Photo courtesy of https://unsplash.com/photos/5QgIuuBxKwM
Arguments as the Social Good: Good arguments even in times of crisis

The COVID 19 crisis confronts decision-makers in politics, society and the economy with the challenge of having to make very quick decisions in a completely new situation under conditions that can change daily. In this situation, any support that provides a well-founded basis for decision-making is of great benefit. The aim of the proposed project is to contribute to this by extracting arguments from the widest possible range of unstructured but daily updated web sources (social media, news and scientific publications).

As a user group, we primarily address decision-makers in politics and business, but also the general public. The project result is made freely available through a publicly accessible web demonstrator. The demonstrator will be an argumentative search engine, which will clearly present current pro- and contra-arguments (i.e. justified options for action) on topics from the COVID 19 crisis (e.g. "curfews", "face masks") and make trends recognizable by means of a suitable visualization (temporal course of the pro- and contra-arguments over the last days/weeks). Sources from all parts of the world are taken into account, so that users can obtain a balanced but nevertheless meaningful mood picture for the search term.

Project members
  • Dr. Johannes Daxenberger, Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/jSKjkV4Oc5Q
Public Opinion Formation about Lockdowns over Time

There is currently a consensus that a lockdown is necessary for humanitarian reasons. This will probably change when people realize that this will be disadvantageous for them in the long run (similar to e.g. the refugee crisis in 2015). Both the rapid emergence of the consensus and its (probable) erosion can be traced with Twitter data (Twitter as an indicator of public opinion). It is also a question of the social groups from which such processes emanate.

Method: Analysis of Twitter data according to social groups (politics, news media, science, citizens; if possible even more detailed: individual media and parties). Record the position and arguments for (protection of the elderly and sick; protection of the health care system) and against (economic consequences; damage to health caused by staying at home) the lockdown (argument mining). Later time series analysis on the mutual influence of social groups (who influences whom?).

Further analysis options: How is this related to the actual infection rates? How do the Twitter data fit with the survey data (is it possible to capture valid public opinion formation with Twitter at all?)

Project members
  • Dr. Christina Viehmann, Communication Sciences, Johannes Gutenberg Universität Mainz
  • Prof. Marcus Maurer, Communication Sciences, Johannes Gutenberg Universität Mainz
  • Prof. Oliver Quiring, Communication Sciences, Johannes Gutenberg Universität Mainz
  • Tilman Beck, M.Sc., Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/qZenO_gQ7QA
Fighting Disinformation: Mining and Evaluating Evidence for COVID-19 from Heterogeneous Sources

In the ongoing COVID-19 crisis, disinformation endangers lives. Social media posts and media reports included wrong information about policy measures, the pandemics threat and even wrong health advice, which was also shared by political actors. This spread of disinformation about infectious diseases not only causes panic and uncertainty in societies, but also causes an increase in the spread of the disease itself.

Our project’s main goal is to fight disinformation by developing the first bilingual system for mining and evaluating evidence for complex COVID-19 claims in English and German. Current approaches are limited due to methodological, theoretical, and technological reasons. We address these challenges by drawing on novel NLP approaches and computational analysis techniques to develop a fair, transparent, fast and reasoned assessment of identified evidence related to COVID-19 in heterogeneous sources.

Project members
  • Simon Kruschinski, M.A., Communication Sciences, Johannes Gutenberg Universität Mainz
  • Prof. Marcus Maurer, Communication Sciences, Johannes Gutenberg Universität Mainz
  • Prof. Oliver Quiring, Communication Sciences, Johannes Gutenberg Universität Mainz
  • Nils Reimers, Dr. Ing., Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/FMJAYeRwtDY
Twitter and Google Search Data for Predicting the Daily Number of New Infections of COVID-19 in Germany

Knowing about the daily number of new infections of COVID-19 is crucial to decide about the next best actions to fight against the virus. Unfortunately, official numbers provided by the Robert Koch Institute (RKI) in Germany are disclosed correctly with several days of delay and are significantly underreported on weekends.

The widely cited other source is provided by the Center for Systems Science and Engineering at Johns Hopkins University (JHU) for Germany. However, it also deviates strongly from the official numbers. In this project, we investigate the use of Twitter and Google Search Data for more accurate and timely predictions of COVID-19 infections spread.

Project members
  • Lukas Jürgensmeier, M.Sc., Goethe University Frankfurt
  • Prof. Bernd Skiera, Goethe University Frankfurt
  • Kevin Stowe, PhD, Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/mcrQoYT8VFE
Answering Critical Questions in the Face of a Global Crisis

Uncertain situations lead to a diverse set of questions, especially if the crisis affects every economic and social aspect of our life. Getting quick answers can be of high importance, however, finding the right information is really hard. Information is often scarce, scattered over many different sources, the situation is rapidly evolving and, as there is a lot of uncertainty, many sources cannot be trusted.

