Chasing Label Shifters: A Change-Aware Framework for Dynamic Graph Node Classification

Junghoon Kim; Seungyoon Choi; Hyunsung Kim; Chanyoung Park

cs.LGOct 13, 2026

Chasing Label Shifters: A Change-Aware Framework for Dynamic Graph Node Classification

Authors: Junghoon Kim, Seungyoon Choi, Hyunsung Kim, Chanyoung Park

Organizations: KAIST

Abstract

Entities in real-world systems often evolve over time: users shift between information consumption patterns, researchers migrate between fields, and firms transition between financial risk states. When such systems are modeled as dynamic graphs, these transitions correspond to nodes whose class labels change between consecutive snapshots, which we call shifters. Correctly classifying shifters is often more consequential than classifying stable nodes with persistent labels, as detecting such transitions enables timely intervention in high-stakes settings. Yet, we identify a systematic failure mode across all major dynamic graph neural network architectures: shifter performance consistently lags behind stable nodes, with errors concentrated on the old label. Our structural analysis traces this failure to embedding inertia: neighborhood aggregation keeps shifter representations anchored to the old-class centroid, and this effect is strongest precisely where it is hardest to correct, namely for shifters whose neighborhoods remain aligned with the old class. Guided by this finding, we propose CHASE, a model-agnostic CHange-Aware framework for Shifting nodEs that wraps any dynamic GNN with targeted components for detecting label shifts and overriding stale neighborhood signals. CHASE consistently improves shifter performance across all tested models and datasets (up to 114.9% and 202.5% improvement on accuracy and F1 score, respectively) while preserving stable-node performance. We additionally contribute three dynamic graph benchmarks with naturally shifting labels, filling a gap in existing resources. The source code is available at https://github.com/jhkim611/CHASE

Figures & tables

Figure 1: In misinformation propagation, recent shifters are more influential than stable entities, making their misclassification much more critical.

Dataset	Domain	# Nodes	# Edges	Snapshots (Split)	Classes	Label meaning	Shifter %
DBLP-S	Academic	34,717	394,385	14 (9/2/3)	2	Research area	2.62%
GossipCop	Social Media	6,091	2,000,390	37 (26/5/6)	2	$>$ 50% fake news consumption	4.06%
Music	Music Streaming	878	1,009,457	24 (15/4/5)	5	Dominant genre	18.77%

Table 1: Dataset statistics. Shifter % is the fraction of nodes whose label differs from the previous snapshot, averaged across snapshots.

Figure 2: (a), (b) The shifter performance gap is universal across all models and datasets. (c) Shifter errors are predominantly stuck in the previous snapshot’s label.

Figure 3: (Left) All four signals distinguish shifters from stable nodes (Top row), but only homophily signals meaningfully separate correctly from incorrectly classified shifters (Bottom row) (Mann-Whitney U test,

p\approx 0

in all cases). (Right) Correctly classified shifters show positive centroid displacement while incorrectly classified ones do not, confirming embedding inertia. Averaged across models and datasets; per-model and per-dataset breakdowns in Appendix B .

Figure 4: The CHASE architecture. Given snapshots

\mathcal{G}^{(t-1)},\mathcal{G}^{(t)}

and a dynamic GNN backbone, the (a) homophily-guided contrastive loss pulls shifter embeddings toward their new-class centroid. Then, the (b) change detection head estimates the probability

p_{\mathrm{change}}

that a node has shifted. Finally, the (c) change-aware classifier receives change signals to guide its prediction.

Figure 5: Performance gap between shifters and stable nodes is significantly reduced (up to

-73.0\%

) under CHASE.

