Improving Decision Making with Argument Mining
October 11th 2017 13:30 - 13:50
Argument Mining is an emerging language technology that receives increasing attention from computer scientists and language technology industries. Current advances show promising results for extracting natural language arguments and their structures from specific types of discourse. These methods open up novel paradigms for exploiting the potential of big data and enable innovative search engines that deliver arguments relevant to a given user query.
In this presentation, we will introduce the current advances of ArgumenText, a project which targets the transfer of current argument mining technologies into industrial applications. In particular, we will present how mining arguments and evidence from various sources can facilitate decision making in complex situations such as buying decisions, fake news recognition and strategic decision making. Furthermore, we will introduce the current state-of-the art in Argument Mining and show how current advances in joint-modeling techniques and end-to-end neural networks improved the extraction of arguments. To enable an easy adaptation of our argument mining models to novel use cases and domains, we established a data generation process using crowdsourcing, which breaks the complex annotation task down into smaller subtasks and allows for generating training data tailored to specific use cases and heterogeneous text sources.
In the final part of the talk, we lay out exemplary use cases and explain how these can be expanded and turned into business workflows. All use cases are based on a common “business pipeline”, with components requiring minimal adaptation for new use case scenarios.