Pith. sign in

REVIEW 2 major objections 5 minor 42 references

Explaining Temporal Graph Neural Networks via Feature-induced Information Flow

T0 review · 2 major / 5 minor · reviewed 2026-07-15 · grok-4.5

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

Pith's one-line read Event relevance scores that follow every pathway an event can influence an ETGNN prediction, including the latent memory updates that carry long-range dependencies.

desk verdict Solid, usable XAI extension for ETGNNs: modular NRM + full event-flow relevance that actually recovers long-range chains the baselines miss. read the letter →

arxiv 2606.27201 v2 pith:KE5OCKBA submitted 2026-06-25 cs.LG

Ping Xiong , Thomas Schnake , Klaus-Robert Müller , Shinichi Nakajima This is my paper
classification cs.LG
keywords explainableAIlayer-wiserelevancepropagationnormalizedmeasuretemporalgraphneuralnetworkseventhigher-orderattribution
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

Event-based temporal graph neural networks learn from streams of time-stamped interactions, yet existing explainers only inspect the final embedding and decoding stages. They therefore miss the pathways that run through event-induced messages and memory updates, which are precisely the routes that carry long-range temporal influence. This paper defines Event Relevance as the total relevance of every walk that either arrives at an event feature or passes through the messages that event produces. The definition is obtained by extending the Normalized Relevance Measure framework with a modular decomposition that lets relevance be specified hierarchically for the complex recurrent architecture. On synthetic infection-chain and attacker-motif graphs with known ground truth, and on a real political-event network, the resulting scores recover the true causal chains more accurately than prior methods and support joint relevance of multiple events.

What carries the argument

Event Relevance (ER) inside the Normalized Relevance Measure framework, obtained by modular decomposition of the ETGNN into Event-Processing, Embedding and Decoding modules so that walk relevance can be defined and propagated through every event-associated variable.

What would settle it

On a synthetic infection graph whose single ground-truth infection chain is known, compute top-k Event Relevance; if that chain is systematically ranked below the chains recovered by a baseline that ignores memory updates, the central claim fails.

Watch

Extended reading notes

Core claim

The contribution of an event to an ETGNN prediction is the relevance of the union of all walks that touch either the event feature itself or any of the messages the event induces during memory evolution; this quantity, called Event Relevance, is strictly more faithful than attributions that look only at the final embedding stage.

Load-bearing premise

That the chosen layer-wise relevance-propagation rules, once the network is rewritten as a proper feed-forward graph, correctly measure how much each event actually drives the model’s output.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper proposes Event Relevance (ER), an LRP-style attribution method for Event-based Temporal Graph Neural Networks that attributes predictions to events by tracing the full event-induced information flow, including pathways through event-induced messages that update node memories in the Event Processing module. Building on the Normalized Relevance Measure (NRM) framework, the authors introduce a modular decomposition procedure (series/parallel connections, hierarchical layer indexing, local relevance) that makes NRM tractable for complex ETGNN architectures. They define ER-feat, ER-msg, ER-Emb, full ER, and joint multi-event relevance, and evaluate on synthetic infection-chain and attacker-motif datasets with ground-truth explanations plus the real-world ICEWS18 political event network. Qualitative heatmaps and quantitative metrics (Recall-chain, Precision/Recall, pruning/activation) show that ER and ER-msg substantially outperform TGNNExplainer, Occlusion, Grad imes Input, and the Emb-only baseline ER-Emb, especially for long-range temporal dependencies.

Significance. If the results hold, the work is a clear advance for XAI on temporal graphs: existing ETGNN explainers largely ignore the EP/memory pathway that carries long-range event interactions, and the paper supplies both a principled NRM-based definition that includes those pathways and synthetic benchmarks with verifiable ground-truth chains/motifs. The modular NRM extension is reusable beyond ETGNNs (illustrated on LSTM), and joint multi-event relevance is a natural higher-order capability that is demonstrated on infection chains. Strengths include explicit relevance factorizations in the appendices, multiple complementary evaluation metrics, and consistent qualitative/quantitative gains over strong baselines. The main free parameters (LRP-γ, ε) are conventional rather than circular.

