Graph Neural Networks (GNNs) have become a leading technique for semi-supervised node classification. However, the distribution of labeled nodes in the graph is often uneven, resulting in only a small portion of unlabeled nodes being able to directly reach labeled nodes. This misalignment between labeled and unlabeled nodes significantly weakens the performance of GNNs. To address this issue, this paper introduces a plug-and-play GNN architecture called RLGNNs. The core idea of RLGNNs consists of two parts. First, by utilizing a graph transformer, we establish direct global connections between unlabeled and labeled nodes, enhancing the reachability of labeled nodes and improving the representations of unlabeled nodes. Second, we incorporate GNNs to capture the original structural information of the graph that was overlooked in the first part, allowing for a more comprehensive understanding of the graph’s contextual information. Through a series of experiments, we demonstrate that RLGNNs achieve superior performance in semi-supervised node classification tasks compared to other state-of-the-art GNNs.