Observing debates across media platforms: tools and models - an ODYCCEUS online event


Sven Banisch
Max Planck Institute for Mathematics in the Sciences, Germany
Biased processing and opinion polarization: experimental refinement and plausibility of argument communication theory.
We look at data from an experiment on biased argument processing from the perspective of the cognitive architecture employed in argument communication models of collective opinion formation. The empirical experiment realized in the context of attitudes toward energy reveals a strong tendency to consider arguments aligned with the current attitude more persuasive and to downgrade those speaking against the current attitude. This is integrated into a theoretical model of cognitive agents by assuming that the coherence of an argument with the current attitude controls the probability to adopt it and to change the attitude accordingly. The strength of this bias is included as a free parameter which can be estimated from experimental data. We find a clear signature of moderate biased processing. Simulations with interacting cognitive agents that exchange arguments show that deliberation with a moderate level of biased processing leads to efficient group decision processes, whereas groups without bias may be trapped in long processes of indecision and strong biases result in persistent intra-group conflict. We relate the opinion distributions emerging in the model to surveyed attitude distributions and find a good match in transitory periods of the model. We shall explore this in an online tool which allows to compare the model dynamics to empirical attitude distributions.

Anna Keuchenius
University of Amsterdam, Netherlands
(joint work with Petter Tornberg and Justus Uitermark)
Negative ties in social networks: the role of contention in group polarization dynamics of the Dutch Black Pete debate
With many of today’s pressing and polarized debates taking place online, specifically on online social networks, there is an unprecedented opportunity to study the dynamics of people self-organizing into groups and agreeing or arguing against each other over important topics. However, previous studies into polarization on social networks have barely considered negative interactions between users: the focus has been on positive interaction, often studied through the lens of networks with positive ties. In this research, we show what the explicit inclusion of negative interactions online can contribute to the study of large scale social dynamics, particularly to that of partisan polarization. We illustrate our approach by a case study of the debate about ’Black Pete’, a caricature from a Dutch tradition with racist connotations, that has divided Dutch society and is an ongoing topic of conflict. We use a dataset of 400.000 Tweets on this topic, many of which contain mentions to specific other users. By means of machine learning, we identify users’ positions in the debate and the sign of user-to-user interactions. Signed network analysis reveals important differences between the group compositions in this network compared to the non signed counterpart, the retweet network, which is typically used for studies into polarization on Twitter. Furthermore, the signs allow for the identification of a rich variety of roles in this debate, defined by combinations ingroup and outgroup sentiments, that stay hidden in an unsigned analysis. Beyond the specific findings of this case study, the present research elucidates the importance of negative interactions in the social realm and illustrates the potential of various research avenues made possible by a signed analysis of contentious debates.

Simon Levis Sullam
Università Ca' Foscari, Italy
Antisemitic Discourse and Narrative, or the Antisemite as Novelist
Modern antisemitism in its founding moment, between the 1880s and the early 1920s, was created and spread as a narrative: from Edouard Drumont, La France juive (1886) to the “Protocols of the Elders of Zion” (1903-1919). A key feature of these narratives was conspirationism: they offered the revelation and description of a supposed hidden role of Jews in world affairs. But narrative is a more general mode of organization, interpretation and transmission of events and knowldege, and anti-Jewish hostility has been based on a set of repetitive narratives or plots at least since the Middle Ages. This papers explores the influence of literary models, and the use of literary and narrative devices, in the creation and spread of modern antisemitism.

Stijn Peeters
University of Amsterdam, Netherlands
(joint work with Sal Hagen)
Multi- and cross-platform data analysis with 4CAT (4cat.oilab.eu)


In this workshop, we discuss the 4CAT Capture and Analysis Toolkit. We start with a discussion of the design philosophy of 4CAT, a tool built to offer 'traceable' research and avoid the 'black-boxing' problems often inherent to tool-based research. Next is an exploration of the tool's web interface and the available data sources (such as 4chan, Reddit or Telegram) and analyses (such as generating word embedding models, simple frequency analysis or generating image walls), as well as its synergy with other tools. Finally we briefly discuss the possibilities for extension of 4CAT or using it via an API.

4CAT is a web tool and requires no prior installation. We recommend you register for an account on our 4CAT instance in advance, so you can engage with the tools discussed in the workshop directly.

