Large⁃scale graph data,such as knowledge graphs and social networks,are ubiquitous. The enormous size of these graphs poses significant challenges for analytical tasks like frequent pattern mining (FPM). Graph reduction techniques have emerged as a key enabler for large⁃scale graph analysis,as they can drastically reduce graph size while preserving critical information. However,existing graph reduction methods primarily focus on retaining specific attribute information or minimizing global information loss,often neglecting the preservation of local high⁃order structures. This limitation leads to suboptimal support for FPM tasks. To address this issue,we propose a Structure⁃Preserving Graph Reduction Framework (SPGRF) that retains a higher quantity of high⁃quality frequent patterns while reducing data scale. First,we introduce a local⁃to⁃global edge importance evaluation method that guides the construction of an initial reduction skeleton by precisely identifying key nodes and core edges. Second,to compensate for structural degradation caused by reduction,we design a neighborhood⁃based skeleton enhancement mechanism to improve the structural diversity and completeness of the reduced graph. Extensive experiments on real⁃world graphs demonstrate the superiority of our framework across multiple metrics. The generated reduced graphs effectively preserve the original "backbone" structure: when the graph is reduced to 10% of its original size the accuracy of frequent patterns on the Wiki dataset reaches 96%.
图稀疏化主要通过去除冗余边或节点,在保留原图结构性质的同时生成相似的小图.早期方法多以结构保真为目标,例如,Althöfer et al[8]提出的Spanner算法采用一种贪心策略,在允许节点间距离被可控拉伸的前提下,以少量边保留原始图的近似距离结构,实现了有效的图稀疏化,奠定了稀疏图构建的理论基础.随后,研究焦点逐渐转向谱相似性保持,旨在在图拉普拉斯矩阵层面维持图的整体性质,代表性成果包括Feng[9]提出的相似性感知构造框架,有效兼顾了理论保证与计算的可扩展性.近年来,GNN的快速发展推动了稀疏化从结构驱动向模型性能导向演化.Razin et al[10]提出的WIS方法引入“步行指数”作为边重要性度量,移除对多步信息传播影响较小的边,提升GNN的效能.Zhao et al[11]进一步将公平性纳入优化目标,实现模型准确性与群体公平性的协调.Zhang et al[12]突破了统一稀疏率的约束,提出节点感知的个性化稀疏策略,更贴合异质图中个体差异的表示需求.随着图数据规模的增长,结构演化,动态图、高效计算以及可解释性成为最新研究热点.为了提升稀疏化的合理性,Akkas and Azad[13]引入Shapley来评估边影响力,在保障GNN性能的同时,实现有效约简.针对动态演化场景,Shabani et al[14]利用强化学习设计了STGS框架以保持关键时空模式.Yuan et al[15]提出dyGRASS算法,借助GPU加速的局部随机游走实现了极高效的动态谱稀疏化.除了基于边评分的稀疏化方法,另一类通过子图采样获取代表性结构的策略也在大规模图处理中得到广泛应用.森林火灾采样(Forest Fire Sampling,FF)[16]以模拟火势蔓延的方式递归选取邻接节点,适用于连通性强的图.钉球采样(Spikyball Sampling,SS)[17]在广度优先搜索中仅激活少量邻居,可以缓解高密度区域的过采样问题.针对传统随机游走存在的收敛速度慢、探索效率低等问题,CNARW方法[18]引入共同邻居信息对游走方向进行引导,以加速采样过程,提高探索多样性.此外,为了在采样中纳入更高阶结构,Wang et al[19]在BMiner挖掘框架中提出基于Motif的节点采样策略,借助Motif模式重新定义节点重要性,融合结构特征与属性语义,在保持信息完整性的同时显著压缩图规模.
