Dual Spatial-Temporal Attribution: Architecture-Aligned Post-Hoc Explainability for Recurrent Graph Anomaly Detection

Hamida Seba; Iyad Assaad Nekka; Karima Amrouche; Khaled Walid Hidouci

REVIEW 4 major objections 5 minor 14 references

Dual Spatial-Temporal Attribution: Architecture-Aligned Post-Hoc Explainability for Recurrent Graph Anomaly Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read X-AddGraph gives AddGraph's anomaly alarms a three-part post-hoc explanation while preserving every detection score exactly.

desk verdict A clean, architecture-aligned post-hoc explainer for AddGraph with a true-by-construction preservation guarantee, but the headline long-term attribution result is partly self-confirming and needs a sharper test. read the letter →

arxiv 2608.12441 v1 pith:GERQLBTL submitted 2026-08-12 cs.LG cs.AI

Iyad Assaad Nekka , Hamida Seba , Khaled Walid Hidouci , Karima Amrouche This is my paper · ORCID

classification cs.LGcs.AI

keywords dynamicgraphanomalydetectionpost-hocexplainabilitydualspatial-temporalattributiongradientbackpropagationthroughtimeattentionnetworkrecurrentneuralnetworksAdd

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that AddGraph, a widely used GCN+GRU anomaly detector for dynamic graphs, can be explained after the fact without any change to its behavior. The authors build X-AddGraph, whose Dual Spatial-Temporal Attribution (DSTA) mechanism produces, for each flagged edge, a spatial explanation (which neighboring nodes drove the score), a short-term temporal explanation (which recent snapshot within the attention window mattered), and a long-term temporal explanation (which historical snapshot before the window contributed through the recurrent memory). They report that the detector's AUC is preserved exactly (ΔAUC = 0 to ten decimal places) and that the long-term attribution picks historical snapshots whose removal changes the score substantially more than a random snapshot (0.127 vs. 0.074 counterfactual divergence). If correct, this gives deployment environments a way to audit alarm decisions without retraining or altering an already validated detector.

What carries the argument

The Dual Spatial-Temporal Attribution (DSTA) mechanism is the framework's core object: for each flagged edge it outputs the triplet (top-k neighbor set N*, most suspicious window step s*, most responsible pre-window lag k*), read as the structural, short-term, and long-term causes. It is carried by three aligned attribution formulas: Eq. (3), an input-weighted gradient $\phi^\mathrm{sp}_{uv} = |A^t_{uv} \cdot \partial f/\partial A^t_{uv}|$ for the spatial part; Eq. (4), the endpoint-averaged CAB attention $\phi^\mathrm{sh}_s = \tfrac{1}{2}(a^{t*}_{i*}[s] + a^{t*}_{j*}[s])$ for the short-term part; and Eq. (5), the normalized Frobenius norm of the BPTT gradient $\phi^\mathrm{lo}_k = g_k / \sum_{k'} g_{k'}$ with $g_k = \|\partial f/\partial H^{t*-\omega-k}\|_F$ for the long-term part. The key machinery is the coupling of these three formulas to the frozen detector's own computation: no retraining, no modified weights, and the short-term component is literally read from the cached attention distribution.

What would settle it

Inject anomalies into a synthetic dynamic graph with known causal lag, such as an edge that becomes anomalous because of a specific snapshot five steps earlier, then compare the lag picked by maximizing Eq. (5) with the true lag across many trials; if the match rate does not exceed random, the long-term explanation is not causally faithful.

Watch

Extended reading notes

Core claim

The central claim is that X-AddGraph is the first post-hoc explainability framework for AddGraph and the GCN+GRU paradigm, and that it achieves exact preservation of detection behavior while adding explanations that no spatially-blind method can provide. Each of its three attribution components is aligned with one architectural module: a gradient-times-input relevance map over the current adjacency for the spatial module, a direct reading of the contextual attention weights already computed during inference for the short-term module (at zero extra cost), and a backpropagation-through-time gradient rollback through the GRU hidden states for the long-term module. The paper argues this decomposition is necessary because AddGraph's score couples a graph attention convolution, a contextual attention block, and a GRU in a single pipeline. The empirical support is the exact score reproduction and the counterfactual temporal tests: the identified historical lag carries a mean counterfactual divergence of 0.127 versus 0.074 for a random lag, and confident true positives show 0.332 versus 0.125.

