| Abstract | With the proliferation of fake news in the last few years, especially during COVID-
19, combating the spread of misinformation has become a social and political urgent
need. Fact-checkers and journalists need to identify claims that were previously verified
by a reputable fact-checking organization before inspecting the claim veracity. Many
claims showed up repeatedly but at different time periods and different forms. In this
thesis, we propose an automated approach to retrieve claims that have been already
manually-verified by professional fact-checkers. Our proposed approach uses recent
powerful BERT (BERT is a Transformer-based machine learning technique that can be
used to address several Natural Language Processing problems effectively) variants as
rerankers in monoBERT fashion. MonoBERT is a point-wise ranking approach that uses
a BERT-based model to assign a relevance score for query-document pair. Additionally,
we study the impact of using different fields of the verified claim during training and
inference phases. Experimental results show that our proposed pipeline outperforms
the state-of-the-art approaches on two public English datasets and one Arabic dataset
by a remarkable margin. Moreover, we are the first to develop a system for the Arabic
language. |