major comments (2)
  1. [Section 5, Appendix C] Section 5 and Appendix C: the central claim that higher ER scores identify the events the model actually uses rests on the faithfulness of the chosen LRP rules (LRP-γ with γ=0.1 for linear layers; LRP-all for element-wise multiplications in GRU/attention) after modular conversion to a proper FFNN. This is a standard LRP assumption, but the manuscript does not report sensitivity to γ (or alternative rules such as LRP-0/ε) nor any sanity check that the modular factorization preserves conservation on the full ETGNN. A short ablation or conservation diagnostic would make the load-bearing faithfulness claim more robust.
  2. [Table 1, Appendix F] Table 1 and Appendix F: quantitative scores are reported as averages over top-k without error bars, confidence intervals, or statistical tests across episodes/graphs. Given that the superiority claim is quantitative (e.g., Recall-chain 0.844 vs ≤0.35 on Infection; large Precision/Recall gaps on Attacker), variability across samples should be shown so that the ranking is not driven by a few easy instances.
minor comments (5)
  1. [Figure 1, Section 4.4] Figure 1 and Section 4.4: the relation ER = ER-msg + ER-Emb (and why ER-feat is weaker) is clear in the text but would benefit from a one-sentence reminder in the figure caption that walks through messages but not the event feature itself are counted only by ER-msg.
  2. [Section 5.3, Table 2] Section 5.3 / Table 2: human-interpretability on ICEWS18 is argued via news-article alignment for a few selected events. A brief note on how the predicted events were chosen (and that Occlusion is infeasible) would reduce selection-bias concerns for readers.
  3. [Sections 3–4] Notation: global vs hierarchical/local layer indices (n^(l), l_msg(e), S^(l)(X)) are carefully defined but dense; a short notation table or consistent reminder when switching hierarchies would help.
  4. [Throughout] Typos / wording: e.g., "genereal architectures" (Sec. 3.1), "straightforwared" (footnote 5), "reall" in Table 1 caption; also arXiv id 2606.27201 appears future-dated relative to typical numbering.
  5. [Section 2.4, Section 5] Related work: TempME is cited but not run as a baseline; a short justification (or a limited comparison) would complete the experimental picture.

Circularity Check

1 steps flagged · score 1.0 of 10

No significant circularity; ER definitions and empirical claims rest on independent ground-truth/perturbation evaluations, with only ordinary self-citation of the authors' concurrent NRM framework.

  1. self citation load bearing [Sec. 2.2, Abstract, Sec. 3 intro, Eq. (21)]
    "Our method is built upon the recent Normalized Relevance Measure (NRM) framework [7], a generalized procedure following the higher-order XAI method GNN-LRP [8] to define the relevance of arbitrary sets of neurons... This framework enables explicit quantification of the entire event-induced information flow, and guarantees comparability of relevance scores across layers."

    NRM [7] is concurrent work by the identical author set (Xiong, Schnake, Müller, Nakajima). The paper treats its signed-measure axioms, marginal/conditional operations, and cross-layer comparability as given premises that underwrite ER. The citation is not machine-checked or externally falsified independently of the present authors; however it is not load-bearing for the empirical superiority claims, which rest on ground-truth chains and pruning curves, so the circularity is minor.

full rationale

The paper is an XAI methods paper whose central claims (ER recovers full event-induced walks including messages, outperforms TGNNExplainer/Occlusion/Grad imes Input on Infection Recall-chain, Attacker precision/recall, and ICEWS pruning) are evaluated against external synthetic ground-truth chains/motifs and model-output perturbations that do not depend on the NRM definitions themselves. The modular series/parallel decomposition (Eqs. 7–9, Sec. 3.2) and hierarchical indexing are constructive extensions that assemble walk relevances once LRP rules are fixed; they do not redefine the target quantities in terms of the outputs. The sole self-citation is to the concurrent NRM framework [7] (same author set), which supplies the signed-measure language and comparability guarantees; this is ordinary framework reuse, not a load-bearing uniqueness theorem or fitted ansatz that forces the reported superiority. γ=0.1 and LRP-all are conventional hyperparameters, not parameters fitted to the evaluation metrics and then re-presented as predictions. No self-definitional loop, fitted-input-as-prediction, or renaming of a known result appears in the derivation chain from proper-FFNN conversion through ER-feat/ER-msg/ER to the tables and figures. Score 1 only for the minor self-citation; the derivation is otherwise self-contained against external benchmarks.

