COOLANT++: concept-bank alignment and confidence-gated reasoning for multimodal fake news detection

Abstract

Model and training pipeline accompanying the Scientific Reports article “COOLANT++: Concept-Bank Alignment and Confidence-Gated Reasoning for Multimodal Fake News Detection.” The release implements the shared cross-modal concept bank, confidence-gated reasoning unit, training losses, benchmark feature loader, input-validation utilities, and training/evaluation entry point. Twitter and Weibo benchmark data are not redistributed; dataset access and preprocessing are provided through the public original-COOLANT repository.