	DBLP-S				GossipCop				Music
Model	Sh.Acc	Sh.F1	St.Acc	St.F1	Sh.Acc	Sh.F1	St.Acc	St.F1	Sh.Acc	Sh.F1	St.Acc	St.F1
GCN (collapsed)	73.0	76.5	96.0	97.1	48.6	22.1	69.7	40.1	38.1	23.7	64.5	22.9
GCN (seq)	71.8	75.5	96.1	97.1	52.8	25.0	75.6	45.3	46.3	29.0	73.6	45.0
+CHASE	83.2 +15.9%	85.9 +13.8%	96.2 +0.1%	97.2 +0.1%	69.6 +31.8%	60.1 +140.4%	87.7 +16.0%	76.9 +69.8%	80.0 +72.8%	78.6 +171.0%	93.8 +27.4%	89.5 +98.9%
EERM (collapsed)	73.5	77.4	96.2	97.3	47.5	20.0	69.6	44.5	39.2	22.4	64.7	25.6
EERM (seq)	73.6	77.3	96.3	97.3	52.5	23.4	76.1	47.1	44.0	24.6	73.8	45.4

Table 2: Main results. Shifter (Sh.) and stable node (St.) accuracy (Acc) and F1 score for each backbone, without and with CHASE augmentation. On shaded rows, bold values indicate improvement over the base, with relative change shown as superscripts. All values are 5-seed means in percentage points; standard deviations are reported in Appendix E .

	DBLP-S		GossipCop		Music
Config	Sh.Acc	Sh.F1	Sh.Acc	Sh.F1	Sh.Acc	Sh.F1
Baseline	63.9 ±1.0	68.4 ±1.1	58.2 ±4.7	34.1 ±12.4	50.1 ±7.0	35.2 ±13.6
+Contr.	65.9 ±0.6	69.9 ±0.6	60.4 ±4.8	45.5 ±3.4	53.6 ±7.0	44.6 ±7.4
+Hom.wt.	68.2 ±1.0	71.9 ±1.1	61.9 ±4.5	48.7 ±3.7	58.1 ±4.6	51.7 ±4.5
+Cls.	70.7 ±1.1	73.3 ±1.8	63.4 ±3.6	51.3 ±3.6	62.2 ±4.2	56.7 ±3.6
+Detect.	76.6 ±1.8	80.2 ±2.2	68.4 ±1.4	59.2 ±1.4	76.5 ±1.0	75.3 ±1.0

Table 3: Ablation study on GCN-GRU.

Figure 6: Performance improvements after CHASE augmentation, averaged across all models. (Left) Accuracy and F1 score gains are greater for nodes with lower current homophily. (Top right) Classification error on shifters is substantially reduced. (Bottom right) On the multiclass Music dataset, shifts away from the dominant class benefit much more than shifts toward it.

Baseline $\Delta_{c}$	Baseline Acc. (%)	Gain from +CHASE (%p)
$<-0.25$ (most inertial)	3.0	+40.1
$-0.25$ to $-0.05$	15.9	+37.5
$-0.05$ to $0.05$	61.1	+17.9
$>0.05$ (non-inertial)	94.0	0.0

Table 4: Shifter accuracy by baseline inertia (

\Delta_{c}

), pooled over all backbones and datasets.

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 7: All four signals distinguish shifters from stable nodes, consistently across all three datasets.

Figure 8: Homophily signals meaningfully separate correctly from incorrectly classified shifters while the drift signals lose their discriminative power, consistently across all three datasets.

Figure 9: The homophily patterns (high current homophily and low previous homophily for correctly classified shifters, vice versa for incorrectly classified shifters) are consistent across all models and datasets.

Figure 10: Correctly classified shifters generally show positive centroid displacement and incorrectly classified shifters generally display negative centroid displacement, consistently across all models and datasets.