图粗化通过节点聚合与信息压缩,生成更简洁的图结构以提升表示与计算效率.传统粗化方法多以最小化结构重建误差为目标.Loukas and Vandergheynst[20]提出的RSS框架首次为粗化过程提供了谱层面的形式化分析.Kumar et al[7]进一步将节点特征纳入粗化过程,提出联合结构与属性优化的粗化策略,提升了特征一致性.针对传统粗化方法易产生稠密摘要图的局限,Lee et al[21]提出的SSumM方法将节点合并与摘要图稀疏化两个过程相结合,在最小描述长度原则的指导下生成稀疏的摘要图.在GNN应用中,图粗化逐渐融入了模型训练流程.Huang et al[22]提出的SCAL方法结合粗化与知识迁移策略,先在小图上进行训练再迁移至原图,显著提升了训练效率.Bacciu et al[23]借鉴卷积神经网络设计,提出图下采样机制,解决了图数据缺乏规则结构的问题,拓展了GNN的层次化设计空间.
图凝聚技术进一步将约简提升至“可训练小图合成”层面,通过合成一个与原图结构无关的全新小图,强调在小图上尽可能重现原图的训练行为,其代价是牺牲了节点与边的直接可解释性.Jin et al[24]提出的GCond框架率先引入知识蒸馏思想,以梯度匹配方式驱动小图训练过程与原图一致,开启了图凝聚研究序章.为了缓解其在大图场景下的效率瓶颈,Jin et al[25]提出的DosCond,通过模型解耦与流程简化显著提升了可扩展性.Gao et al[26]创新性地将凝聚图作为反馈机制,引导原图去噪与学习协同优化.考虑到单步蒸馏误差累积,Zheng et al[27]进一步提出SFGC框架,引入“专家轨迹”引导小图模拟原图的长期训练动态,提升了模型的稳定性.Zhang et al[28]与Liu et al[29]分别从节点评分与结构可解释性的角度提出了新型蒸馏范式.为了规避高昂的双循环训练开销,Wang et al[30]提出GC⁃SNTK框架,采用图正切核逼近原图行为来形成封闭解方案,大幅简化计算.后续如KiDD[31],OpenGC[32]等方法,进一步扩展其在动态图与增量学习等复杂场景中的应用.
另一路线以分布匹配为核心,通过对齐特征分布来替代梯度监督.Xiao et al[33]提出更简洁的对齐策略SimGC,Gao et al[34]提出CGC,进一步摆脱了训练依赖,通过结构先验直接生成可用小图,极大地提升了图凝聚方法的效率与可落地性.
(1)大规模图稀疏化摘要算法(Sparse Summarization of Massive Graphs,SSumM)[21].在最小描述长度原则指导下,交替执行节点合并与超边稀疏化,以比特数为约束生成稀疏图摘要.为了使输出能应用于频繁模式挖掘下游任务,从超边的两个超点中选取支持度最大边作为超边的替代.
(2)行走指数稀疏化算法(Walk Index Sparsification,WIS)[10].按边对划分边界游走数的贡献由小到大排序,依次删边,以最小结构改动保住GNN对节点交互的建模能力.
(3)共同邻居感知随机游走采样算法(Common Neighbor Aware Random Walk,CNARW)[18].改进了传统随机游走,引入共同邻居信息指导游走方向.每一步选择下一跳节点时,计算当前节点与候选节点的共同邻居数量.根据共同邻居数量对候选节点加权,优先选择共同邻居多的节点.
(4)森林火灾采样算法(Forest Fire Sampling,FF)[16].模拟森林火灾蔓延的随机遍历,从一个随机种子节点开始,以设定的“燃烧概率”决定是否“点燃”每个邻居.被点燃的节点继续以同样方式向外蔓延,直至达到预设规模.
HashemiM, GongS B, NiJ T,et al. A comprehensive survey on graph reduction:Sparsification,coarsening,and condensation∥Proceedings of the 33rd International Joint Conference on Artificial Intelligence. Jeju,Korea (South):IJCAI,2024:8058-8066.
[3]
LiuY L, QiuR H, HuangZ. Cat:Balanced continual graph learning with graph condensation∥2023 IEEE International Conference on Data Mining. Shanghai,China:IEEE,2023:1157-1162.
[4]
WickmanR, ZhangX F, LiW Z. A generic graph sparsification framework using deep reinforcement learning∥2022 IEEE International Conference on Data Mining. Orlando,FL,USA:IEEE,2022:1221-1226.
