Collective behaviour in digital societies

The number of social and economic activities on digital platformshas been increasing since the last two decades, and especiallyin the last decade. Several such platforms also provideopportunities for interactions among participating users,either as a part of the primary activities on such platformsor as a secondary feature. Such interactions form the basisof the emergence of collective behaviour on individual platforms,wherein users’ interactions affect the aggregate state ofthe platform and at the same time the aggregate state of platformfeeds into how users interact among themselves. Digitalplatforms turn into digital societies by providing many waysto conduct users’ interactions driven activities, that were traditionallyconducted in the physical world, in efficient andscalable ways. Apparently, these two societies – digital andreal world – exist simultaneously and have feedback effectsin shaping outcomes in both the societies on various factorsincluding behaviour, beliefs, opinions, and others. This thesiscontains three essays including (1) peer influence on creationof new projects in the learning environment Scratch, (2) polarizationof climate change beliefs due to homophily in interactionson the social media Twitter, and (3) effects on collectiveattention due to political framing of climate change by thenews media Guardian. Particular emphasis is laid on statisticalinference of the effects and hypothesizing mechanismsbehinds such effects.The production and consumption of projects in the Scratchcommunity, a digital platform developed by MIT Media Labwhere users, usually young children, learn to program by creatingand sharing projects, is analysed using the data for thefirst five years after its launch. In particular, investigationis done to discover if users are influenced by the popularityof their peers’ projects and their peers’ preferences for consumingspecific baskets of projects. The major challenges inthis type of analysis is to provide parsimonious models forcomplexities of interactions on the platform and to disentanglepeer influence from homophily in the vast network of behavioursand friendships. Homophily is a term widely usedin social networks studies to describe friendship or tie formationsthat arise due to similarities in behaviours or commonattributes between participating agents in the formationof such ties. The analysis reveals that while Scratchers’ consumptionpreference is not influenced by their peers, the popularityof their projects is significantly influenced by theirpeers in short and long terms. A large proportion of the influencefrom peers is mediated via Scratchers’ creation of newprojects, which highlights Scratchers’ subsequent decisionsin response to existing popularity of peers’ projects. Theseinsights can potentially help in incorporation of behaviourdrivendesigns in future educational technologies.Producer-consumer business models is at the heart of severalsocial networking sites. Activities on such sites range frommeeting friends, exchanging messages, propagating messages,advertisements, and others. Lack of regulations on informationposting and limitations of computer-assisted informationchecks therefore provide opportunities for people’s beliefsto be polarized due to the spread of fake information insuch social networks. Homophily in communication createsgroups of people or agents with bounded beliefs about thereality, and hence can polarize a society. Such homophily indigital media has been termed as echo chambers which intuitivelypromotes the notion that people hear nothing morethan what they already believe. Using evidence from 11 yearsof Twitter conversations on the climate change topic, an empirical analysis is conducted on the effect of homophily incommunication patterns on the polarization of beliefs aboutthe reality of climate change. The analysis reveals a counterintuitiveresult that increasing levels of homophily in communicationpredicts decreasing levels of polarization in beliefsin the long run. To understand better the mechanism ofthe effect of homophily on polarization, a model is developedthat shows how polarization can emerge due to the joint effectsof precision of misinformation propagating in a socialnetwork and homophily in communication among agents inthe social network with differing beliefs. Credibility of fakenews, modelled as precision of misinformation, circulating inthe social network can lead to acceptance of the fake news(depending on agents’ susceptibility to it), thereby changingbeliefs and creating polarization. The model shows that fakenews can not polarize the society unless it has a minimallevel of credibility, irrespective of the level of homophily incommunication patterns. This throws a light on perhaps themost intuitive but usually the forgotten factor of information– credibility. While the results show that the climate changesceptic exchanges of messages on the social media Twitter donot carry enough credibility to create large scale polarizationin society, they also provide useful indications to directly orindirectly quantify the emergence of credibility of informationin digital platforms, and also to shift attention of technologyfrom detecting fake stories to detecting fake storieswhich the society might find them to be credible.Digital news platforms have become outlets for engaging discussions.They provide journalists and publishers with severaldimensions to gauge the acceptance of their articles andat the same time provides readers with simple tools to participatein discussions. It has been long acknowledged thatnews media frame their articles. As an instance of politicalframing, to understand how the politicization climate change articles influences the collective attention and discussion onsuch articles, articles published in the Guardian until 2018 areused to examine whether and how politicization influencesreaders’ collective discussion. The results suggest that (unknown)factors of perception associated to an article when itis categorized in ‘Politics’ section positively impacts the collectiveattention and engagement received on articles. Estimatesalso suggest that mentions of political inclinations ofthe entities within an article impact such collective attention.In particular, a large proportion of such participation is foundto be mediated by discussions becoming politicized by pastcontextual entities related to the article but strictly absent inthe article, thereby suggesting a temporal effect of perception.In addition, a large proportion of the impact of political mentionson users’ engagement is mediated by users who join discussionbeing influenced by past politically oriented contextualentities. Although no evidence is found to support thatauthors or journalists might be enjoying increasing marginalbenefits from collective attention to their articles that resultfrom their choices about mentions of political entities withinthe main texts of climate change articles, the results highlightthat their choices do impact the readers’ perceptions and participations(and potentially the intensity of climate change actionin real world).Overall, this series of research has contributed to improvinghow collective behaviour is shaped in digital societies in relationto peer influence in educational media, polarization ofbeliefs in social media, and political framing of articles innews media. Hopefully, these results would be of interest togeneral audience and researchers in fields of social sciences,economics, marketing, media and communications, and applieddata science.