	DBLP-S				GossipCop				Music
Model	Sh.Acc	Sh.F1	St.Acc	St.F1	Sh.Acc	Sh.F1	St.Acc	St.F1	Sh.Acc	Sh.F1	St.Acc	St.F1
GCN (collapsed)	73.0 ${}_{\pm\textbf{0.2}}$	76.5 ${}_{\pm\textbf{0.4}}$	96.0 ${}_{\pm\textbf{0.0}}$	97.1 ${}_{\pm\textbf{0.0}}$	48.6 ${}_{\pm\textbf{0.9}}$	22.1 ${}_{\pm\textbf{4.9}}$	69.7 ${}_{\pm\textbf{1.3}}$	40.1 ${}_{\pm\textbf{6.5}}$	38.1 ${}_{\pm\textbf{0.7}}$	23.7 ${}_{\pm\textbf{3.2}}$	64.5 ${}_{\pm\textbf{0.7}}$	22.9 ${}_{\pm\textbf{0.3}}$
GCN (sequential)	71.8 ${}_{\pm\textbf{0.7}}$	75.5 ${}_{\pm\textbf{1.0}}$	96.1 ${}_{\pm\textbf{0.0}}$	97.1 ${}_{\pm\textbf{0.0}}$	52.8 ${}_{\pm\textbf{0.1}}$	25.0 ${}_{\pm\textbf{1.2}}$	75.6 ${}_{\pm\textbf{0.1}}$	45.3 ${}_{\pm\textbf{0.7}}$	46.3 ${}_{\pm\textbf{2.9}}$	29.0 ${}_{\pm\textbf{3.5}}$	73.6 ${}_{\pm\textbf{1.7}}$	45.0 ${}_{\pm\textbf{3.9}}$
+CHASE	83.2 ${}_{\pm\textbf{0.6}}$	85.9 ${}_{\pm\textbf{0.5}}$	96.2 ${}_{\pm\textbf{0.1}}$	97.2 ${}_{\pm\textbf{0.1}}$	69.6 ${}_{\pm\textbf{0.6}}$	60.1 ${}_{\pm\textbf{3.8}}$	87.7 ${}_{\pm\textbf{1.2}}$	76.9 ${}_{\pm\textbf{2.5}}$	80.0 ${}_{\pm\textbf{0.8}}$	78.6 ${}_{\pm\textbf{0.9}}$	93.8 ${}_{\pm\textbf{0.3}}$	89.5 ${}_{\pm\textbf{0.5}}$
EERM (collapsed)	73.5 ${}_{\pm\textbf{0.4}}$	77.4 ${}_{\pm\textbf{0.5}}$	96.2 ${}_{\pm\textbf{0.0}}$	97.3 ${}_{\pm\textbf{0.0}}$	47.5 ${}_{\pm\textbf{2.4}}$	20.0 ${}_{\pm\textbf{6.5}}$	69.6 ${}_{\pm\textbf{1.4}}$	44.5 ${}_{\pm\textbf{8.5}}$	39.2 ${}_{\pm\textbf{2.3}}$	22.4 ${}_{\pm\textbf{0.2}}$	64.7 ${}_{\pm\textbf{0.3}}$	25.6 ${}_{\pm\textbf{2.4}}$
EERM (seq)	73.6 ${}_{\pm\textbf{0.5}}$	77.3 ${}_{\pm\textbf{0.7}}$	96.3 ${}_{\pm\textbf{0.0}}$	97.3 ${}_{\pm\textbf{0.0}}$	52.5 ${}_{\pm\textbf{2.7}}$	23.4 ${}_{\pm\textbf{9.4}}$	76.1 ${}_{\pm\textbf{4.9}}$	47.1 ${}_{\pm\textbf{3.0}}$	44.0 ${}_{\pm\textbf{3.4}}$	24.6 ${}_{\pm\textbf{6.8}}$	73.8 ${}_{\pm\textbf{2.5}}$	45.4 ${}_{\pm\textbf{8.0}}$

Appendix

Table 5: Full results with standard deviations (5 seeds) corresponding to Table 2 . All values in percentage points.

Table 16

Figure 11: Relative improvement in shifter F1 (+CHASE over baseline), stratified by number of label shifts during test snapshots. Averaged across all seven dynamic backbones. CHASE’s benefit increases with shift frequency across all datasets.

Figure 12: Shifter and stable F1 across test snapshots for baseline, baseline with retraining, and baseline + CHASE. CHASE outperforms both alternatives consistently, despite the retraining baseline having access to ground-truth labels from previous test snapshots. Averaged across all backbones.

Figure 13: Hyperparameter sensitivity on GCN-GRU. Shifter accuracy and F1 remains stable across all tested values of

\lambda_{\mathrm{hom}}

\lambda_{\mathrm{change}}

, and

m

. Dashed lines indicate default values.

	shifter (node 46)	stable (node 688)
	label: $0\rightarrow 1$	label: $0\rightarrow 0$
$\mathrm{self\_drift}$	91.65	6.78
$\mathrm{nbr\_drift}$	121.67	5.70
$h_{\mathrm{prev}}$	0.76	0.77
$h_{\mathrm{curr}}$	0.05	0.80
$p_{\mathrm{change}}$	0.83	0.04

Appendix

Table 8: Comparison of a shifter vs. a stable node.

