GraphKAN is a novel graph neural network (GNN) that replaces traditional multi-layer perceptrons (MLPs) and activation functions with Kolmogorov-Arnold Networks (KANs) for enhanced feature extraction. The paper argues that MLPs and activation functions hinder feature extraction due to information loss and limited representational capacity. Inspired by KANs, which use spline-based univariate functions for information aggregation, GraphKAN aims to improve the efficiency and interpretability of GNNs. The KANs are designed to replace MLPs in the node representation phase, enabling more effective feature extraction. The paper introduces GraphKAN, which incorporates KANs into GNNs and adds LayerNorm for stable learning. Experiments on real-world graph-like temporal signal data show that GraphKAN outperforms traditional GNNs like GCN in terms of classification accuracy, particularly when the number of labeled nodes is small. The results indicate that GraphKAN is a powerful tool for feature extraction in graph-like data, with potential applications in few-shot classification tasks. The paper also demonstrates that GraphKAN's feature extraction capability surpasses that of GCN, as evidenced by clustering comparisons of intermediate features. The experiments show that while GraphKAN has higher computational time than GCN, its improved accuracy makes it a valuable tool for graph-based tasks.GraphKAN is a novel graph neural network (GNN) that replaces traditional multi-layer perceptrons (MLPs) and activation functions with Kolmogorov-Arnold Networks (KANs) for enhanced feature extraction. The paper argues that MLPs and activation functions hinder feature extraction due to information loss and limited representational capacity. Inspired by KANs, which use spline-based univariate functions for information aggregation, GraphKAN aims to improve the efficiency and interpretability of GNNs. The KANs are designed to replace MLPs in the node representation phase, enabling more effective feature extraction. The paper introduces GraphKAN, which incorporates KANs into GNNs and adds LayerNorm for stable learning. Experiments on real-world graph-like temporal signal data show that GraphKAN outperforms traditional GNNs like GCN in terms of classification accuracy, particularly when the number of labeled nodes is small. The results indicate that GraphKAN is a powerful tool for feature extraction in graph-like data, with potential applications in few-shot classification tasks. The paper also demonstrates that GraphKAN's feature extraction capability surpasses that of GCN, as evidenced by clustering comparisons of intermediate features. The experiments show that while GraphKAN has higher computational time than GCN, its improved accuracy makes it a valuable tool for graph-based tasks.