TU Berlin

Quality and Usability Lab01_02_2021_Prabhakar

Inhalt des Dokuments

zur Navigation

Claim Extraction from Text using Transfer Learning

Location:  Zoom link (Please ask Saman Zadtootaghaj for access)  

Date/Time: 01.02.2021, 15:00-15:30 

SPEAKER: Acharya Ashish Prabhakar (TU Berlin)

Abstract: Fake news detection and argument mining are extensively using claim detection as a pre-processing step. Very high accuracy is necessary to ensure final performance after downstream tasks are high and critical claims are not falsely omitted in the proceeding steps. Recent advancements in transfer learning have shown promising results achieving the state of the art results with a smaller dataset. We apply these methods to perform claim extraction in English and German and quantify its results against a simple LSTM based baseline. We also introduce a large scale dataset of non-claims and make it publicly available for future research. The final result indicates that highly accurate models can be trained with smaller datasets. The introduced dataset of non-claims eradicates any scarcity of training data for the given scenario. We hope this work will benefit any future research that would want to include claim extraction in its pipeline.






Schnellnavigation zur Seite über Nummerneingabe