	DBLP-S			GossipCop			Music
$\hat{\mathbf{y}}^{(t{-}1)}$	$\Delta$ F1	$h_{\mathrm{proxy}}$	$p_{\mathrm{change}}$	$\Delta$ F1	$h_{\mathrm{proxy}}$	$p_{\mathrm{change}}$	$\Delta$ F1	$h_{\mathrm{proxy}}$	$p_{\mathrm{change}}$
Correct	+27.7	0.707	0.750	+40.0	0.910	0.909	+30.9	0.442	0.664
Incorrect	-4.6	0.589	0.562	-8.0	0.711	0.696	+22.5	0.472	0.643

Appendix

Table 9: Effect of incorrect

\hat{\mathbf{y}}^{(t{-}1)}

on shifters: change in F1 (

\Delta

F1, in percentage points) from baseline, mean

h_{\mathrm{proxy}}

, and mean

p_{\mathrm{change}}

. Averaged across backbones.

	DBLP-S				GossipCop				Music
	$\mathrm{self\_drift}$		$\mathrm{nbr\_drift}$		$\mathrm{self\_drift}$		$\mathrm{nbr\_drift}$		$\mathrm{self\_drift}$		$\mathrm{nbr\_drift}$
Noise	Sh.	St.	Sh.	St.	Sh.	St.	Sh.	St.	Sh.	St.	Sh.	St.
Feature	10.61	8.47	3.86	3.22	9.42	4.85	1.35	0.16	178.05	149.63	42.32	21.47
Edge	9.25	5.97	4.39	4.01	8.06	0.57	3.33	2.49	118.19	66.68	38.89	31.46

Appendix

Table 10: Mean drift signal values for shifters (Sh.) and stable nodes (St.) at noise level

\alpha=0.5

. Separation persists under both perturbation types.

Figure 14: Shifter F1 under increasing feature noise (Top) and edge noise (Bottom) on GCN-GRU. CHASE ’s degradation is minimal compared to the baseline across all noise levels and datasets.

	DBLP-S		GossipCop		Music
Backbone	$\Delta$ s/ep	$\Delta$ MB	$\Delta$ s/ep	$\Delta$ MB	$\Delta$ s/ep	$\Delta$ MB
GCN-GRU	+0.24	+25.6	+0.35	+6.8	+0.24	+2.2
EvolveGCN	+0.40	+42.5	+0.50	+12.7	+0.22	+7.4
SiGNN	+0.25	+137.9	+0.37	+10.5	+0.24	+3.1
GCN-SE	+0.30	+42.9	+0.38	+18.4	+0.21	+3.1
DySAT	+0.24	+184.9	+0.43	+85.2	+0.19	+8.8

Appendix

Table 11: CHASE overhead: additional wall-clock time (seconds per epoch) and peak GPU memory (MB) over the baseline.

Dynamic Graph Learning

Dynamic graph learning models how networks and their node features change over time to predict links, events, labels, and system behavior.

See all 124 papers

May 15, 2026

Attention Dispersion in Dynamic Graph Transformers: Diagnosis and a Transferable Fix

Transformer-based architectures have become the dominant paradigm for Continuous-Time Dynamic Graph (CTDG) learning, yet their performance remains limited on temporally shifted datasets. In this work, we identify attention dispersion as a shared failure mode of dynamic graph Transformers under temporal distribution shift. Through controlled ablation contrasting structurally and temporally distinguished historical neighbors against random ones, we show that prediction depends on a class of critical nodes that carry consistently more predictive signal than arbitrary neighbors. However, existing Transformers fail to focus on these nodes even when they are present in the input, as temporal shift weakens attention contrast and produces overly dispersed attention distributions. This diagnosis suggests a simple and transferable fix: replace standard attention with differential attention, which suppresses common-mode attention and amplifies distinctive token-level signals. When added to three representative CTDG Transformer baselines, differential attention consistently improves performance, with gains concentrated on high-shift datasets. Attention-level measurements further confirm the mechanism, showing reduced attention entropy and increased attention mass on critical nodes. Building on these findings, we introduce DiffDyG, a reference implementation combining differential attention with standard input encodings. Across 9 benchmarks and three negative sampling protocols, DiffDyG achieves SOTA performance, with especially large gains on the most shifted datasets.

