gravino, Pietro; Lo Sardo, D. Ruggiero; Loreto, Vittorio
The Diffusion of Propaganda on Social Media: Analyzing Russian and Chinese Influence on X (Twitter) during Xi Jinping’s visit to Moscow in 2023, Luliia Alieva
How algorithms recommend political content on social network, Tim Faverjon
Faverjon, Tim; Ramaciotti, Pedro
Unveiling emerging moderation dynamics in Mastodon’s federated instance network, Beatriz Arregui Garcia
Arregui Garcia, Beatriz; La Cava, Lucio; Baqir, Anees; Gallotti, Riccardo; Meloni, Sandro
Sampled datasets risk substantial bias in the identification of political polarization on social media, Gabriele Di Bona et al.
Gabriele Di Bona, Emma Fraxanet, Björn Komander, Andrea Lo Sasso, Virginia Morini, Antoine Vendeville, Max Falkenberg and Alessandro Galeazzi
The study of phenomena related to public opinion online and especially political polarization garners significant interest in Computational social sciences. The undertaking of several studies of political phenomena in social media mandates the operationalization of the notion of political stance of users and contents involved. Relevant examples include the study of segregation and polarization online, or the study of political diversity in content diets in social media. While many research designs rely on operationalizations best suited for the US setting, few allow addressing more general design, in which users and content might take stances on multiple ideology and issue dimensions, going beyond traditional Liberal-Conservative or Left-Right scales. To advance the study of more general online ecosystems, we present a methodology for the computation of multidimensional political positions of social media users and web domains. We perform a case study on a large-scale X/Twitter population of users in the French political Twittersphere and web domains, embedded in a political space spanned by dimensions measuring attitudes towards immigration, the EU, liberal values, elites and institutions, nationalism and the environment. We provide several benchmarks validating the positions of these entities (based on both LLM and human annotations), as well as a discussion of the case studies in which they can be used, including, e.g., AI explainability, political polarization and segregation, and media diets. To encourage reproducibility and further studies on the topic, we publicly release our anonymized data.
Science dynamics : from reconstruction to social processes
The robust-fragile duality of the ATLAS collaboration network, Rodríguez-Casañ, Rubén et al.
Rodríguez-Casañ, Rubén; Palazzi, María; Solé-Ribalta, Albert; Canals, Agustí; Borge-Holthoefer, Javier Borge-Holthoefer, Dr. Javier
Do states make scientific fields? McMahan, Peter; Lévesque, Gabriel Lévesque, Gabriel
Gender differences in scientific recognition: authorship and acknowledgment, Yukie; Kusumegi, Keigo; Acuna, Daniel E. Sano, Dr. Yukie
The stagnation of a science, Gillespie, Ryder Gillespie, Ryder
Project ARCH: Optimizing the Design of Virtual Scientific Ecosystems for Team Formation and Innovation, Zajdela, Emma Rosa; Mojeed, Sodiq Abiodun; Kim, Joan Zajdela
Loading...