Node Classification Based on Kolmogorov-Arnold Networks

doi:10.3778/j.issn.1673-9418.2408026

Abstract

Abstract: Most graph deep learning methods extract feature information from graph data by using learnable weights and specific activation functions. The selection of different activations has a significant impact on performance. Aiming at the above problems, this paper proposes a fully connected neural model G-KAN based on the Kolmogorov-Arnold network (KAN), which does not require specific activations and explicit node message passing modules. It dynamically learns activations through KAN and introduces a contrastive loss guided by node similarity to implicitly extract the original graph information. Firstly, G-KAN maps the graph data to the feature space through a linear layer. Secondly, the KAN layers extract the latent features from the input data. Thirdly, G-KAN maps the output to a probability distribution of labels using a linear layer and a Softmax function. Finally, the contrastive loss is used to optimize the output of the KAN layer, encouraging high-similarity nodes to be close to each other and low-similarity nodes to be far away. In the node classification task, G-KAN outperforms the currently advanced baselines and raises Micro-F1 and Macro-F1 by 50.42 and 52.84 percentage points respectively compared with graph convolutional networks (GCN) on BlogCatalog. In activation comparison experiments, the method introduced by KAN outperforms the variants that use different activation functions and demonstrates stronger generalization across various datasets. The results show that the learnable activation strategy used by G-KAN can enhance the representation capacity of fully connected neural networks and generate low-dimensional node representations with higher discriminability.

Key words: graph convolutional networks, multi-layer perceptron, Kolmogorov-Arnold networks, contrastive learning, node classification

摘要： 多数图深度学习模型通过可学习权重加固定激活函数的方式提取图数据的特征信息，采用不同激活函数时对模型性能有较为显著的影响。针对上述问题，提出了一种基于Kolmogorov-Arnold网络（KAN）的全连接神经网络模型G-KAN，无需特定的激活函数和显式的节点信息传递策略，通过KAN动态学习激活函数，并引入节点相似度引导的对比损失隐式提取原始图特征信息。G-KAN通过线性层将图数据映射到特征空间，通过KAN层提取输入数据中的潜在特征，通过线性层和Softmax函数将KAN层的输出映射为节点标签的概率分布，并引入对比损失对KAN层的输出进行优化，推动高相似度节点彼此接近、低相似度节点彼此远离。在节点分类任务中，G-KAN优于当前较为先进的基线模型，特别是在BlogCatalog数据集上，G-KAN的Micro-F1和Macro-F1相较图卷积网络（GCN）提高了50.42和52.84个百分点。在激活函数对比实验中，引入KAN的方法不仅优于采用不同激活函数的变体，对不同数据集的泛化能力也更强。上述实验结果表明，G-KAN采用的可学习激活函数策略能够提高全连接神经网络的表征能力，使生成的低维节点表示具有更高的区分性。

关键词: 图卷积网络, 多层感知机, Kolmogorov-Arnold网络, 对比学习, 节点分类

YUAN Lining, FENG Wengang, LIU Zhao. Node Classification Based on Kolmogorov-Arnold Networks[J]. Journal of Frontiers of Computer Science and Technology, 2025, 19(3): 645-656.

袁立宁, 冯文刚, 刘钊. 基于Kolmogorov-Arnold网络的节点分类算法[J]. 计算机科学与探索, 2025, 19(3): 645-656.

