ArgumenText is a validation project of the Ubiquitous Knowledge Processing (UKP) Lab at the Technische Universität Darmstadt. It is our goal to validate the latest research breakthroughs in Argument Mining and Text Analytics in industrial applications and to develop innovative products to unleash the potential of unstructured data. Starting with a previously developed joint-modeling method for identifying argument structures in student essays (Stab and Gurevych 2017), we develop robust end-to-end approaches for mining arguments from web-scale corpora.

We extent our current methods to other languages like German using Language Adaptation and evaluate their robustness across various topics and text types. The resulting methods and software are intended to extract arguments from dynamic text sources such as news streams or social media in real-time and to prepare them for the user in a comprehensive summary.

ArgumenText is funded by the Federal Ministry of Education and Research (BMBF) as part of the VIP+ programme.

Project goals

Infographic Argument Mining

Argument Mining

We develop novel deep learning methods for robustly mining arguments from heterogeneous Web sources and text streams like social media and news. Our approach allows for fast adaptation to new use cases and mining arguments robustly across different domains.

Language Adaptation

Our argument mining methods will be optimized for finding arguments in different languages. We build on the latest multilingual representations allowing us to make optimal use of our training resources and to apply a model trained in one language to many other language.

Infographic Language Adaption
Infographic Argument Aggregation

Argument Aggregation

For allowing a fast and comprehensive overview of topic-relevant arguments, we develop summarization methods that group similar pro and con arguments from different sources. This allows easy access to relevant arguments without reading through long result lists.

Realtime Analytics

To find trending arguments, we apply our methods to text streams like news or social media. We seek to develop a scalable infrastructure that continuously searches millions of different sources for the most recent arguments and delivers the latest trends in real-time.

Infographic Realtime Analytics
argumentext background svg

Use Cases

Our technology allows to unleash the potential of unstructured data and to integrate unused assets into decision-making processes in various different application scenarios:
1

Advanced Online Research

The search for well-grounded information on (potentially) controversial topics in the web is often limited by the huge number of expressions of plain opinion or sentiment. However, the web also contains sufficient reasonable arguments – but they are much harder to detect in the sheer mass of information available. Argument extraction is able to mine these valuable information nuggets for any topic you can think of.

2

Trending Topics

The potential of emerging technologies or products is a frequently discussed topic on the web. Such discussion can be very valuable for decision makers and innovation management, as they find diverse and well-reasoned opinions on new products and technologies. While simple research about innovation might lead to having to scan through mostly irrelevant advertisement or opinion spam, argument mining uncovers the essential information necessary to discover promising innovation. By scanning arguments for given products or technologies over time, ArgumenText is able to instantly reveal changes as well as trends in the public perception of the latter.

3

Customer Feedback Analysis

Opinion mining and sentiment analysis are redefining customer feedback analysis in companies around the world. However, current approaches to (semi-)automatically mining feedback from social media or other public communication channels can only reveal what people like or dislike about a certain product, brand or company. Argument extraction goes an important step further – it detects the reason behind why people like or dislike the targeted product or brand. This information can be of massive value for decision makers in marketing or management.

Partners

Want to request a demo?

We'd love to hear from you.