5. Fact Verification

[This project is a part of my tenure at University of South Carolina as an affiliate Researcher]

Problem

  • Automatic fact verification has received significant attention recently. Traditional automatic fact-checking systems focus on estimating truthfulness using numerical scores which are not human-interpretable. Manual fact-checking is a time-consuming task. Professional fact-checkers take a considerable amount of time sometimes several days to verify a fact.

Proposed Solution

  • In this work, an automatic fact verification system is proposed using question answering based on 5W (Who, When, What, Where, Why) approach


Implementation

  • Data is collected from random a mixture of publicly available fake news dataset

  • Experimentation for finding 5Ws from a given claim is conducted using a previously implemented work.

  • Experimented with question and answering system to take the context from the claim and generate automatic 5W-based questions

  • This is to be followed by an information retrieval system and a textual entailment model to generate evidence so that a final verdict is predicted

My contribution

  • Ideation on the framework, managing the team

  • Experimentation on 5W SRL and question answering model and evaluating the results

  • Working on the manuscript

Uniqueness

  • It would be the first system proposed for 5W Aspect based fact verification and would have an added explainability through question answer generation framework

Results

  • We plan to publish this work at ACL'23 and due to confidentiality a lot of information couldn't be added into this project here at the moment.