[5]
SeoH, YunJ H, YangE. TEDDY:Trimming edges with degree⁃based discrimination strategy∥The 12th International Conference on Learning Represen⁃tations. Vienna,Austira:ICLR,2023:1-12.
[6]
MengY C, LiR H, LinL L,et al. Topology⁃preserving graph coarsening:An elementary collapse⁃based approach. Proceedings of the VLDB Endowment,2024,17(13):4760-4772.
[7]
KumarM, SharmaA, SaxenaS,et al. Featured graph coarsening with similarity guarantees∥Proceedings of the 40th International Conference on Machine Learning. New York,NY,USA:PMLR,2023:17953-17975.
[8]
AlthöferI, DasG, DobkinD,et al. On sparse spanners of weighted graphs. Discrete & Compu⁃tational Geometry,1993,9(1):81-100.
[9]
FengZ. GRASS:Graph spectral sparsification leveraging scalable spectral perturbation analysis. IEEE Transactions on Computer:Aided Design of Integrated Circuits and Systems,2020,39(12):4944-4957.
[10]
RazinN, VerbinT, CohenN. On the ability of graph neural networks to model interactions between vertices. Advances in Neural Information Processing Systems,2023,36:26501-26545.
[11]
ZhaoJ X, HuangT J, LiuS W,et al. FS⁃GNN:Improving fairness in graph neural networks via joint sparsification. Neurocomputing,2025,648:130641.
[12]
ZhangG B, SunX G, YueY W,et al. Graph sparsification via mixture of graphs∥The 13th International Conference on Learning Represen⁃tations. Singapore:ICLR,2025:92735-92763.
[13]
AkkasS, AzadA. Explainable graph sparsification with Shapley values∥Proceedings of the ACM Web Conference 2026. New York,NY,USA:Association for Computing Machinery,2026:8557-8560.
[14]
ShabaniN, BeheshtiA, QiY K,et al. STGS:Spatio⁃temporal graph sparsification using reinforcement learning∥Proceedings of the 34th ACM International Conference on Information and Knowledge Management. New York,NY,USA:Association for Computing Machinery,2025:2546-2555.
[15]
YuanY H, AghdaeiA, FengZ. dyGRASS:Dynamic spectral graph sparsification via localized random walks on GPUs∥2025 IEEE/ACM International Conference on Computer Aided Design.Munich,Germany:IEEE,2025:1-9.
[16]
ZhangX, NiY, LiS Y,et al. A survey of large graph sampling techniques. Journal of Computer⁃Aided Design & Computer Graphics,2022,34(12):1805-1814.
[17]
RicaudB, AspertN, MizV. Spikyball sampling:Exploring large networks via an inhomogeneous filtered diffusion. Algorithms,2020,13(11):275.
[18]
LiY K, WuZ Y, LinS,et al. Walking with perception:efficient random walk sampling via common neighbor awareness∥2019 IEEE 35th International Conference on Data Engineering.Macao,China:IEEE,2019:962-973.
[19]
WangX, ShiJ H, ZouJ J,et al. Supports estimation via graph sampling. Expert Systems with Applica⁃tions,2024,240:122554.
[20]
LoukasA, VandergheynstP. Spectrally approxi⁃mating large graphs with smaller graphs∥Proceedings of the 35th International Conference on Machine Learning. New York,NY,USA:PMLR,2018:3237-3246.
[21]
LeeK, JoH, KoJ,et al. Ssumm:Sparse summarization of massive graphs∥Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York,NY,USA:Association for Computing Machinery,2020:144-154.
[22]
HuangZ F, ZhangS Z, XiC,et al. Scaling up graph neural networks via graph coarsening∥Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. New York,NY,USA:Association for Computing Machinery,2021:675-684.
[23]
BacciuD, ConteA, LandolfiF. Generalizing downsampling from regular data to graphs. Proceedings of the AAAI Conference on Artificial Intelligence,2023,37(6):6718-6727.
[24]
JinW, ZhaoL X, ZhangS C,et al. Graph condensation for graph neural networks∥The 10th International Conference on Learning Represen⁃tations. Online:ICLR,2022:1-19.
