No description available.

This thesis investigates how group-level differences in willingness of opinion expression shape the extent to which certain standpoints are visible in public debate online. Against the backdrop of facilitated communication and connection to like-minded others through digital technologies, models and methods are developed and case studies are carried out – by and large from a network perspective. To this end, we first propose a model of opinion dynamics that examines social- structural conditions for public opinion expression or even predominance of different groups. The model focuses not on opinion change, but on the decision of individuals whether to express their opinion publicly or not. Groups of agents with different, fixed opinions interact with each other, changing the willingness to express their opinion according to the feedback they receive from others. We formulate the model as a multi-group game, and subsequently provide a dynamical systems perspective by introducing reinforcement learning dynamics. We show that a minority can dominate public discourse if its internal connections are sufficiently dense. Moreover, increased costs for opinion expression can drive even internally well-connected groups into silence. We then focus on how interaction networks can be used to infer political and social positions. For this purpose, we develop a new type of force-directed network layout algorithm. While being widely used, a rigorous interpretation of the outcomes of existing force-directed algorithms has not been provided yet. We argue that interpretability can be delivered by latent space approaches, which have the goal of embedding a network in an underlying social space. On the basis of such a latent space model, we derive a force-directed layout algorithm that can not only be used for the spatialisation of generic network data – exemplified by Twitter follower and retweet networks, as well as Facebook friendship networks – but also for the visualization of surveys. Comparison to existing layout algorithms (which are not grounded in an interpretable model) reveals that node groups are placed in similar configurations, while said algorithms show a stronger intra-cluster separation of nodes, as well as a tendency to separate clusters more strongly in retweet networks. In two case studies, we observe actual public debate on the social media platform Twitter – topics are the Saxon state elections 2019, and violent riots in the city of Leipzig on New Year’s Eve in the same year. We show that through the interplay of retweet and reply networks, it is possible to identify differences in willingness of opinion expression on the platform between opinion groups. We find that for both events, propensities to get involved in debate are asymmetric. Users retweeting far-right parties and politicians are significantly more active, making their positions disproportionately visible. Said users also act significantly more confrontational in the sense that they reply mostly to users from different groups, while the contrary is not the case. The findings underline that naive reliance on what others express online can be collectively dangerous, especially in an era in which social media shapes public discourse to an unprecedented extent.

De nombreux domaines ont intérêt à étudier les points de vue exprimés en ligne, que ce soit à des fins de marketing, de cybersécurité ou de recherche avec l'essor des humanités numériques. Nous proposons dans ce manuscrit deux contributions au domaine de la fouille de points de vue, axées sur la difficulté à obtenir des données annotées de qualité sur les médias sociaux. Notre première contribution est un jeu de données volumineux et complexe de 22853 profils Twitter actifs durant la campagne présidentielle française de 2017. C'est l'un des rares jeux de données considérant plus de deux points de vue et, à notre connaissance, le premier avec un grand nombre de profils et le premier proposant des communautés politiques recouvrantes. Ce jeu de données peut être utilisé tel quel pour étudier les mécanismes de campagne sur Twitter ou pour évaluer des modèles de détection de points de vue ou des outils d'analyse de réseaux. Nous proposons ensuite deux modèles génériques semi-supervisés de détection de points de vue, utilisant une poignée de profils-graines, pour lesquels nous connaissons le point de vue, afin de catégoriser le reste des profils en exploitant différentes proximités inter-profils. En effet, les modèles actuels sont généralement fondés sur les spécificités de certaines plateformes sociales, ce qui ne permet pas l'intégration de la multitude de signaux disponibles. En construisant des proximités à partir de différents types d'éléments disponibles sur les médias sociaux, nous pouvons détecter des profils suffisamment proches pour supposer qu'ils partagent une position similaire sur un sujet donné, quelle que soit la plateforme. Notre premier modèle est un modèle ensembliste séquentiel propageant les points de vue grâce à un graphe multicouche représentant les proximités entre les profils. En utilisant des jeux de données provenant de deux plateformes, nous montrons qu'en combinant plusieurs types de proximité, nous pouvons correctement étiqueter 98% des profils. Notre deuxième modèle nous permet d'observer l'évolution des points de vue des profils pendant un événement, avec seulement un profil-graine par point de vue. Ce modèle confirme qu'une grande majorité de profils ne changent pas de position sur les médias sociaux, ou n'expriment pas leur revirement. / Numerous domains have interests in studying the viewpoints expressed online, be it for marketing, cybersecurity, or research purposes with the rise of computational social sciences. We propose in this manuscript two contributions to the field of stance detection, focused around the difficulty of obtaining annotated data of quality on social medias. Our first contribution is a large and complex dataset of 22853 Twitter profiles active during the French presidential campaign of 2017. This is one of the rare datasets that considers a non-binary stance classification and, to our knowledge, the first one with a large number of profiles, and the first one proposing overlapping political communities. This dataset can be used as-is to study the campaign mechanisms on Twitter, or used to test stance detection models or network analysis tools. We then propose two semi-supervised generic stance detection models using a handful of seed profiles for which we know the stance to classify the rest of the profiles by exploiting various proximities. Indeed, current stance detection models are usually grounded on the specificities of some social platforms, which is unfortunate since it does not allow the integration of the multitude of available signals. By infering proximities from differents types of elements available on social medias, we can detect profiles close enough to assume they share a similar stance on a given subject. Our first model is a sequential ensemble algorithm which propagates stances thanks to a multi-layer graph representing proximities between profiles. Using datasets from two platforms, we show that, by combining several types of proximities, we can achieve excellent results. Our second model allows us to observe the evolution of profiles' stances during an event with as little as one seed profile by stance. This model confirms that a large majority of profiles do not change their stance on social medias, or do not express their change of heart.

