Claim polarity analysis from conflicting sources

Automating fact-checking is a challenging task that often requires the retrieval of multiple pieces of evidence from a reliable corpus for verifying the truthfulness of a claim. Most of the current automated fact-checking systems produce true-or-false verdicts for certain types of factual claims in order to provide accuracy and transparency. However, some of the claim fragments can be true and others false. We propose ClaimAnalyse, the first fully automated claim polarity analysis system, an easy-to-use and an efficient tool for analyzing a claim according to the conformity and non-conformity of query sub-claims with the sources. Given a query claim and its related sources, ClaimAnalyse (I) selects the common sub-claims between the query claim and the sources, (II) composes the sources to cover the query claim and detects conform or non-conform source sub-claims with it, and (III) provides the polarity analysis of the query sub-claims over sources without any judgment or verdict on the entire claim. Our system provides the user with a complete and accurate analysis of the claim, which allows it to provide its own decision according to the polarity assigned to each query sub-claim. Using FEVER dataset, we experimentally verify the efficiency and the effectiveness of the proposed model. The experimental results have shown that among the state of-the-art methods, our proposed ClaimAnalyse model has achieved the best performance.