Load-bearing premise

The load-bearing premise is that counterfactual score divergence is a faithful measure of how much a historical snapshot actually caused the anomaly score; since the long-term component selects snapshots by gradient magnitude and then evaluates them by score divergence, the test could partly reward the selection criterion rather than reveal true causal influence.

Editorial extensions

If this is right

Deployed AddGraph detectors can be retrofitted with explanations at zero behavioral risk: every anomaly score and every alarm is identical before and after explanation, since the explainer never modifies the frozen model.
Analysts get a concrete answer to three questions per alarm: which neighbors, which recent step, and which historical snapshot drove the score, enabling audit and triage of individual flags.
The long-term component fills a gap that static explainers cannot fill: they have no mechanism for ranking historical snapshots, so on temporal questions they are reduced to random selection (0.074), while the rollback reaches 0.127 counterfactual divergence.
The architecture-aligned design transfers to the wider GCN+GRU family: any model whose convolution consumes the previous hidden state admits the same spatial and long-term components, and any model exposing an internal attention distribution can reuse the short-term reading.
The zero-cost short-term component shows that faithful explanations can be free when a model already computes a normalized attention distribution during inference.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural extension the paper leaves implicit is that the same architecture-aligned recipe, matching each attribution component to one module and exploiting cached endogenous signals, should apply to other recurrent graph architectures whose scoring fuses structural and temporal streams; a testable prediction is that any detector exposing an attention-like distribution will admit a near-free and f
The temporal fidelity test could be sharpened into a direct causal benchmark: if datasets with annotated culprit snapshots existed, one could compare the identified lag against ground truth rather than against random selection; the paper itself notes such benchmarks do not yet exist, and synthetic dynamic graphs with injected anomalies of known lag would supply one.
From the reported per-population numbers, a reasonable inference is that confident true positives concentrate their historical signal strongly (0.332 versus 0.125), so in practice the long-term explanation is most useful for high-scoring alarms, while near-threshold and false-positive alarms have more distributed temporal causes; one could test whether thresholding explanations by alarm confidence
The sparsity pattern, rising from 0.000 for confident true positives to 0.750 for random edges, suggests that confidently flagged anomalies are structurally isolated and their entire local neighborhood is explanation-relevant; this could be turned into a confidence heuristic that flags maximally non-sparse explanations as probable alarms even before consulting the detector score.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague — this one is worth a look if you care about explainability for dynamic graph detectors, but the headline temporal result needs a grain of salt.

What's genuinely new: X-AddGraph is, as far as I can tell, the first post-hoc explainability framework for AddGraph, the 2019 GCN+GRU model that still defines a lot of the edge-level anomaly detection space. The design is clean: one attribution component per architectural module—input-weighted gradients for the spatial GAT, a direct read of the CAB attention weights for the short-term window (free, since the weights are already computed), and BPTT gradient rollback for the long-term hidden state. The decomposition is natural, and the paper gives a plausible argument for why a generic static explainer would miss the temporal components. They also release code, which is good.

The strictly post-hoc guarantee is true by construction: the detector is frozen, so performance is preserved. The 'Delta AUC = 0 verified to ten decimal places' is a bit of theater, but harmless. On the UCI Message benchmark, their trained AddGraph reaches 0.8705 AUC, which is better than the original 0.8083—that's a useful datapoint.

Now the soft spot. The central empirical claim for long-term attribution is that the identified historical snapshot carries more counterfactual signal than a random one (0.127 vs. 0.074, Table 3). The problem is that the selection criterion in Eq. (5) is the Frobenius norm of the BPTT gradient—a local sensitivity measure—and the evaluation then measures the counterfactual score divergence of that same chosen snapshot against a random pick. That's partly circular: to first order, a perturbation's effect scales with exactly that gradient norm, so an argmax will beat a random draw even if the snapshot isn't the true causal driver. The authors themselves concede in Section 7.3 that the fidelity metrics are only counterfactual proxies. To make the result convincing, they'd need to compare against per-lag counterfactual effects across all lags, or at least report the full distribution. Also, the evaluation is thin: a single dataset, ten edges per population, two seeds, and no variance reported in Table 3. The AUC mismatch between Table 1 (0.8705) and Section 6.1 (0.8491) is explained but looks sloppy at first glance.

That said, the method itself is sound as a post-hoc explanation toolbox. The qualitative walkthrough gives some independent plausibility, and the limitations are honestly stated. This is a conditional accept—the long-term attribution result needs a sharper test, but the rest of the framework stands.