Stijn Peeters
University of Amsterdam, Netherlands
(joint work with Marc Tuters and Tom Willaert (Vrije Universiteit Brussel))
Historicising the Reactionary Turn: tracing the radicalisation of image board culture
‘Image board culture’, originating from niche internet forums like 4chan, has been implicated as a source of political activism and more recently also radicalisation, having been positioned as an incubator of the ‘alt-right’ turn in international politics. Studying the discourse on these forums is therefore valuable, but complicated by the fact that it is spread out over multiple sites and the fact that image boards are sometimes ephemeral; content is automatically deleted from them after a short while.
We propose that instead of studying the image board culture directly from the forums it comprises, we can use its encyclopedia, Encyclopedia Dramatica. This wiki is not so much a collection of knowledge in the usual sense, but rather a ‘performative archive’ in which the styles, preoccupations and attitudes of the subculture writing it are archived and catalogued. In this paper, we use diachronic word embeddings to study the semantic shift of key concepts and terms across 10 years of content from Encyclopedia Dramatica. We indeed observe this shift for some of the discourse, particularly that related to the masculinist ‘manosphere’, providing an empirical account of the ‘reactionary turn’ of image board culture. We also reflect on how the affordances of wikis and this wiki in particular complicate the application of state-of-the-art NLP techniques on this type of data.

Armin Pournaki
Max Planck Institute for Mathematics in the Sciences, Germany
Tutorial: Collect, transform and visualize data with the twitter explorer
In this tutorial, you will learn how to collect, transform and visualize Twitter data as interactive networks using the (twitterexplorer). Even though it will briefly cover the installation, we encourage users that want to participate in this hands-on tutorial to install the twitter explorer beforehand. Please visit our (newly launched blog), where you will find an introduction and all instructions for the tutorial.

For users who want to reproduce all the results of the tutorial:
- a Twitter developer account
- a machine that runs the twitter explorer
For users who don't want to collect their own data:
- a machine that runs the twitter explorer
For users who just want to try the interface with the provided dataset:
- a modern web browser

Please contact me [pournaki at mis dot mpg dot de] if you have questions.

Denise Pumain
Université Paris 1 Panthéon-Sorbonne, France
From discourse to facts through simulation of complex systems
The Covid pandemic19 has awakened recurrent debates about the future of cities. Some speculate on a gradual abandonment of large metropolises and a return of populations to the countryside, or to small or medium-sized towns. Similar predictions invoking a trend of "counter-urbanisation" had been made in the mid-1970s but did not materialise. Before risking predictive hypotheses, a thorough analysis of the dynamics of cities is necessary. We observe how cities, which are organised in systems of cities, maintain their size relationships over time. The main explanation for this persistence lies in the importance of the interactions between cities and urban stakeholders. Dynamic models of complex systems can simulate this development. They provide elements to explain the cognitive dissonance between individual and collective ways of living in the world. They also make it possible to envisage reasoned planning for a successful ecological transition and to reduce inequalities.

Monika Rakoczy
Sorbonne Université, France
(joint work with Robin Lamarche-Perrin and Armin Pournaki (MPI MiS))
Observing the evolution of the climate change debate across different public spheres


Climate change is an undeniable truth in the scientific community. However, even though predictions of its impacts have been persisting for years, the debate seems to have gained significant traction in the public sphere only in the last few years.One explanation for this could be the increase of immediate effects of climate change, such as rising temperatures, floods and bushfires.Meanwhile, the increasing interest in the climate change debate could as well have come along with a discursive shift, enhanced by emerging actors on the international scene such as Greta Thunberg. This shift can take place and manifest itself differently in various areas of debate, ranging from political discussions in parliaments to newspaper articles and social media outlets.Therefore, we present our contribution to a climate change observatory currently under development within the ODYCCEUS project. Our focus is to allow the users, in particular social science researchers and data journalists to explore data connected to climate change.
Our contribution consists of accessible methods and Web services to visualize and analyze the climate change debate on different media spheres.

We consider three large-scale datasets spanning from July 2016 to September 2019:
(1) For the perspective of social media, a collection of 80M tweets related to climate change (courtesy of DMI),
(2) For the perspective of mass media, all articles published by The Guardian (courtesy of UVB)
(3) For the political debate, a corpus of 92 287 UK parliamentary speeches (courtesy of MPI MIS)
All documents in these three corpora have been selected by the use of keyword "climate change".

For each of them, we identify and observe the dynamics of subtopics over time, employing methods ranging from outlier exploration and topic modeling to network representations.
This allows us to investigate the possible existence and dynamics of discursive shifts.
In particular, we focus on three aspects of debates: How much is climate change discussed in a given area? What vocabulary / terminology is used to do so? What kind of interaction patterns do we observe in such discussions?