abstract: What causes social systems to resist change? Studies of the emergence of social complexity in archaeology have focused primarily on drivers of change with much less emphasis on drivers of stability. Social stability, or the persistence of social systems, is an essential feature without which human society is not possible. By combining quantitative modeling (Exponential Random Graph Modeling) and the comparative archaeological record where the social system is represented by networks of relations between settlements, this research tests several hypotheses about social and geographic drivers of social stability with an explicit focus on a better understanding of contexts and processes that resist change. The Valencian Bronze Age in eastern Spain along the Mediterranean, where prior research appears to indicate little, regional social change for 700 years, serves as a case study. The results suggest that social stability depends on a society’s ability to integrate change and promote interdependency. In part, this ability is constrained or promoted by social structure and the different, relationship dependencies among individuals that lead to a particular social structure. Four elements are important to constraining or promoting social stability—structural cohesion, transitivity and social dependency, geographic isolation, and types of exchange. Through the framework provided in this research, an archaeologist can recognize patterns in the archaeological data that reflect and promote social stability, or lead to collapse. Results based on comparisons between the social networks of the Northern and Southern regions of the Valencian Bronze Age show that the Southern Region’s social structure was less stable through time. The Southern Region’s social structure consisted of competing cores of exchange. This type of competition often leads to power imbalances, conflict, and instability. Strong dependencies on the neighboring Argaric during the Early and Middle Bronze Ages and contributed to the Southern Region’s inability to maintain social stability after the Argaric collapsed. Furthermore, the Southern Region participated in the exchange of more complex technology—bronze. Complex technologies produce networks with hub and spoke structures highly vulnerable to collapse after the destruction of a hub. The Northern Region’s social structure remained structurally cohesive through time, promoting social stability. / Dissertation/Thesis / Webpage with data tables and R code / Doctoral Dissertation Anthropology 2020

abstract: Internet memes have become a widespread tool used by people for interacting and exchanging ideas over social media, blogs, and open messengers. Internet memes most commonly take the form of an image which is a combination of image, text, and humor, making them a powerful tool to deliver information. Image memes are used in viral marketing and mass advertising to propagate any ideas ranging from simple commercials to those that can cause changes and development in the social structures like countering hate speech. This work proposes to treat automatic image meme generation as a translation process, and further present an end to end neural and probabilistic approach to generate an image-based meme for any given sentence using an encoder-decoder architecture. For a given input sentence, a meme is generated by combining a meme template image and a text caption where the meme template image is selected from a set of popular candidates using a selection module and the meme caption is generated by an encoder-decoder model. An encoder is used to map the selected meme template and the input sentence into a meme embedding space and then a decoder is used to decode the meme caption from the meme embedding space. The generated natural language caption is conditioned on the input sentence and the selected meme template. The model learns the dependencies between the meme captions and the meme template images and generates new memes using the learned dependencies. The quality of the generated captions and the generated memes is evaluated through both automated metrics and human evaluation. An experiment is designed to score how well the generated memes can represent popular tweets from Twitter conversations. Experiments on Twitter data show the efficacy of the model in generating memes capable of representing a sentence in online social interaction. / Dissertation/Thesis / Masters Thesis Computer Science 2020