Who benefits: anyone working on explainability for temporal graphs, or deploying AddGraph-family models in regulated settings. It's not a landmark, but it's a legitimate contribution. A serious referee should engage with it, provided they push on the circularity issue.

Recommendation: send to peer review; expect a revision that either fixes the temporal validation or softens the claim.

Referee Report

4 major / 5 minor

Summary. The manuscript presents X-AddGraph, a post-hoc explainability framework for the AddGraph dynamic graph anomaly detector. The framework decomposes an explanation for a flagged edge into three components aligned with AddGraph's modules: a gradient-based spatial attribution over the current adjacency, a direct reading of the contextual attention weights for short-term temporal attribution, and a backpropagation-through-time gradient rollback through the GRU hidden states for long-term temporal attribution. The central claims are that X-AddGraph is the first post-hoc explainer for AddGraph and the GCN+GRU paradigm, that it preserves detection performance exactly because the detector is frozen, and that the long-term attribution identifies historical snapshots carrying more counterfactual signal than random selection (0.127 vs. 0.074). The evaluation is conducted on the UCI Message benchmark with four edge populations, comparing against a flat-gradient baseline, and includes a qualitative walkthrough and a public code release.

Significance. If the claims hold, the paper makes a useful engineering contribution: it offers a way to explain a deployed AddGraph detector without retraining or modification, including a zero-cost short-term temporal signal and a long-term temporal capability that spatially-blind explainers cannot provide. The code release and the explicit architectural alignment are strengths, and the post-hoc property is genuinely valuable for operational deployment. However, the headline long-term attribution result is supported by a partly self-confirming evaluation, so the current evidence does not yet establish that the identified historical snapshot is the true causal driver. With additional controls and more careful statistical reporting, the framework could be a solid contribution to the explainability literature for dynamic graph anomaly detection.

major comments (4)

[Section 5.3 / Table 3 / Eq. (5)] The long-term temporal fidelity test is partly self-confirming. The selection rule in Eq. (5) ranks lags by the Frobenius norm of the BPTT gradient, and the evaluation in Section 5.3 then measures counterfactual score divergence at the argmax lag versus a random lag. To first order, any perturbation of the hidden state changes the score by an amount proportional to that same gradient norm, so the selected lag will mechanically show a larger divergence than a random lag even if the gradient ranking carries no causal information. Please add a control that removes this selection bias, for example a rank correlation between gradient norms and counterfactual divergences over all K lags, a comparison against the second-best lag, or a comparison against the maximum over a random set of lags. Without such a control, the headline advantage 0.127 vs. 0.074 is not evidence of explanatory faithfulness beyond the already-known fact that large local gradients produce larger local effects.
[Section 5.2 / Table 3] The paper promises mean±standard deviation over two seeds in Section 5.2, but Table 3 reports only means labeled 'mean over seeds' with no standard deviations and no significance test, and each edge population contains only 10 edges. The 0.127 vs. 0.074 difference, and especially the confident-TP comparison 0.332 vs. 0.125, should be accompanied by standard deviations and a paired significance test across edges or seeds. As reported, the '73% relative advantage' is not statistically grounded.
[Section 5.3] The counterfactual operation used in the long-term temporal test is not defined precisely enough for reproducibility. The text says that the divergence between the counterfactual score at the identified lag and at a random lag is measured, but it does not state how the hidden state at a lag is perturbed, how many random lags are sampled per edge, or whether the same random lag is reused across seeds. Please specify the exact protocol, including the perturbation construction and the random-lag sampling procedure.
[Section 7.3 / Section 6.3] Section 7.3 concedes that the fidelity metrics are only counterfactual proxies and that ground-truth causal benchmarks do not yet exist for this task. Given that concession, the results section, particularly the 'decisive comparison' language in Section 6.3, should be reworded to present the long-term result as a proxy comparison with known selection bias rather than as identification of the true historical cause. This is not a request to remove the experiment, but to align the claims with the acknowledged evidential status.

minor comments (5)

[Section 6.1 / Table 1] The statement that Delta AUC = 0 is 'verified empirically to ten decimal places' is presented as an empirical result, but because the detector is frozen by construction, the equality holds a priori; the verification is a sanity check of the implementation, and the claim should be phrased accordingly.
[Section 7.2] The generalization claims to StrGNN and EvolveGCN are not tested in the paper and should be framed as hypotheses or future work rather than as established properties of DSTA.
[Section 5.2 / Section 6] The evaluation uses a single benchmark dataset; while Section 7.3 acknowledges this, the discussion in Sections 6 and 7.2 should consistently avoid language that implies general effectiveness across datasets.
[Table 2] Fidelity+ values slightly above 1 are interpreted as evidence that removed edges were suppressing the anomaly signal, but no analysis is given to distinguish this from numerical noise; a sentence describing the typical magnitude of numerical error would strengthen this interpretation.
[Eq. (1)] The notation in Eq. (1) uses r and Q without defining their dimensions; please define all variables used in the attention formula.

