In the domain of graph neural networks, the issue of heterophily poses a considerable challenge. Combining intermediate node representations serves as an effective approach to mitigate graph heterophily. However, existing methods of this kind typically focus on intra-node level combinations, while overlooking the fact that inter-node intermediate representations (i.e., those from different nodes) are topologically connected and can also provide rich semantic information for the learning of target node’s representation. To leverage such inter-node information and further explore deeper topological structural information, we propose MRW-GNN, a novel GNN framework that extends the input graph into a multi-range graph comprising multiple subgraphs with identical topology but diverse node representations. MRW-GNN employs random walk algorithms on the multi-range graph to sample inter-node intermediate representations, which are then processed by RNN-based models to achieve effective fusion of semantic and structural information. Experiments on various datasets demonstrate that MRW-GNN achieves state-of-the-art performance on both homophilic and heterophilic graphs.