References

[1] YANG X H, WANG Y Q, LIU Y, et al. Mixed graph contrastive network for semi-supervised node classification[J]. ACM Transactions on Knowledge Discovery from Data, 2024, 18(7): 162.
[2] XU M J. Understanding graph embedding methods and their applications[J]. SIAM Review, 2021, 63(4): 825-853.
[3] DU G D, ZHANG J, ZHANG N, et al. Semi-supervised imbalanced multi-label classification with label propagation[J]. Pattern Recognition, 2024, 150: 110358.
[4] KHEMANI B, PATIL S, KOTECHA K, et al. A review of graph neural networks: concepts, architectures, techniques, challenges, datasets, applications, and future directions[J]. Journal of Big Data, 2024, 11(1): 18.
[5] WU Z H, LIN X C, LIN Z H, et al. Interpretable graph convolutional network for multi-view semi-supervised learning[J]. IEEE Transactions on Multimedia, 2023, 25: 8593-8606.
[6] CHEN F H, LI S L, HAN J L, et al. Review of lightweight deep convolutional neural networks[J]. Archives of Computational Methods in Engineering, 2024, 31(4): 1915-1937.
[7] LIANG Q, WANG W, BAO F, et al. L2GC: lorentzian linear graph convolutional networks for node classification[C]//Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation, 2024: 9988-9998.
[8] HUANG A P, LU J L, WU Z H, et al. Geometric localized graph convolutional network for multi-view semi-supervised classification[J]. Information Sciences, 2024, 677: 120769.
[9] CHEN D, O’BRAY L, BORGWARDT K. Structure-aware transformer for graph representation learning[C]//Proceedings of the 39th International Conference on Machine Learning, 2022: 3469-3489.
[10] 文竹, 袁立宁, 黄伟, 等. 基于图多层感知机的节点分类算法[J]. 广西科学, 2023, 30(5): 942-950.
WEN Z, YUAN L N, HUANG W, et al. Node classification based on graph multi-layer perceptron[J]. Guangxi Sciences, 2023, 30(5): 942-950.
[11] LIU Z M, WANG Y X, VAIDYA S, et al. KAN: Kolmogorov-Arnold networks[EB/OL]. [2024-06-23]. https://arxiv.org/abs/2404.19756.
[12] 谢逢洁, 杨盼盼, 崔文田. 复杂网络理论中的高阶网络研究综述[J]. 系统科学学报, 2024, 32(1): 121-127.
XIE F J, YANG P P, CUI W T. Review of higher-order network research in complex network theory[J]. Journal of Systems Science, 2024, 32(1): 121-127.
[13] 高阳, 张宏莉. 基于随机游走的社区发现方法综述[J]. 通信学报, 2023, 44(6): 198-210.
GAO Y, ZHANG H L. Survey on community detection method based on random walk[J]. Journal on Communications, 2023, 44(6): 198-210.
[14] GUI C. Link prediction based on spectral analysis[J]. PLoS One, 2024, 19(1): e0287385.
[15] 刘俊奇, 涂文轩, 祝恩. 图卷积神经网络综述[J]. 计算机工程与科学, 2023, 45(8): 1472-1481.
LIU J Q, TU W X, ZHU E. Survey on graph convolutional neural network[J]. Computer Engineering & Science, 2023, 45(8): 1472-1481.
[16] WU F, SOUZA A, ZHANG T, et al. Simplifying graph convolutional networks[C]//Proceedings of the 36th International Conference on Machine Learning, 2019: 6861-6871.
[17] VELIČKOVIĆ P, CUCURULL G, CASANOVA A, et al. Graph attention networks[EB/OL]. [2024-06-23]. https://arxiv.org/abs/1710.10903.
[18] SOYDANER D. Attention mechanism in neural networks: where it comes and where it goes[J]. Neural Computing and Applications, 2022, 34(16): 13371-13385.
[19] BRODY S, ALON U, YAHAV E. How attentive are graph attention networks?[C]//Proceedings of the 2022 International Conference on Learning Representations, 2022: 1-26.
[20] KIM D, OH A. How to find your friendly neighborhood: graph attention design with self-supervision[EB/OL]. [2024- 06-23]. https://arxiv.org/abs/2204.04879.
[21] SHEHZAD A, XIA F, ABID S, et al. Graph transformers: a survey[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2407.09777.
[22] AGHAEI A A. rKAN: rational Kolmogorov-Arnold networks[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2406.14495.
[23] ABUEIDDA D W, PANTIDIS P, MOBASHER M E. DeepOKAN: deep operator network based on Kolmogorov Arnold networks for mechanics problems[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2405.19143.
[24] DROKIN I. Kolmogorov-Arnold convolutions: design principles and empirical studies[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2407.01092.
[25] WANG Y Z, SUN J, BAI J S, et al. Kolmogorov Arnold informed neural network: a physics-informed deep learning framework for solving forward and inverse problems based on Kolmogorov Arnold networks[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2406.11045.
[26] KIAMARI M, KIAMARI M, KRISHNAMACHARI B. GKAN: graph Kolmogorov-Arnold networks[EB/OL]. [2024- 08-13]. https://arxiv.org/abs/2406.06470.
[27] ZHANG F, ZHANG X. GraphKAN: enhancing feature extraction with graph Kolmogorov Arnold networks[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2406.13597.
[28] XU J F, CHEN Z Y, LI J Z, et al. FourierKAN-GCF: Fourier Kolmogorov-Arnold network: an effective and efficient feature transformation for graph collaborative filtering[EB/OL]. [2024-08-13]. https://arxiv.org/abs/2406.01034.
[29] YUAN L N, JIANG P, HOU W L, et al. G-MLP: graph multi-layer perceptron for node classification using contrastive learning[J]. IEEE Access, 2024, 12: 104909-104919.
[30] 袁立宁, 蒋萍, 莫嘉颖, 等. 基于二阶图卷积自编码器的图表示学习[J]. 计算机工程与应用, 2024, 60(10): 180-187.
YUAN L N, JIANG P, MO J Y, et al. Graph representation learning using second-order graph convolutional autoencoders[J]. Computer Engineering and Applications, 2024, 60(10): 180-187.
[31] 陈琪, 郭涛, 邹俊颖. 基于双重视图耦合的自监督图表示学习模型[J]. 计算机工程与设计, 2023, 44(12): 3738-3744.
CHEN Q, GUO T, ZOU J Y. Self-supervised graph representation learning model with dual view coupling[J]. Computer Engineering and Design, 2023, 44(12): 3738-3744.
[32] WANG X, ZHU M Q, BO D Y, et al. AM-GCN: adaptive multi-channel graph convolutional networks[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2020: 1243-1253.
[33] YUAN L N, JIANG P, WEN Z, et al. Improving variational graph autoencoders with multi-order graph convolutions[J]. IEEE Access, 2024, 12: 46919-46929.
[34] LI P Z, PEI Y, LI J Q. A comprehensive survey on design and application of autoencoder in deep learning[J]. Applied Soft Computing, 2023, 138: 110176.
[35] HU Y, YOU H X, WANG Z C, et al. Graph-MLP: node classification without message passing in graph[EB/OL]. [2024-06-23]. https://arxiv.org/abs/2106.04051.
[36] ZHANG X, XU Y H, HE W, et al. A comprehensive review of the oversmoothing in graph neural networks[C]//Proceedings of the 18th CCF Conference on Computer Supported Cooperative Work and Social Computing. Singapore: Springer, 2023: 451-465.