MDTE: Minority-Aware Diffusion over Temporal Edge Events for Imbalanced Node Classification

Zhou Zelong; Zhang Tianming; Yang Zhengyi; Tang Yifu; Hou Chenyu; Cao Bin; Fan Jing

cs.LGAug 25, 2026

MDTE: Minority-Aware Diffusion over Temporal Edge Events for Imbalanced Node Classification

Authors: Zhou Zelong, Zhang Tianming, Yang Zhengyi, Tang Yifu, Hou Chenyu, Cao Bin, Fan Jing

Organizations: Zhejiang University of Technology Hangzhou, China · The University of Sydney Sydney, Australia · Vecton AI Australia

Abstract

Class-imbalanced node classification on temporal graphs is challenging because majority-dominated temporal propagation progressively assimilates minority representations, while conventional node and neighborhood information provides insufficient discriminative evidence for minority classes. To address these issues, we propose MDTE, a minority-aware diffusion framework that reconstructs stable and discriminative temporal edge-event representations through conditional diffusion denoising. Specifically, MDTE introduces Distribution-Aware Selective Propagation, which combines Local Outlier Factor (LOF)-based propagation filtering with cluster-aware low-frequency propagation. The module preserves informative neighborhood dependencies while mitigating harmful propagation and majority-class information assimilation. It further develops Multi-View Discriminative Fusion, which exploits feature reconstruction and topology prediction to characterize class-wise differences in distribution learning and extracts complementary discriminability signals to guide denoising. Experiments on five real-world datasets demonstrate that MDTE consistently achieves the best performance on minority-class-oriented metrics, improving minority-class recall by up to 23.53 percentage points, minority-class F1 by 8.68 percentage points, and AUPRC by 2.67 percentage points over the strongest baselines.

Figures & tables

Dataset	Nodes	Temporal edges	Minority	Majority	IR
DGraph-Fin	3,700,550	4,300,999	15,509	1,210,092	78.03
Elliptic	203,769	234,355	4,545	42,019	9.25
Elliptic++ Transactions	203,769	234,355	4,545	42,019	9.25
Elliptic++ Actors	822,942	2,868,964	14,266	251,088	17.60
Ethereum	2,973,489	13,551,303	1,165	2,972,324	2,551.35

Table 1. Dataset statistics.

Dataset	Metric	TGC	DyGFormer	TGAT	TGN	GraphPAE	DGMAE	GraphSMOTE	LTE4G	GraphENS	DDM	SDMG	LGD	TGAT+Conda	MDTE
DGraph-Fin	AUROC	$72.89{\pm}.27$	$76.10{\pm}.23$	$72.06{\pm}.12$	$73.41{\pm}.15$	$70.92{\pm}.31$	$70.61{\pm}.18$	$72.46{\pm}.08$	$71.90{\pm}.04$	$71.85{\pm}.06$	$72.34{\pm}.06$	$74.59{\pm}.60$	$75.48{\pm}.11$	$74.38{\pm}.14$	$\boldsymbol{81.19{\pm}0.23}$
	AUPRC	$2.89{\pm}.09$	$3.69{\pm}.11$	$3.14{\pm}.02$	$3.12{\pm}.04$	$2.92{\pm}.05$	$3.00{\pm}.05$	$2.73{\pm}.02$	$2.64{\pm}.01$	$3.07{\pm}.03$	$3.32{\pm}.06$	$3.54{\pm}.13$	$3.67{\pm}.04$	$3.25{\pm}.06$	$\boldsymbol{4.76{\pm}0.07}$
	Recall-min.	$21.47{\pm}12.03$	$32.01{\pm}6.46$	$30.37{\pm}6.03$	$28.95{\pm}7.42$	$21.02{\pm}11.21$	$18.30{\pm}4.44$	$21.90{\pm}14.72$	$20.68{\pm}11.92$	$17.33{\pm}4.00$	$19.25{\pm}5.21$	$19.60{\pm}4.41$	$25.16{\pm}7.80$	$24.73{\pm}5.59$	$\boldsymbol{55.54{\pm}0.95}$
	F1-min.	$5.64{\pm}.32$	$7.91{\pm}.51$	$6.82{\pm}.04$	$6.61{\pm}.06$	$5.90{\pm}.25$	$6.75{\pm}.20$	$5.50{\pm}.29$	$5.45{\pm}.30$	$6.34{\pm}.20$	$6.95{\pm}.09$	$7.18{\pm}.23$	$7.47{\pm}.29$	$6.64{\pm}.11$	$\boldsymbol{8.24{\pm}0.10}$
	Macro-F1	$49.69{\pm}0.02$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$49.68{\pm}0.00$	$\boldsymbol{49.72{\pm}0.06}$
Elliptic	AUROC	$98.38{\pm}.13$	$92.74{\pm}.44$	$96.95{\pm}.13$	$\boldsymbol{98.47{\pm}.05}$	$97.07{\pm}.14$	$94.88{\pm}.24$	$95.53{\pm}.07$	$96.54{\pm}.16$	$98.28{\pm}.04$	$97.94{\pm}.14$	$98.29{\pm}.09$	$95.23{\pm}.24$	$98.33{\pm}.06$	$98.21{\pm}0.06$

Table 2. Node classification results (%). The best and second-best results in each metric row are shown in bold and underlined, respectively.