Circularity Check

1 steps flagged · score 5.0 of 10

Long-term temporal fidelity advantage is partly self-confirming: argmax gradient-norm selection makes 0.127 vs. 0.074 largely a selection artifact.

fitted input called prediction [Section 4.4 (Eq. 5), Section 5.3, Table 3]
"We attribute the score to each pre-window snapshot by the Frobenius norm of the gradient obtained through backpropagation through time ... The most influential historical snapshot is k∗ = arg max_k φ_lo_k. ... we compare the counterfactual effect of the identified window step / historical snapshot against that of a randomly selected one."

The identified lag is selected as the argmax of the gradient norm in Eq. (5), and the validation in Table 3 measures the counterfactual score divergence at that same lag versus a random lag. To first order, the counterfactual divergence of a perturbation to H_{t*−ω−k} is proportional to the same Frobenius norm of ∂f/∂H_{t*−ω−k} used for selection, so choosing the maximum mechanically yields a larger divergence than a random draw. The reported 0.127 vs. 0.074 therefore partly reflects max-vs-random selection, not independent evidence that the identified snapshot is the true historical cause. The paper itself concedes in Section 7.3 that the fidelity metrics are only counterfactual proxies, not ground-truth causal evaluation.

full rationale

The paper is not broadly circular: the post-hoc preservation claim is honestly stated as a by-construction guarantee, the spatial attribution follows external DGExplainer rules, the short-term component is a direct reading of the model's own attention weights, and no load-bearing self-citation chain appears. The one significant circular step is confined to the long-term temporal fidelity evaluation: Eq. (5) selects k* by maximizing a local gradient-norm sensitivity proxy, and the validation then measures a counterfactual effect that is, to first order, the same gradient quantity. Thus the headline long-term advantage is partly a selection artifact rather than an independent causal confirmation. The score of 5 reflects partial circularity in one central empirical claim while the framework's other components retain independent content.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The paper contributes a method rather than a derivation with fitted constants. The hand-chosen values are K and top-k, plus inherited AddGraph hyperparameters from the original protocol. The load-bearing assumptions concern gradient and attention fidelity as proxies for explanatory importance, and the paper itself flags most of these as proxies or limitations.

free parameters (2)

Lookback horizon K = 5
K=5 in Eq. 5 bounds the long-term gradient rollback. No sensitivity analysis is reported, and the headline long-term result (0.127 vs 0.074) could depend on this hand-chosen value.
Top-k for structural explanation = 5 (implied by 'top-5' in Section 6.2)
The structural explanation returns the top-k nodes, and the experiments use k=5 without a sensitivity study. This choice affects the reported Fidelity and Sparsity values.

assumptions (5)

domain assumption The AddGraph score function is differentiable with respect to adjacency entries and recurrent hidden states, so the gradient-based attribution terms are well-defined.
Equations 3 and 5 compute partial derivatives of the score; this requires differentiability, which holds for GAT and GRU with smooth activations but is not stated or proven in the paper.
ad hoc to paper The CAB attention weights computed during inference are a faithful measure of short-term temporal importance.
Section 4.3 reads cached attention directly as the short-term explanation, asserting it is 'exactly faithful' to the model's internal weighting, without empirical validation. Section 7.3 concedes the attention is nearly uniform with the detector's window size of 2.
domain assumption Backpropagation through time gradient norms through the GRU are a valid proxy for historical influence on the anomaly score.
Equation 5 uses the Frobenius norm of the gradient to rank historical snapshots. This assumes gradients reflect counterfactual importance, a standard but non-trivial assumption in attribution literature.
domain assumption Counterfactual score divergence is a valid fidelity metric for explanations.
Section 5.3 defines temporal fidelity via counterfactual score changes, and Section 7.3 admits these are proxies, not ground-truth causal evaluation.
domain assumption UCI Message with the original anomaly injection protocol is representative enough to support the paper's generalization claims.
The evaluation uses a single benchmark. Section 7.3 acknowledges that extending to Digg and heterogeneous settings is planned, so generalization is assumed rather than demonstrated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Dual Spatial-Temporal Attribution: Architecture-Aligned Post-Hoc Explainability for Recurrent Graph Anomaly Detection." pith.science (2026). https://pith.science/paper/GERQLBTL