The aim of this project is to develop a question answering platform that helps finding answers to the most pressing questions. In the project, we address two user groups.

The first group are researchers who want to source scientific literature to find new ways to tackle the critical situation. For this, we index over 45,000 peer-reviewed scientific articles and provide useful meta-data to help analyzing the situation more rapidly.

At the same time, understanding scientific literature is challenging and only possible with domain expertise. To provide a benefit to a wider public audience, the system also has a "user mode" that provides answers to everyday questions about the crisis. For this, we source various trustworthy sources like the WHO, CDC, or FDA.

Project members
  • Nils Reimers, Dr.-Ing., Computer Science, TU Darmstadt
  • Gregor Geigle, B.Sc., Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/gHrEvF7Ugb4
Real-Time Multilingual Multi-Modal Twitter Content Analysis for COVID-19 Crisis Response

The impact of the COVID-19 is global, affecting all people from all language backgrounds. Similarly, data revolving around the event is inherently diverse in many, representing hundreds of languages, and consists of texts, images, and videos. Social media in particular contains data from a wide variety of languages, coupled with user-generated images and videos. The speed with which the pandemic is evolving necessitates a platform that is capable of expanding to new languages and new types of data quickly.

The aim of this project is twofold. First, we aim to employ state-of-the-art language technology to adapt big data to small: can we adapt these models designed for high-resource languages such as English and German to provide accurate classification and support for low-resource languages? Second, we incorporate modern computer vision to merge analysis of text, images and videos. There exists a vast wealth of information in images: empty shelves, prevalence of mask use, infographics of pandemic data; and we use multi-modal methods to build prediction based on images along with their related texts, giving us a more complete understanding of the available data. Through both of these methods, we aim to build a system that can quickly and accurately analyze social media data, leveraging relevant information from texts and images, allowing researchers as well as the general public to better understand the crisis, it’s impacts, and possible solutions.

Project members
  • Jonas Pfeiffer, M. Sc., Computer Science, TU Darmstadt
  • Jan-Martin Steitz, M. Sc., Computer Science, TU Darmstadt
  • Kevin Stowe, PhD., Computer Science, TU Darmstadt
  • Prof. Stefan Roth, Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/2FPjlAyMQTA
Interactive Annotation Platform for Rapidly Creating COVID-19 Training and Evaluation Data

Rapid creation of training and evaluation data is the key requirement for Content Analytics in real-time on the Web content. For example, Twitter streams, news articles or scientific literature need to be analysed for COVID-19-related time-critical information. This requires a large-scale effort of the community rather than closed groups of expert annotators or effort-consuming crowdsourcing.

For this purpose, we present a thematically focused interactive annotation platform to be jointly used by the researchers in this project. The platform will monitor and continuously index a range of data sources. The researchers can then search these data sources for relevant content, e.g. by keywords, and then import this content into the platform for manual annotation. Access to annotations contributed on the platform will be granted to the academic community by means of an academic authentication service (DFN-AAI, eduGAIN, CLARIN-AAI, or similar), thereby adhering to the conditions for the copyright exception for academic research. We shall base this platform on the open-source INCEpTION annotation software.

Project members
  • Richard Eckart de Castilho, Dr.-Ing., Computer Science, TU Darmstadt
  • Jan-Christoph Klie, M.Sc., Computer Science, TU Darmstadt
  • Ute Winchenbach, M.Sc., Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/w59nZxIkw8M
On the Pulse of Time: Public Opinion Monitoring of COVID-19 Government Measures in the German News

In the face of COVID-19, governments decide to implement measures of varying degrees of stringency, which in some cases severely restrict public life. Although these measures may seem (and are) necessary at certain points in time to relieve the current backbone of our society - the medical sector - we still have to weigh up the long-term consequences. It is specifically the societal acceptance of the decisions taken that is crucial for the success of the strategies with regard to the crisis. In this context, we aim to measure the public opinion on governmental actions taken in Germany.

The public opinion shall be represented by a sentiment analysis of comments to COVID-19 related articles of leading German news portals. The results of this analysis will be put into context with a dataset of government measures over time. With our analyses we hope to be able to provide insights on an effectiveness-stringency trade-off of policy actions.

Project members
  • Hendrik Jöntgen, Goethe University Frankfurt
  • Prof. Oliver Hinz, Goethe University Frankfurt
  • Nicolas Pfeuffer, Goethe University Frankfurt
  • Johannes Schaffrath, Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/M9EctVUPrp4
Twitter data for predicting social solidarity during the Corona crisis

The COVID 19 crisis poses not only a severe threat to peoples' health, it profoundly changes our professional and private lives. The lockdowns in countries are putting their citizens on for humanitarian reasons have repercussions for families, work, businesses, and civic participation.

