<p>Fact verification refers to the process of detecting false information based on evidence texts, a task that presents significant challenges. Current research faces issues such as the neglect of long-distance semantic dependencies and interference caused by redundant information. To address these challenges, this paper proposes a knowledge-enhanced method that incorporates question answering for fact verification. The method first introduces a question answering protocol pipeline to capture and classify fine-grained information while filtering out redundant data. Then it improves knowledge using this fine-grained information and constructs a graph with various types of entity nodes to uncover the potential semantic relationships between evidence and claims. The experimental results on the FEVER dataset demonstrate that the proposed method outperforms existing comparative models in both Fever Score (FS) and Label Accuracy (LA), thus validating the effectiveness of the model. On the test set, LA reached 0.748 and FS reached 0.707.</p>