Réseaux sociaux numériques / Digital social networks

Seminar organized by Francis Bloch from November 7 to 11, 2022.

Participants

Francis BLOCH (Université Paris 1, FRANCE), Yann BRAMOULLE (Université d’Aix-Marseille, FRANCE), Damon CENTOLA (University of Pennsylvania – ÉTATS-UNIS), Gabrielle DEMANGE (PSE – EHESS – Paris, FRANCE), Bhaskar DUTTA (Ashoka University and University of Warwick – Coventry, ROYAUME-UNI), Marcin DZIUBIŃSKI (University of Warsaw – Warsaw, POLOGNE), Dean ECKLES (MIT – Cambridge, ÉTATS-UNIS), Sanjeev GOYAL (University of Cambridge, ROYAUME-UNI et Nyu Abu Dhabi, EMIRATS ARABES UNIS), Nicole IMMORLICA  (Microsoft Research – Cambridge, MA, USA, ÉTATS-UNIS), Matthew JACKSON (Stanford University – California, ÉTATS-UNIS), Emmanuel LAZEGA (Sciences Po – Paris, FRANCE), Ankur MANI (University of Minnesota – Minneapolis, ÉTATS-UNIS), Camille ROTH (CNRS / Centre Marc Bloch – Berlin, Germany, ALLEMAGNE), Shaden SHABAYEK (Sciences Po – Paris, FRANCE), Yves ZENOU (Monash University – Caulfield, AUSTRALIE).

Résumé

Le séminaire a réuni quinze chercheurs économistes, sociologues et informaticiens travaillant sur les réseaux sociaux numériques. Les participants ont discuté les travaux empiriques récents exploitant les données de Facebook et Twitter pour étudier l’influence des réseaux sociaux sur les résultats économiques et les situations politiques. Plusieurs présentations ont été consacrées à des modélisations du comportement des plates-formes. Certaines étudient la polarisation des opinions engendrée par les réseaux sociaux, s’appuyant sur le modèle de formation des opinions de De Groot. D’autres présentations ont couvert divers aspects des réseaux sociaux conventionnels (non numériques). Grâce à l’environnement exceptionnel des Treilles, le séminaire a rencontré un grand succès, comme en témoignent les nombreuses discussions et projets de collaboration entre chercheurs venant de trois disciplines différentes.

Summary

The seminar brought together fifteen scholars from economics, sociology and computer science working on digital social networks. The participants discussed recent advances in the empirical analysis of Facebook and Twitter data and the influence of social media on economic and political outcomes. Other presentations were devoted to theoretical models of the platforms’ behavior. Polarization of opinions in social networks was also studied by extending the De Groot model of opinion formation to allow for dissensus. Other presentations covered a diverse range of topics on conventional social networks. Due to the wonderful environment at Les Treilles, the seminar was an unmitigated success, with lively discussions and seeds of collaboration among researchers from three distinct academic communities.

Report

How do friendships on Facebook influence agents’ economic prospects? Has Twitter increased polarization of the public debate? Is the business model of social media platforms viable and how will it evolve over time? To answer those questions, and raise new ones, a group of fifteen scholars from six different countries (France, United States, United Kingdom, India, Poland, Australia) and three different academic communities (economics, computer science and sociology) met at Les Treilles between November 7 and November 11, 2022. Researchers were at different levels of seniority, with a large majority of senior researchers (thirteen out of fifteen) and two junior researchers. Three out of the fifteen participants were women.

The objective of the seminar was to foster exchange among economists, sociologists and computer scientists working on digital (and conventional) social networks. Each participant gave a presentation of one hour and a half, and there was ample time for discussion both in the seminar room, during meals, walks on the beautiful domain of Les Treilles and during the excursion to the Abbaye of Thoronet on Thursday afternoon. The objective was fully achieved, with lively discussions involving recent work by economists (Matthew Jackson) on Facebook data, and by computer scientists (Dean Eckles) and sociologists (Camille Roth) on Twitter data. The participants uncovered strong connections between the theoretical work conducted by economists and computer scientists, and the empirical work conducted by economists, sociologists and computer scientists on digital social networks.

A first group of presentations dealt with the analysis of Facebook and Twitter data. Matt Jackson shows the importance of homophily – the tendency to form links with agents with similar characteristics – on social and economic prospects in the United States. His work documents the geographical variation of social capital and homophily at the county level and shows that economic connectedness (the fraction of friends with high income) is the main determinant of social mobility. Dean Eckles focuses attention on long ties, for which connected agents do not have common neighbors. He identifies the fraction of long ties across geographical counties in the US and Mexico and shows that long ties are correlated with positive economic outcomes. Camille Roth studies the polarization of messages on Twitter, by distinguishing among separate linguistic groups in Europe. He shows that some topics result in more integration and others in more polarization of the linguistic communities. In a separate study of Twitter data, he investigates the difference between quotes and re-tweets, and shows that re-tweets tend to polarize the political spectrum whereas quotes typically re-center the political debate.

The study of the optimal decisions of social media platforms was the topic of three presentations, which proposed different theoretical models of platform behavior. Sanjeev Goyal introduces a model of verification and information sharing in networks. Inspired by the fact-checking tools in digital social networks, Goyal distinguishes between two situations: one where verification efforts are substitutes (the more the other agents check, the less incentive an agent must check) and complements (the more the other agents check, the more incentive an agent has to check because his reputation is on the line). Gabrielle Demange distinguishes between two strategies of a platform: targeting (when the platform discriminates among users) and visibility (when the platform promotes or deletes links) and two different models of engagement. She considers both situations where participants engage in a single activity and in two activities (contribute and read). Ankur Mani analyzes the role of social connections in viral marketing with many consumers. He identifies network architectures for which the optimal price is uniform across nodes, so that the value of information is zero. A study of large random graphs shows that the value of information in Erdos Renyi network is highest when the level of connectivity is low and compares the value of the network for different networks with scale-free distributions.

Two presentations were devoted to the formation of polarized opinions in social networks, proposing variants of the De Groot model. Nicole Immorlica extends the model by letting agents receive in addition to their neighbors’ opinion, a newsfeed from a media outlet. There are two media outlets (one conservative and one liberal) and the platform chooses the outlet according to the agent’s current opinion. The work shows that high levels of polarization are obtained if the influence of the media outlet is sufficiently large. Shaden Shabayek distinguishes between two types of agents. Elite agents form opinions by either being attracted to agents who are close to them or being repelled by agents who have different opinions. Normal agents form their opinions simply by listening to their neighbors. The analysis of the model shows that dissensus and polarization can occur for some parameter values.

Other presentations were based on conventional rather than digital social networks. Yann Bramoullé characterizes which transfers emerge from the rational decisions of altruistic agents who care for their neighbors (friends, parents..) in a social network. Damon Centola reports on the importance of “wide bridges” involving cohesive groups of agents to spread behavior in complex contagion. Emmanuel Lazega studies the interaction between personal links and links across organizations in an extensive case study of social links between researchers working on cancer in France in the 2000s. Bhaskar Dutta and Marcin Dziubinski discuss models involving two players: an attacker (or seeker) and a defender (or hider) in networks.  Bhaskar Dutta focuses attention on an attacker whose objective is to transport contraband goods on a fixed network, whereas Marcin Dziubinski studies the problem of a seeker who explores a network to find a hidden object. Yves Zenou reports on a field experiment done in Bangladesh to identify whether students who are more central in the social network generate more educational spillovers.