@misc{pith2026260812441,
  author       = {Pith},
  title        = {Pith review of: Dual Spatial-Temporal Attribution: Architecture-Aligned Post-Hoc Explainability for Recurrent Graph Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GERQLBTL}},
  note         = {Machine review of arXiv:2608.12441}
}

read the original abstract

Deep learning detectors for anomalies in dynamic graphs have reached strong accuracy, yet they remain opaque: when an edge is flagged, the analyst receives a score but no reason. This opacity is untenable in the cooperative, regulated information systems where such detectors are deployed, where automated decisions must be auditable and trustworthy. We address this gap for AddGraph, the foundational GCN+GRU framework for edge-level anomaly detection in dynamic graphs, which to our knowledge has never been equipped with any form of explainability. We present a strictly post-hoc explainability framework, X-AddGraph, built on a Dual Spatial-Temporal Attribution (DSTA) mechanism whose three components are each aligned with one of AddGraph's architectural modules: a gradient-based relevance attribution over the current adjacency structure (spatial), a direct reading of the contextual attention weights already computed during inference (short-term temporal, at zero additional cost), and a gradient rollback through the recurrent hidden states (long-term temporal). Because the detector is frozen, detection performance is preserved exactly (Delta AUC = 0, verified empirically to ten decimal places). On the UCI Message benchmark, our trained AddGraph baseline reaches an average per-snapshot AUC of 0.8705, exceeding the originally published result; X-AddGraph reproduces every score identically while adding explanations where none existed. Evaluated across four edge populations - confident true positives, low-confidence true positives, false positives, and random samples - the long-term attribution identifies historical snapshots carrying significantly more counterfactual signal than random selection (0.127 vs. 0.074), a capability that no spatially-blind explainer can provide. We release our implementation for full reproducibility.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 13 canonical work pages

[1]

In: Proc

Zheng, L., Li, Z., Li, J., Li, Z., Gao, J.: AddGraph: Anomaly detection in dynamic graph using attention-based temporal GCN. In: Proc. 28th International Joint Con- ference on Artificial Intelligence (IJCAI), pp. 4419–4425 (2019) 12 I. A. Nekka et al

work page 2019
[2]

3747–3756 (2021)

Cai, L., Chen, Z., Luo, C., Gui, J., Ni, J., Li, D., Chen, H.: Structural temporal graphneuralnetworksforanomalydetectionindynamicgraphs.In:Proc.30thACM International Conference on Information and Knowledge Management (CIKM), pp. 3747–3756 (2021)

work page 2021
[3]

In: Proc

Pareja, A., Domeniconi, G., Chen, J., Ma, T., Suzumura, T., Kanezashi, H., Kaler, T., Schardl, T.B., Leiserson, C.E.: EvolveGCN: Evolving graph convolutional net- works for dynamic graphs. In: Proc. 34th AAAI Conference on Artificial Intelligence, pp. 5363–5370 (2020)

work page 2020
[4]

IEEE Transactions on Knowledge and Data Engineering35(12), 12081–12094 (2023)

Liu, Y., Pan, S., Wang, Y.G., Xiong, F., Wang, L., Chen, Q., Lee, V.C.S.: Anomaly detection in dynamic graphs via transformer. IEEE Transactions on Knowledge and Data Engineering35(12), 12081–12094 (2023)

work page 2023
[5]

ACM Transactions on Knowledge Discovery from Data18(8), 1–44 (2024)

Ekle, O.A., Eberle, W.: Anomaly detection in dynamic graphs: A comprehensive survey. ACM Transactions on Knowledge Discovery from Data18(8), 1–44 (2024)

work page 2024
[6]

In: Advances in Neural Information Processing Systems 32 (NeurIPS) (2019)

Ying, Z., Bourgeois, D., You, J., Zitnik, M., Leskovec, J.: GNNExplainer: Gener- ating explanations for graph neural networks. In: Advances in Neural Information Processing Systems 32 (NeurIPS) (2019)

work page 2019
[7]

In: Advances in Neural Information Processing Systems 33 (NeurIPS) (2020)

Luo, D., Cheng, W., Xu, D., Yu, W., Zong, B., Chen, H., Zhang, X.: Parameterized explainer for graph neural network. In: Advances in Neural Information Processing Systems 33 (NeurIPS) (2020)

work page 2020
[8]

