ExClaim: Explainable Neural Claim Verification Using Rationalization

Explainability matters for misinformation detection: The paper demonstrates that claim verification systems can achieve high accuracy (0.93 F1 score) while remaining interpretable to users, challenging the common trade-off between performance and transparency that has plagued AI-based fact-checking tools.

Black-box models undermine trust in automated fact-checking: Current claim verification systems use complex neural networks whose decision-making processes are hidden from users, making it difficult for people to understand why a claim was labeled true or false—a critical gap when these systems influence public understanding of information.

Natural language explanations can justify AI verdicts: ExClaim uses "rationalization" (inspired by legal reasoning) to provide human-readable explanations alongside fact-check verdicts, helping users understand not just what the system decided, but why it reached that conclusion.

Evidence source bias is a systemic problem: The paper identifies that existing fact-checking systems often rely on mainstream news as evidence, which can carry specific agendas and biases, limiting the objectivity of automated claim verification.

Intermediate steps need explanation too: Beyond explaining final verdicts, the system provides explanations for intermediate reasoning steps, allowing users to scrutinize the full decision-making process rather than just the final output.