Jinhao Zhang, Kangfei Zhao, Qiuhao Zeng +1

Beijing Institute of Technology · University of Toronto · Hong Kong Baptist University

Distribution Shift Robustness Temporal GNNs Dynamic Graph Learning Attention Mechanisms Differential Attention Temporal Distribution Shift

Ask PDF

Jun 22, 2026

PromptDyG: Test-Time Prompt Adaptation on Dynamic Graphs

Activities in numerous evolving systems can be represented as dynamic graphs in snapshot form at different time intervals, i.e., discrete-time dynamic graphs (DTDGs). Existing methods show impressive advances in capturing historical temporal evolution patterns in DTDGs, but they focus on addressing an offline learning setting, where models are trained using historical snapshots once and then evaluated to all subsequent graph snapshots without further updating. This fails to capture 1) the nature of evolving complexities across graph snapshots and 2) the distribution shift in the testing graph snapshots. To address these problems, we propose PromptDyG, a novel framework that leverages unsupervised test-time Prompt adaptation for Dynamic Graph learning under a live-update online setting. The key insight is that an expressive dynamic graph prompt can be learned on a frozen backbone via minimization of feature-wise, label-free entropy to efficiently and continuously model the evolving patterns. We show theoretically that this unsupervised prompt adaptation can guarantee a larger similarity margin between positive and negative pairs, facilitating more accurate dynamic predictions. It is further confirmed by our extensive empirical results on six benchmark datasets that show consistent and significant improvements of PromptDyG over state-of-the-art baselines.

Guoguo Ai, Chaoxi Niu, Hui Yan +3

School of Computer Science and Engineering, Nanjing University of Science and Technology, China · School of Computing and Information Systems, Singapore Management University, Singapore · Faculty of Data Science, City University of Macau, Macau, China +3

Continual Test-Time Adaptation Dynamic Graph Learning Test-Time Prompt Tuning

Ask PDF

Mar 2, 2026

Towards OOD Generalization in Dynamic Graphs via Causal Invariant Learning

Although dynamic graph neural networks (DyGNNs) have demonstrated promising capabilities, most existing methods ignore out-of-distribution (OOD) shifts that commonly exist in dynamic graphs. Dynamic graph OOD generalization is non-trivial due to the following challenges: 1) Identifying invariant and variant patterns amid complex graph evolution, 2) Capturing the intrinsic evolution rationale from these patterns, and 3) Ensuring model generalization across diverse OOD shifts despite limited data distribution observations. Although several attempts have been made to tackle these challenges, none has successfully addressed all three simultaneously, and they face various limitations in complex OOD scenarios. To solve these issues, we propose a Dynamic graph Causal Invariant Learning (DyCIL) model for OOD generalization via exploiting invariant spatio-temporal patterns from a causal view. Specifically, we first develop a dynamic causal subgraph generator to identify causal dynamic subgraphs explicitly. Next, we design a causal-aware spatio-temporal attention module to extract the intrinsic evolution rationale behind invariant patterns. Finally, we further introduce an adaptive environment generator to capture the underlying dynamics of distributional shifts. Extensive experiments on both real-world and synthetic dynamic graph datasets demonstrate the superiority of our model over state-of-the-art baselines in handling OOD shifts.

Xinxun Zhang, Pengfei Jiao, Mengzhou Gao +2

School of Cyberspace, Hangzhou Dianzi University, China · State Key Laboratory of AI Safety, Beijing · School of Computer Science, Wuhan University, China +1

OOD Generalization Dynamic Graph Learning

Ask PDF

Abstract

Dynamic Graph Learning

Explore similar work

Attention Dispersion in Dynamic Graph Transformers: Diagnosis and a Transferable Fix

PromptDyG: Test-Time Prompt Adaptation on Dynamic Graphs

Towards OOD Generalization in Dynamic Graphs via Causal Invariant Learning