In: Proc

Duval, A., Malliaros, F.D.: GraphSVX: Shapley value explanations for graph neu- ral networks. In: Proc. European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), pp. 302–318 (2021)

work page 2021

Show all 14 references

[9]

In: Proc

Pope, P.E., Kolouri, S., Rostami, M., Martin, C.E., Hoffmann, H.: Explainability methods for graph convolutional neural networks. In: Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10772–10781 (2019)

2019
[10]

In: Proc

Xie, J., Liu, Y., Shen, Y.: Explaining dynamic graph neural networks via rele- vance back-propagation. In: Proc. 34th International Joint Conference on Artificial Intelligence (IJCAI) (2025)

2025
[11]

In: Proc

Xia, W., Lai, M., Shan, C., Zhang, Y., Dai, X., Li, X., Li, D.: Explaining temporal graph models through an explorer-navigator framework. In: Proc. 11th International Conference on Learning Representations (ICLR) (2023)

2023
[12]

arXiv preprint arXiv:2310.16375 (2023)

Wang, T., Lin, X., Liu, S., et al.: DyExplainer: Explainable dynamic graph neural networks with sparse attentions. arXiv preprint arXiv:2310.16375 (2023)

2023 arXiv
[13]

IEEE Transactions on Knowledge and Data Engineering (2026)

Han, X., Hu, X., Ma, C., Shang, X.: Explainable anomaly detection in dynamic heterogeneous graphs via relation evolution. IEEE Transactions on Knowledge and Data Engineering (2026)

2026
[14]

Social Networks31(2), 155–163 (2009)

Opsahl, T., Panzarasa, P.: Clustering in weighted networks. Social Networks31(2), 155–163 (2009)

2009

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [1]

In: Proc

Zheng, L., Li, Z., Li, J., Li, Z., Gao, J.: AddGraph: Anomaly detection in dynamic graph using attention-based temporal GCN. In: Proc. 28th International Joint Con- ference on Artificial Intelligence (IJCAI), pp. 4419–4425 (2019) 12 I. A. Nekka et al

work page 2019

[2] [2]

3747–3756 (2021)

Cai, L., Chen, Z., Luo, C., Gui, J., Ni, J., Li, D., Chen, H.: Structural temporal graphneuralnetworksforanomalydetectionindynamicgraphs.In:Proc.30thACM International Conference on Information and Knowledge Management (CIKM), pp. 3747–3756 (2021)

work page 2021

[3] [3]

In: Proc

Pareja, A., Domeniconi, G., Chen, J., Ma, T., Suzumura, T., Kanezashi, H., Kaler, T., Schardl, T.B., Leiserson, C.E.: EvolveGCN: Evolving graph convolutional net- works for dynamic graphs. In: Proc. 34th AAAI Conference on Artificial Intelligence, pp. 5363–5370 (2020)

work page 2020

[4] [4]

IEEE Transactions on Knowledge and Data Engineering35(12), 12081–12094 (2023)

Liu, Y., Pan, S., Wang, Y.G., Xiong, F., Wang, L., Chen, Q., Lee, V.C.S.: Anomaly detection in dynamic graphs via transformer. IEEE Transactions on Knowledge and Data Engineering35(12), 12081–12094 (2023)

work page 2023

[5] [5]

ACM Transactions on Knowledge Discovery from Data18(8), 1–44 (2024)

Ekle, O.A., Eberle, W.: Anomaly detection in dynamic graphs: A comprehensive survey. ACM Transactions on Knowledge Discovery from Data18(8), 1–44 (2024)

work page 2024

[6] [6]

In: Advances in Neural Information Processing Systems 32 (NeurIPS) (2019)

Ying, Z., Bourgeois, D., You, J., Zitnik, M., Leskovec, J.: GNNExplainer: Gener- ating explanations for graph neural networks. In: Advances in Neural Information Processing Systems 32 (NeurIPS) (2019)

work page 2019

[7] [7]

In: Advances in Neural Information Processing Systems 33 (NeurIPS) (2020)

Luo, D., Cheng, W., Xu, D., Yu, W., Zong, B., Chen, H., Zhang, X.: Parameterized explainer for graph neural network. In: Advances in Neural Information Processing Systems 33 (NeurIPS) (2020)

work page 2020

[8] [8]

In: Proc

Duval, A., Malliaros, F.D.: GraphSVX: Shapley value explanations for graph neu- ral networks. In: Proc. European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), pp. 302–318 (2021)

work page 2021