RATIO Hackathon 2023
From 24.04 to 28.04. a group of ~10 persons gathered in the University of Bielefeld (CITEC-Building) in order to tackle two tasks in a Hackathon. Looking back to this very productive week with lots of excellent ideas, brainstorming, programming together (and framing already a paper), it was a pleasure to have this event in Bielefeld. Together with funny and tasty socials in two evenings (game evening and exploring the Sparrenburg), we initialized cooperations in the field of argument mining. In particular, we investigated two tasks:
Argument Ranking via Argument Quality Assessment
One group looked into the shared task of Argument Retrieval for Controversial Questions 2023. Hence, given a controversial question, the is to return an ordered list of (argumentative) documents (PRO and CON) in order to discuss that question. While the initial ranking can be done by argument-agnostic state-of-the-art-retrieval methods as BM25, the task requires to retrieve well-argumentative documents, too. The group implemented traditional as well as deep-neural methods to assess the argument quality of comprehensive web documents.
Aspect Identification in short argumentative texts
The other group had a closer look at the paper by Mattes Ruckdeschel “Boundary Detection and Categorization of Argument Aspects via Supervised Learning”. Clustering arguments of a certain topic into emphasized main points/ subtopics is an important step in the argument mining pipeline to overview the important aspects in a debate. Mattes Ruckdeschel created a dataset containing more than 1000 multi-labeled arguments each for 4 different topics. However, until yet, supervised approaches detecting emphasized aspects in arguments can only handle generic aspect groups or topic-specific aspects lacking the ability to generalize across topics. Hence, this group closes this gap by concepting a supervised approach of cross-topic topic-specific aspects-identification by the final idea of applying contrastive learning.