🤖 AI Summary
To address low accuracy and poor cross-domain generalization in automated fact-checking of complex real-world claims, this paper proposes the Progressive Adaptive Search Framework (PASS). PASS introduces three core innovations: (1) temporal- and entity-aware contextual enhancement for claim rewriting; (2) dynamic question generation triggered by reflective labeling; and (3) multi-granularity evidence aggregation with iterative reflective verification. Integrated with retrieval-augmented generation (RAG), language adaptation, and multilingual support modules, PASS enables end-to-end adaptive verification. Evaluated across six heterogeneous datasets—including general knowledge, scientific claims, real-world scenarios, and multilingual tasks—PASS consistently outperforms state-of-the-art baselines, achieving significant gains in accuracy. All code and experimental artifacts are publicly released.
📝 Abstract
Automated fact-checking faces challenges in handling complex real-world claims. We present PASS-FC, a novel framework that addresses these issues through claim augmentation, adaptive question generation, and iterative verification. PASS-FC enhances atomic claims with temporal and entity context, employs advanced search techniques, and utilizes a reflection mechanism. We evaluate PASS-FC on six diverse datasets, demonstrating superior performance across general knowledge, scientific, real-world, and multilingual fact-checking tasks. Our framework often surpasses stronger baseline models. Hyperparameter analysis reveals optimal settings for evidence quantity and reflection label triggers, while ablation studies highlight the importance of claim augmentation and language-specific adaptations. PASS-FC's performance underscores its effectiveness in improving fact-checking accuracy and adaptability across various domains. We will open-source our code and experimental results to facilitate further research in this area.