[25]
JinW, TangX F, JiangH M,et al. Condensing graphs via one⁃step gradient matching∥Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York,NY,USA:Association for Computing Machinery,2022:720-730.
[26]
GaoX Y, YinH Z, ChenT,et al. RobGC:Towards robust graph condensation. IEEE Transactions on Knowledge and Data Engineering,2025,37(8):4791-4804.
[27]
ZhengX, ZhangM, ChenC Y,et al. Structure⁃free graph condensation:From large⁃scale graphs to condensed graph⁃free data. Advances in Neural Information Processing Systems,2023,36:6026-6047.
[28]
ZhangY C, ZhangT L, WangK,et al. Navigating complexity∥Proceedings of the 41st International Conference on Machine Learning. Vienna,Austria:JMLR.org,2024:60379-60395.
[29]
LiuZ Y, ZengC L, ZhengG J. Graph data condensation via self⁃expressive graph structure reconstruction∥Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York,NY,USA:Association for Computing Machinery,2024:1992-2002.
[30]
WangL, FanW Q, LiJ T,et al. Fast graph condensation with structure⁃based neural tangent kernel∥Proceedings of the ACM Web Conference 2024. New York,NY,USA:Association for Computing Machinery,2024:4439-4448.
[31]
XuZ, ChenY Z, PanM H,et al. Kernel ridge regression⁃based graph dataset distillation∥Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York,NY,USA:Association for Computing Machinery,2023:2850-2861.
[32]
GaoX Y, ChenT, ZhangW T,et al. Graph condensation for open⁃world graph learning∥Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York,NY,USA:Association for Computing Machinery,2024:851-862.
[33]
XiaoZ B, WangY, LiuS Y,et al. Simple graph condensation∥Machine Learning and Knowledge Discovery in Databases. Research Track. Cham,Switzerland:Springer,2024:53-71.
[34]
GaoX Y, YeG H, ChenT,et al. Rethinking and accelerating graph condensation:A training⁃free approach with class partition∥Proceedings of the ACM on Web Conference 2025. New York,NY,USA:Association for Computing Machinery,2025:4359-4373.
[35]
WangX, LanZ, HeY A,et al. A cost⁃effective approach for mining near⁃optimal top⁃k patterns. Expert Systems with Applications,2022,202:117262.
[36]
BringmannB, NijssenS. What is frequent in a single graph?∥Advances in Knowledge Discovery and Data Mining. Heidelberg,Germany:Springer,2008:858-863.
[37]
RossiR A, AhmedN K. The network data repository with interactive graph analytics and visualization∥Proceedings of the 29th AAAI Conference on Artificial Intelligence. Menlo Park,CA,USA:AAAI Press,2015:4292-4293.
[38]
YangJ, LeskovecJ. Defining and evaluating network communities based on ground⁃truth∥Proceedings of the ACM SIGKDD Workshop on Mining Data Semantics. Beijing,China:Association for Computing Machinery,2012:1-8
[39]
TalukderN, ZakiM J. A distributed approach for graph mining in massive networks. Data Mining and Knowledge Discovery,2016,30(5):1024-1052.
YinH, BensonA R, LeskovecJ,et al. Local higher⁃order graph clustering∥Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York,NY,USA:Association for Computing Machinery,2017,2017:555-564.
[42]
LinM K, LiW Z, LuS L. Balanced influence maximization in attributed social network based on sampling∥Proceedings of the 13th International Conference on Web Search and Data Mining. New York,NY,USA:Association for Computing Machinery,2020:375-383.
[43]
ZhouZ G, ShiC, ShenX L,et al. Context⁃aware sampling of large networks via graph representation learning. IEEE Transactions on Visualization and Computer Graphics,2021,27(2):1709-1719.
[44]
KioucheA E, BasteJ, HaddadM,et al. Neighborhood⁃preserving graph sparsification. Proceedings of the VLDB Endowment,2024,17(13):4853-4866.
[45]
DehmerM, MowshowitzA. A history of graph entropy measures. Information Sciences,2011,181(1):57-78.