abstract: Millions of users leave digital traces of their political engagements on social media platforms every day. Users form networks of interactions, produce textual content, like and share each others' content. This creates an invaluable opportunity to better understand the political engagements of internet users. In this proposal, I present three algorithmic solutions to three facets of online political networks; namely, detection of communities, antagonisms and the impact of certain types of accounts on political polarization. First, I develop a multi-view community detection algorithm to find politically pure communities. I find that word usage among other content types (i.e. hashtags, URLs) complement user interactions the best in accurately detecting communities. Second, I focus on detecting negative linkages between politically motivated social media users. Major social media platforms do not facilitate their users with built-in negative interaction options. However, many political network analysis tasks rely on not only positive but also negative linkages. Here, I present the SocLSFact framework to detect negative linkages among social media users. It utilizes three pieces of information; sentiment cues of textual interactions, positive interactions, and socially balanced triads. I evaluate the contribution of each three aspects in negative link detection performance on multiple tasks. Third, I propose an experimental setup that quantifies the polarization impact of automated accounts on Twitter retweet networks. I focus on a dataset of tragic Parkland shooting event and its aftermath. I show that when automated accounts are removed from the retweet network the network polarization decrease significantly, while a same number of accounts to the automated accounts are removed randomly the difference is not significant. I also find that prominent predictors of engagement of automatically generated content is not very different than what previous studies point out in general engaging content on social media. Last but not least, I identify accounts which self-disclose their automated nature in their profile by using expressions such as bot, chat-bot, or robot. I find that human engagement to self-disclosing accounts compared to non-disclosing automated accounts is much smaller. This observational finding can motivate further efforts into automated account detection research to prevent their unintended impact. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2019

Web surveys in higher education are particularly important for assessing the quality of academic teaching and learning. Traditionally, mainly quantitative data is used for quality assessment. Increasingly, questions are being raised about the impact of attitudes of the individuals involved. Therefore, especially the analysis of open-ended text responses in web surveys offers the potential for impact evaluation. Despite the fact that qualitative text mining and sentiment analysis are being introduced in other research areas, these instruments are still slowly gaining access to evaluation research. On the one hand, there is a lack of methodological expertise to deal with large numbers of text responses (e.g. via semantic analysis, linguistically supported coding, etc.). On the other hand, deficiencies in interdisciplinary expertise are identified in order to be able to contextualize the results. The following contribution aims to address these issues. The presentation will contribute to the field of impact evaluation and reveals methodological implications for the development of text mining and sentiment analysis in evaluation processes.

The WikiLeaks Afghanistan war logs contain nearly 77,000 reports of incidents in the US-led Afghanistan war, covering the period from January 2004 to December 2009. The recent growth of data on complex social systems and the potential to derive stories from them has shifted the focus of journalistic and scientific attention increasingly toward data-driven journalism and computational social science. In this paper we advocate the usage of modern statistical methods for problems of data journalism and beyond, which may help journalistic and scientific work and lead to additional insight. Using the WikiLeaks Afghanistan war logs for illustration, we present an approach that builds intelligible statistical models for interpretable segments in the data, in this case to explore the fatality rates associated with different circumstances in the Afghanistan war. Our approach combines preprocessing by Latent Dirichlet Allocation (LDA) with model trees. LDA is used to process the natural language information contained in each report summary by estimating latent topics and assigning each report to one of them. Together with other variables these topic assignments serve as splitting variables for finding segments in the data to which local statistical models for the reported number of fatalities are fitted. Segmentation and fitting is carried out with recursive partitioning of negative binomial distributions. We identify segments with different fatality rates that correspond to a small number of topics and other variables as well as their interactions. Furthermore, we carve out the similarities between segments and connect them to stories that have been covered in the media. This gives an unprecedented description of the war in Afghanistan and serves as an example of how data journalism, computational social science and other areas with interest in database data can benefit from modern statistical techniques. (authors' abstract)

