Although substantial progress has been made in rumor detection and misinformation identification, existing methods still exhibit notable limitations in both information processing and feature representation. Specifically, current methods often struggle to comprehend complex semantics or implicit expressions, and they typically lack deep modeling of the underlying rumor propagation mechanisms, thereby limiting their capacity to handle dynamic information evolution. To address these challenges, we propose a novel rumor detection framework that integrates commonsense reasoning with feedback-enhanced propagation modeling. First, we simulate the human information processing process by introducing ConceptNet to enhance the model’s commonsense reasoning capabilities. Then, we construct a feedback-enhanced propagation graph that captures both local and global structural information within the dissemination process, enabling comprehensive extraction of propagation-related features. Finally, these features are integrated for rumor classification. Experimental results on three benchmark datasets show that the proposed method consistently outperforms state-of-the-art baselines, highlighting its robustness and effectiveness across diverse data scenarios.