Variant	DGraph-Fin					Elliptic
Variant	AUROC	AUPRC	Recall-min.	F1-min.	Macro-F1	AUROC	AUPRC	Recall-min.	F1-min.	Macro-F1
w/o Temporal Edge-Event Construction	75.54 $\pm$ 0.12	3.52 $\pm$ 0.09	41.35 $\pm$ 3.54	6.79 $\pm$ 0.18	49.69 $\pm$ 0.02	97.77 $\pm$ 0.11	91.83 $\pm$ 0.24	83.99 $\pm$ 1.51	85.59 $\pm$ 0.30	92.10 $\pm$ 0.56
w/o Distribution-Aware Selective Propagation	80.64 $\pm$ 0.23	4.57 $\pm$ 0.08	50.88 $\pm$ 3.34	8.11 $\pm$ 0.07	49.68 $\pm$ 0.00	98.11 $\pm$ 0.09	93.13 $\pm$ 0.13	85.00 $\pm$ 1.11	86.71 $\pm$ 0.48	92.85 $\pm$ 0.56
w/o Multi-View Discriminative Fusion	80.90 $\pm$ 0.29	4.67 $\pm$ 0.06	53.44 $\pm$ 0.99	8.08 $\pm$ 0.04	49.71 $\pm$ 0.02	97.93 $\pm$ 0.18	92.35 $\pm$ 0.50	83.96 $\pm$ 1.92	86.29 $\pm$ 0.41	92.91 $\pm$ 0.39
w/o Debiased Contrastive Learning	81.01 $\pm$ 0.39	4.71 $\pm$ 0.19	50.07 $\pm$ 7.76	8.09 $\pm$ 0.05	49.69 $\pm$ 0.02	97.75 $\pm$ 0.14	91.63 $\pm$ 0.18	83.34 $\pm$ 1.25	86.04 $\pm$ 0.66	92.31 $\pm$ 0.38
Full Model	81.19 $\pm$ 0.23	4.76 $\pm$ 0.07	55.54 $\pm$ 0.95	8.24 $\pm$ 0.10	49.72 $\pm$ 0.06	98.21 $\pm$ 0.06	93.61 $\pm$ 0.15	86.44 $\pm$ 0.37	88.54 $\pm$ 0.31	93.68 $\pm$ 0.17

Table 3. Ablation results (%). The best and second-best results in each column are shown in bold and underlined, respectively.

Node Classification

Node classification assigns labels to graph entities by combining their features with information from connected nodes and graph structure.

See all 100 papers

NodeImport: Imbalanced Node Classification with Node Importance Assessment
Jul 15, 2026Nan Chen, Zemin Liu, Bryan Hooi +3Class-Imbalanced Learning Graph Neural Networks
Johns Hopkins University Baltimore, Maryland, USA · Zhejiang University Hangzhou, Zhejiang, China · National University of Singapore Singapore, Singapore +1
GraphIFE: Rethinking Graph Imbalance Node Classification via Invariant Learning
Sep 28, 2025Fanlong Zeng, Wensheng Gan, Kangjie Chen +1Class-Imbalanced Learning Graph Neural Networks
School of Intelligent Systems Science and Engineering, Jinan University, Zhuhai 519070, China · School of Cybersecurity, Tianjin University, Tianjin 300354, China · University of Illinois Chicago, Chicago, USA
Contrastive Representation-Guided Genetic Minority Oversampling for Imbalanced Time-Series Classification
Aug 24, 2026Wenbin Pei, Yunrong Hao, Zhen Liu +4Class-Imbalanced Learning Synthetic Data Augmentation
School of Computer Science and Technology, Dalian University of Technology, Dalian 116024, China; Key Laboratory of Social Computing and Cognitive Intelligence (Dalian University of Technology), Ministry of Education, Dalian 116024, China · College of Computing and Data Science, Nanyang Technological University, Singapore · School of Mechanics and Aerospace Engineering, Dalian University of Technology, Dalian 116024, China +3

Jul 15, 2026J/K move · Enter open · S save

NodeImport: Imbalanced Node Classification with Node Importance Assessment

Nan Chen, Zemin Liu, Bryan Hooi, Bingsheng He+2

Johns Hopkins University Baltimore, Maryland, USA · Zhejiang University Hangzhou, Zhejiang, China · National University of Singapore Singapore, Singapore+1

Class-Imbalanced Learning Graph Neural Networks Data Valuation Node Classification

Open paper AskPDF arXiv

In real-world applications, node classification on graphs often faces the challenge of class imbalance, where majority classes dominate training, resulting in biased model performance. Traditional GNNs often struggle in such scenarios, as they tend to overfit to majority classes while underrepresenting minority classes. Existing solutions, which either prioritize nodes based on class size or synthesize new nodes for minority classes, often fall short of effectively addressing this imbalance issue. This paper introduces an approach to class-imbalanced node classification by utilizing a balanced meta-set for importance measurement, where a training node is considered significant if it enhances model performance under an unbiased setting. Our method identifies important nodes that can counteract class imbalance and utilizes them for model training, allowing for fine-grained and dynamic node selection throughout the training process. We theoretically derive a formula to directly assess node importance, reducing computational overhead and providing an intuitive threshold for node selection. Guided by this metric, we develop a novel framework that filters valuable labeled, unlabeled, and synthetic nodes that enhance model performance in an unbiased context. A key advantage of this framework is its separation of the synthetic node generation process from the filtering process, ensuring compatibility with various node generation methods. Furthermore, we introduce a strategy to construct a high-quality meta-set that closely approximates the overall feature distribution, ensuring robust representation of each class. We evaluate our framework, NodeImport, across multiple datasets using popular GNN architectures, demonstrating its superiority over existing baselines. Our results highlight the flexibility and effectiveness of the framework in mitigating class imbalance, leading to improved outcomes.

Abstract

Node Classification

Explore similar work

NodeImport: Imbalanced Node Classification with Node Importance Assessment

GraphIFE: Rethinking Graph Imbalance Node Classification via Invariant Learning

Contrastive Representation-Guided Genetic Minority Oversampling for Imbalanced Time-Series Classification