Fact‑checking on social media is constantly hampered by evidence‑level and aggregation‑level conflicts, where misleading evidence masquerades as authoritative news. To capture this problem and enable conflict verification, we built the ContraNote dataset from X’s Community Notes system. It comprises 33,686 posts for evidence‑level conflict resolution and 54,474 instances for aggregation‑level prioritization.
We introduce CoVer, a factual adjudication framework organized as a three‑stage pipeline: (1) evidence schema normalization, mapping heterogeneous evidence into a unified representation; (2) factual consensus construction, aggregating votes or confidence scores to produce an initial verdict; (3) support verification, ensuring the consensus is backed by high‑quality evidence, thereby preventing noise from contaminating the final decision.
Technical evaluation shows CoVer achieves 86.0% accuracy, 68.0% macro‑F1, and 64.5% balanced accuracy on the conflict task, and 88.5% accuracy, 88.5% macro‑F1, and 89.2% balanced accuracy on the prioritization task. It also outperforms CONFACT‑HumC (88.4% accuracy) and CONFACT‑ModC (89.4% accuracy).
Review: By normalizing evidence and applying a multi‑level consensus mechanism, CoVer effectively mitigates the spread of conflicting misinformation on social platforms, offering a scalable approach for automated fact‑checking.