Identification and Aggregation of Arguments in Scientific Literature on Basis of Natural Language Processing

Identification and Aggregation of Arguments in Scientific Literature on Basis of Natural Language Processing (Sci-Arg)

Overview

With the rapid development in science in the last decades, the number of scientific publications in conferences, journals and other publication platforms rises, too. The amount of information poses challenges for young researchers as it is a very time-consuming process to identify relevant information. This is especially true for interdisciplinary research, as it combines the research of several fields. In this research project, “Identification and Aggregation of Arguments in Scientific Literature on Basis of Natural Language Processing” (Sci-Arg), we will focus on developing Argument Mining (AM) techniques from an information-seeking perspective to detect relevant arguments in large-scale scientific literature.

Goal

The greater goal of this project is to facilitate the search for relevant information for researchers. Similar to a search engine, users should be able to crawl relevant arguments in vast amounts of scientific literature and thus acquire an overview of the discourse in related work. The focus here is on a interdisciplinary search for arguments.

Using Natural Language Processing methods, we aim to:

  • identify topic-relevant arguments in vast amount of scientific literature
  • aggregate the arguments to eliminate redundant content and group them according their topical aspects
  • integrate the functionalities in a prototype

In related work it has been shown that generating and annotating a dataset in the scientific domain is difficult, even for domain experts. Thus, we are interested in leveraging approaches like Multi-Task Learning and Transfer Learning to re-use existing datasets and models.

Further, in the course of this project we are continuously evaluating our approach in a real-world setup. For that purpose, we are developing a prototype which enables researchers to search for relevant arguments in vast amount of scientific literature. Using this prototype, we will collect feedback about the quality of the search results and adapt our models accordingly.

Funding

Software Campus program (BMBF)

Partners

People

  • Prof. Dr. Iryna Gurevych, Mentor
  • Tilman Beck, Doctoral Researcher