Abstract

Automatic fact verification has attracted recent research attention as the increasing dissemination of disinformation on social media platforms. The FEVEROUS shared task introduces a benchmark for fact verification, in which a system is challenged to verify the given claim using the extracted evidential elements from Wikipedia documents. In this paper, we propose our 3rd place three-stage system consisting of document retrieval, element retrieval, and verdict inference for the FEVEROUS shared task. By considering the context relevance in the fact extraction and verification task, our system achieves 0.29 FEVEROUS score on the development set and 0.25 FEVEROUS score on the blind test set, both outperforming the FEVEROUS baseline.

Highlights

  • The large-scale dissemination of disinformation on social media platforms intended to mislead or deceive the general population has become a major societal problem (Tan et al, 2020)

  • We propose a three-stage system as Figure 1 shows to improve the FEVEROUS baseline in two aspects

  • There- k = 5 for the downstream element retrieval using fore, we adopt the RoBERTa (Liu et al, 2019) NLI the BERT model and at the same time experiment model pre-trained on well-known NLI datasets, in- with different settings for the downstream element cluding SNLI, MNLI, FEVER-NLI, ANLI

Read more

Summary

Introduction

The large-scale dissemination of disinformation on social media platforms intended to mislead or deceive the general population has become a major societal problem (Tan et al, 2020). To answer the increasing demand for such systems, the FEVER (Fact Extraction and VERification) dataset (Thorne et al, 2018) was introduced and used for the shared task of the FEVER Workshop 2018 It consists of 185,445 annotated claims with a label of "SUPPORTED", "REFUTED", or "NOT ENOUGH INFO" as well as sets of evidential sentences from the given pre-processed Wikipedia pages. Among the participated teams a system to be able to retrieve structured information from Wikipedia as evidence for each claim, which differs from the shared task in 2018 These two shared tasks still share the similar setting as a fact extraction and verification problem, which makes the pipelines and methods of the early proposed systems worth referring to. Systems are evaluated by jointly considering how complete the relevant Wikipedia elements are retrieved and how correct the final verification verdicts are

Methods
Results
Conclusion
Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.