Using current Twitter data (Twitter as an indicator of public opinion), we predict how social solidarity develops under these new conditions. At a time at which many fear that social cohesion and solidarity are at risk, how do societies cope with this unforeseen challenge?

Project members
  • Prof. Daniela Grunow, Goethe University Frankfurt
  • Alexandra Ils, M.A., Goethe University Frankfurt
  • Dr. Steffen Eger, Computer Science, TU Darmstadt
  • Kevin Stowe, PhD., Computer Science, TU Darmstadt
  • Prof. Iryna Gurevych, Computer Science, TU Darmstadt
Read More
Photo source: https://unsplash.com/photos/TrhLCn1abMU
Across Countries and Regions: Public Reactions to Policy Decisions in Times of Crisis

COVID-19 has caused enormous pressure upon policy-makers to make critical strategy decisions under hard time constraints. In times of crisis, policy-makers may use different strategies to attempt to constrain the crisis. The effect of these strategies can hardly be anticipated. Beforehand it is unclear which strategies are more effective to convince the citizens of implementing protective measures for reducing the infection rate.

This project studies the evolution of public opinions in light of the events (e.g., new government measures, a high lethality rate etc.) related to COVID-19 and the respective strategies adopted by different countries and regions. We aim to answer the following question: How do citizens respond to the strategies adopted by policy-makers in different geographical areas? We rely on a large volume of unstructured data from COVID-19 Twitter posts in multiple languages, as well as on news articles, official press releases and public addresses from policy-makers.

Project members
  • Gabriela Alves Werb, Goethe University Frankfurt
  • Marcus Dombois, Faculty of Civil and Environmental Engineering Sciences, TU Darmstadt
  • Christopher Klamm, Computer Science, TU Darmstadt
  • Lennart Kraft, Goethe University Frankfurt
  • Jens Stappenbeck, Hessische Stiftung Friedens- und Konfliktforschung
Read More
Human and Machine in the Loop: Understanding Covid-19 through Interactive Machine Learning

We use deep learning with interactive machine learning and explainable AI components to analyze lungs x-rays. In particular, we use Grad-CAMs as explanation method for the AlexNet model trained on a rich COVID-19 x-ray data set that contains three different diagnosis classes: COVID-19, NORMAL and Viral Pneumonia.

We propose an innovative process: First, we use deep learning and create a standard classifier. We then apply interactive machine learning to annotate relevant areas in the x-rays and re-train then the classifier. Based on the differences, we can identify the influence of confounding factors on classification results. We subsequently use explainable AI to create heat maps that highlight the relevant areas on the x-rays that are responsible for the classification of the particular case. We present these data to pneumologists who can then try to identify patterns and 1.) compare these patterns with current diagnosis recommendations for radiologists and 2.) potentially identify new patterns that have not been identified by human experts so far.

This sub-project is partly financed by the Volkswagen Stiftung as part of the seeding project “From Machine Learning to Machine Teaching (ML2MT) – Making Machines AND Humans Smarter” in the call “Artificial Intelligence and the Society of the Future”.

Project members
  • Prof. Oliver Hinz, Goethe University Frankfurt
  • Nicolas Pfeuffer, Goethe University Frankfurt
  • Benjamin Abdel-Karim, Goethe University Frankfurt
  • Prof. Kristian Kersting, TU Darmstadt
  • Wolfgang Stammer, TU Darmstadt
  • Patrick Schramowski, TU Darmstadt
  • Prof. Gernot Rohde, University Hospital and Goethe University Frankfurt
Read More
Public and legal reasoning strategies for Covid-19 policies

Legal practitioners have been discussing whether the Covid-19 policies are lawful since their enforcement in the beginning of March 2020. While at the beginning the public accepted these policies as a safety measure, complaints have been issued against certain policies ever since. Furthermore, several public discussions have broken out about the necessity and lawfulness of the measures. The proposed project aims to analyze if reasoning strategies and arguments form the legal side are used in public and vice versa.

We will analyse jurisdiction and Twitter data with respect to critical topics, such as the right of assembly or freedom of religion. We plan to compare arguments and reasoning strategies on both sides to find mutually used arguments and reasoning strategies.

Project members
  • Prof. Indra Spiecker genannt Döhmann, Goethe University Frankfurt
  • Prof. Christoph Burchard, Goethe University Frankfurt
  • Dr. Sebastian Bretthauer, Goethe University Frankfurt
  • Dr. Nicola Recchia, Goethe University Frankfurt
  • Dr. Ivan Habernal, TU Darmstadt
  • Fabian Kaiser, MSc., TU Darmstadt
  • Prof. Iryna Gurevych, TU Darmstadt
Read More