The paper introduces a novel retrieval method called GNN-Ret, which enhances the retrieval process for question answering (QA) by leveraging graph neural networks (GNNs) to consider the relatedness between passages. The authors construct a graph of passages, connecting those that are structurally and keyword-related, and use a GNN to exploit the relationships between passages, improving the retrieval of supporting passages. For multi-hop reasoning questions, they extend the method to RGNN-Ret, which uses a recurrent GNN to integrate the graphs of passages from previous steps, enhancing the retrieval coverage over multiple steps. Extensive experiments on benchmark datasets demonstrate that GNN-Ret and RGNN-Ret achieve higher accuracy for QA tasks compared to strong baselines, with RGNN-Ret achieving state-of-the-art performance on the 2WikiMQA dataset. The contributions of the paper include the introduction of GNN-Ret and RGNN-Ret, which effectively enhance retrieval coverage and improve the accuracy of QA tasks.The paper introduces a novel retrieval method called GNN-Ret, which enhances the retrieval process for question answering (QA) by leveraging graph neural networks (GNNs) to consider the relatedness between passages. The authors construct a graph of passages, connecting those that are structurally and keyword-related, and use a GNN to exploit the relationships between passages, improving the retrieval of supporting passages. For multi-hop reasoning questions, they extend the method to RGNN-Ret, which uses a recurrent GNN to integrate the graphs of passages from previous steps, enhancing the retrieval coverage over multiple steps. Extensive experiments on benchmark datasets demonstrate that GNN-Ret and RGNN-Ret achieve higher accuracy for QA tasks compared to strong baselines, with RGNN-Ret achieving state-of-the-art performance on the 2WikiMQA dataset. The contributions of the paper include the introduction of GNN-Ret and RGNN-Ret, which effectively enhance retrieval coverage and improve the accuracy of QA tasks.