Richard Rogers
University of Amsterdam, Netherlands
Mainstreaming the fringe, or how misinformation propagates on social media


The presentation takes up the question of the current state of the misinformation problem on social media, and its study. Employing digital methods and data journalism techniques, it examines how misinformation manifests itself on Facebook, Twitter, Instagram, Google Web Search, Reddit, 4chan and TikTok during the early COVID-19 pandemic period and the run-up to the US presidential elections. Judging from the most engaged-with content on the platforms, it appears the authority of mainstream media is waning. Broadly speaking, TikTok parodies it, 4chan and Reddit dismiss it and direct users to alternative influencer networks and extreme YouTube content. Twitter prefers the hyperpartisan over it. Facebook’s ‘fake news’ problem also concerns declining amounts of mainstream media referenced. Instagram has influencers (rather than, say, experts) dominating user engagement. By comparison, Google Web Search buoys the liberal mainstream (and sinks conservative sites), but generally gives special interest sources the privilege to provide information.

Livia van Vliet
University of Amsterdam, Netherlands
(joint work with Petter Tornberg and Justus Uitermark)
The Twitter parliamentarian database: Analyzing Twitter politics across 26 countries


We have built the Twitter Parliamentarian Database (TPD), a multi-source and manually validated database of parliamentarians on Twitter, designed to go beyond the one-off nature of most Twitter-based research. The TPD incorporates data from Twitter’s streaming API, government websites, data from the Manifesto Project Database; the Electoral System Design Database; the ParlGov database; and the Chapel Hill Expert Survey. It includes parliamentarians from all European Free Trade Association countries where over 45% of parliamentarians are on Twitter, as well as a selection of English-speaking countries. By compiling these different data sources it becomes possible to compare different countries, political parties, political party families, and different kinds of democracies. The database can be used in providing data for questions regarding differences between countries in parliamentarian Twitter interactions, difference in political parties use of hashtags, as well as the structure of parliamentarian interactions in transnational debates. Thus far, we have found striking cross-party and cross-national differences in how parliamentarians engage in politics on Twitter. For instance, different countries may create different retweet network structures, which may indicate more cohesion in certain countries over others. Moreover, there is an imbalance in the amount of retweeting between national parliamentarians (NPs) and MEPs, where MEPs retweet their national party members significantly more than other NPs. Overall, the TPD opens up new avenues of research in comparative politics on Twitter that were previously not possible.

Tom Willaert
Vrije Universiteit Brussel, Belgium
(joint work with Katrien Beuls, Paul Van Eecke, Jeroen Van Soest)
Tutorial: Penelope climate change debate observatory (see Penelope climate change opinion observatory)


The Penelope climate change opinion observatory offers a low-barrier collection of instruments for studying opinion landscapes on the climate crisis using data from a range of digital sources. These cover news media (e.g. the Guardian), social news media (e.g. Reddit), microblogging sites (e.g. Twitter), and political discussions (e.g. Transcripts of parliamentary debates from the UK and Germany). The platform thus aims to cater to the information needs of researchers, journalists, policy makers and engaged citizens with varying degrees of technical proficiency.
The observatory’s interface offers users the flexibility to chain together data analytics components developed in the ODYCCEUS project into custom pipelines. These components cover the research cycle from the phase of data gathering up to the actual reporting on analyses. Current tools within the climate change opinion observatory include:

  • Data from the aforementioned sources and an interface for selecting and filtering data by keywords, date range and number of posts or articles.
  • Preprocessing components for extracting textual information of interest, including named entities, word co-occurrences, and causal frames.
  • Components for in-depth analysis and visualization of the extracted information, including causal maps, plots of causal distributions, and statement graphs.
  • A reporting tool that fosters reproducibility of the research by keeping track of the activated components and their settings.

In this session, we provide a tutorial of the observatory and present some of the incorporated data analytics components that are part of the Penelope ecosystem of tools for computational social science.


Date and Location

October 26 - 28, 2020
Max Planck Institute for Mathematics in the Sciences
Virtual event - Videobroadcast

Scientific Organizer

Eckehard Olbrich
MPI for Mathematics in the Sciences
Leipzig (Germany)

Local Organization

Felix Gaisbauer
MPI for Mathematics in the Sciences
Leipzig (Germany)

Administrative Contact

Antje Vandenberg
MPI for Mathematics in the Sciences
Leipzig (Germany)
Contact by Email

03.11.2020, 11:16