This thesis investigates the role of network structure, brokerage, and framing in high-risk collective action. I use the protest movement that emerged in Russia following falsified national elections in 2011 and 2012 as an empirical case study. I draw on a unique dataset of nearly 30,000 online documents and the linking structure of over 3,500 Russian Web sites. I employ a range of computational social science methods, including Exponential Random Graph Modeling, an advanced statistical model for social networks, social network analysis, machine learning, and latent semantic analysis. I address three research questions in this thesis. The first asks if a protest network challenging a hybrid regime will have a polycentric or hierarchical structure, and if that structure changes over time. Polycentric networks are conducive to high-risk collective action and are robust to the targeted removal of key nodes, while hierarchical networks can more easily mobilize protesters and spread information. I find that the Russian protest network has a polycentric structure only at the beginning of the protests, and moves towards a less effective hierarchical structure as the movement loses popular support. The second research question seeks to understand if brokered text is actually novel, and if that text is more novel in polycentric networks than in hierarchical ones. Brokers are the individuals or nodes in a network that connect disparate groups through weak ties and close structural holes. Brokers are advantageous because they have access to and spread novel information. I find that the text among nodes in brokered relationships is indeed novel, but that information novelty decreases when networks have a hierarchical structure. The last research question asks if a protest movement in a high-risk political setting can be more successful than the government at spreading its preferred frames, and within such a movement, whether moderate or extremist framing is more prevalent. I find that the opposition is far more effective than the government in spreading its frames, even when the government organizes massive counter protests. Within the movement, moderates are more likely to have their framing adopted online than extremists, unless violence occurs at protests. The findings suggest that movements should build flatter, more diffuse networks by ensuring that brokers tie together diverse protest constituencies. The findings also provide evidence against those who claim that authoritarian governments are more effective in shaping online discourse than oppositional movements, and also suggest that movements should advance moderate framing in order to attract a wider base of support among the general population.

Several calamities occur throughout the world each year, resulting in varying losses. Disasters wreak havoc on infrastructures and impair operation. They result in human deaths and injuries and stress people's mental and emotional states. These negative impacts of natural disasters induce significant economic losses, as demonstrated by the $ 423 billion loss in 2011 in Tohoku, Japan, and the $ 133 billion loss in hurricane Harvey, U.S.A. Every year, hurricanes and tropical storms result in 10,000 human deaths worldwide. To mitigate losses, communities' readiness, flexibility, and resilience must be strengthened. To this end, appropriate techniques for forecasting a community's capacity and functionality in the face of impending crises must be developed and suitable community resilience metrics and their quantification must be established. Collaboration between critical infrastructures such as power systems and emergency services and social networks is critical for building a resilient community. As a result, we require metrics that account for both the social and infrastructure aspects of the community. While the literature on critical infrastructures such as power systems discusses the effect of social factors on resilience, they do not model these social factors and metrics due to their complexity. On the other hand, it turns out that the role of critical infrastructures and some critical social characteristics is overlooked in the computational social science literature on community resilience. Thus, this dissertation presents a multi-agent socio-technical model of community resilience, taking into account the interconnection of power systems, emergency services, and social communities. We offer relevant measures for each section and describe dynamic change and its dependence on other metrics using a variety of theories and expertise from social science, psychology, electrical engineering, and emergency services. To validate the model, we used data on two hurricanes (Irma and Harvey) collected from Twitter, GoogleTrends, FEMA, power utilities, CNN, and Snopes (a fact-checking organization). We also describe methods for quantifying social metrics such as anxiety, risk perception, cooperation using social sensing, natural language processing, and text mining tools. / Doctor of Philosophy / Power systems serve social communities that consist of residential, commercial, and industrial customers. The social behavior and degree of collaboration of all stakeholders, such as consumers, prosumers, and utilities, affect the level of preparedness, mitigation, recovery, adaptability, and, thus, power system resilience. Nonetheless, the literature pays scant attention to stakeholders' social characteristics and collaborative efforts when confronted with a disaster and views the problem solely as a cyber-physical system. However, power system resilience, which is not a standalone discipline, is inherently a cyber-physical-social problem, making it complex to address. To this end, in this dissertation, we develop a socio-technical power system resilience model based on neuroscience, social science, and psychological theories and use the threshold model to simulate the behavior of power system stakeholders during a disaster. We validate our model using datasets of hurricane Harvey of Category 4 that hit Texas in August 2017 and hurricane Irma of Category 5 that made landfall in Florida in September 2017. We retrieve these datasets from Twitter and GoogleTrend and then apply natural language processing and language psychology analysis tools to deduce the social behavior of the end-users.