FicClaim: A Framework for Claim Verification in Fictional Domains Using Synthetic Data Generation
Abstract: The spread of misinformation and disinformation on social medial platforms has made automatic claim verification an important concern in various domains. We study the problem of claim verification in the context of claims about fictional stories for the purpose of uncovering logical inconsistencies also known as plot holes. To this end, we first introduce FicClaim, a synthetic dataset containing plot holes. FicClaim is generated in part by large language models (LLMs) for learning how to apply claim verification to fictional settings. We then develop the FicVer algorithm for finding inconsistencies in a story based on our dataset. We benchmark our algorithm against various claim verification methods and demonstrate that the proposed algorithm leads to state-of-the-art performance. Our code is available at https://anonymized.
Submission Length: Regular submission (no more than 12 pages of main content)
Changes Since Last Submission: Addressed reviewer comments and added requested changes to our work.
Assigned Action Editor: ~Novi_Quadrianto1
Submission Number: 1858
Loading