H. Veeramani, S. Thapa, R. Kanagasabai, U. Naseem,SRLFactQA at Factify5WQA: Composite Claim-Evidence Consistency Aware Semantic Role Labelling based Question-Answering Entailment,in: Proceedings of De-Factify 3.0: Third Workshop on Multimodal Fact Checking and Hate Speech Detection, CEUR, 2024.
Abstract:
In today’s digital age, the rampant spread of fake news and misinformation poses significant challenges.
Fact-checking, a crucial tool in combatting misinformation, has benefited from advances in machine
learning and NLP techniques. There is also a growing emphasis on making fact-checking models
transparent and interpretable. FACTIFY-5WQA 2024, a shared task within DeFactify, is built upon a 5W
framework (who, what, when, where, and why) for question-answer-based fact explainability. In the task,
participants are tasked with developing fact-checking systems that can automatically process claims, use
the associated evidence and questions to determine the veracity of the claims, and assign the correct
labels. The task uses the FACTIFY-5WQA dataset, a semi-automatically generated fact verification
dataset designed to determine the veracity of a claim based on the evidence / documents provided and
the questions and answers generated from both the claim and the evidence. In this paper, we report
our approach using a twin semantic frame-based fine-grained inconsistency detection model, achieving
an accuracy of 45.51%, a significant improvement over the baseline securing the second position in the
shared task.
License type:
Attribution 4.0 International (CC BY 4.0)
Funding Info:
There was no specific funding for the research done