Assumptions & free parameters 2 free parameters · 4 assumptions · 2 invented entities

The central claim rests on the NRM signed-measure axioms, the standard LRP propagation rules, the conversion of ETGNNs into proper FFNNs via duplication layers, and the modeling assumption that event-induced messages are the correct intermediate variables to attribute. No free parameters are fitted to produce the claimed superiority; γ is a conventional hyper-parameter.

free parameters (2)
  • LRP-γ hyper-parameter γ = 0.1
    Set to 0.1 by hand for all linear layers; conventional but not derived.
  • numerical stabilizer ε = 1e-6
    Added to LRP-γ numerator for stability; set to 1e-6.
assumptions (4)
  • domain assumption Normalized Relevance Measure axioms (R(∅)=0, R(W)=1, additivity over disjoint walk sets) and the backward-Markov factorization of walk relevance.
    Taken from the cited NRM framework [7] and used throughout Sections 3–4 to define all ER quantities.
  • domain assumption Any architecture can be virtually converted into a proper FFNN by unfolding, copying intermediate inputs/outputs and inserting duplication layers for skip connections.
    Stated in Section 3.1 citing Xiong et al. [7]; required for modular decomposition of ETGNNs.
  • domain assumption LRP-γ (γ=0.1) for linear layers and LRP-all for element-wise multiplications correctly redistribute relevance.
    Standard LRP practice invoked in Section 5; not re-derived.
  • ad hoc to paper Event-induced messages m(e) are the intermediate variables that mediate long-range temporal dependencies and therefore must be included in event relevance.
    Core modeling choice of the paper (Section 4.4); justified by architecture inspection and empirical gains, but not independently proven.
invented entities (2)
  • Event Relevance (ER) and its variants ER-feat, ER-msg, ER-Emb, joint multi-event ER
    purpose: To quantify the total relevance of all walks associated with one or more events, including pathways through messages.
    Defined in Section 4.4 as the central explanatory quantity; no independent existence outside the NRM construction.
  • Modular decomposition procedure for NRM
    purpose: To define relevance hierarchically for complex multi-module networks such as ETGNNs and LSTMs.
    Introduced in Section 3.2; enables the rest of the paper but is a methodological construct rather than an external entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Explaining Temporal Graph Neural Networks via Feature-induced Information Flow." pith.science (2026). https://pith.science/paper/KE5OCKBA

@misc{pith2026260627201,
  author       = {Pith},
  title        = {Pith review of: Explaining Temporal Graph Neural Networks via Feature-induced Information Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KE5OCKBA}},
  note         = {Machine review of arXiv:2606.27201}
}
read the original abstract

Event-based Temporal Graph Neural Networks (ETGNNs) have demonstrated strong performance across a wide range of applications, including social network analysis, epidemic tracing, recommender systems, and political event forecasting. However, their increasing complexity poses significant challenges for explainability. Existing explanation methods focus only on a subset of the information flow within ETGNNs, typically tracing contributions from the event-related embeddings to the output. Consequently, they overlook the important pathways through event-induced variables, which mediate interactions between nodes and thereby play a central role in capturing long-range temporal dependencies. To overcome this limitation, we propose a novel attribution method that analyzes the entire information flow through all event-associated variables. Our method is built upon the recent Normalized Relevance Measure (NRM) framework, which enables explicit quantification of information flow originating from event embeddings as well as information flow passing through event-induced variables. It also ensures comparability of latent variables across layers, and supports higher-order analysis of interactions between events. To handle the architectural complexity of ETGNNs, we extend the NRM framework with a modular decomposition procedure that facilitates the systematic construction of relevance structure for complex neural architectures. We evaluate our approach on two synthetic datasets for epidemic tracing and social dynamics, as well as a real-world dataset of political event networks. Our qualitative and quantitative experiments show that our method consistently outperforms existing explanation approaches while producing more human-interpretable explanations.

Figures

Figures reproduced from arXiv: 2606.27201 by the authors.

Figure 1
Figure 1. Overview of the proposed Event Relevance (ER). Our ER (bottom row) is defined within the NRM framework, which facilitates tracing information flow [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. A series connection (left) and a parallel connection (right) of modules. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Modularization applied to an LSTM network (a). Its proper FFNN [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Overview of ETGNN: (a) forward computation process, and its (b) [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Illustration of ER-feat for the event e, explaining the node-level prediction of node v. (a) ER-feat (b) ER-msg (c) ER-Emb (d) ER ...... ...... ...... [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Illustration of four ER definitions. (a) ER-feat is the relevance of all [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Joint relevance of events e1 and e2, where e1 is associated with nodes u and w and e2 is with nodes v and w. 5. Empirical Evaluation of Explaining ETGNNs In this section, we conduct qualitative and quantitative experi￾ments with synthetic and real-world temporal graph …
Figure 8
Figure 8. Figure 8: Top-20 most relevant event edges for node prediction for an example episode in the Infection dataset. Nodes are depicted as circles except special ones: [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Marginal and joint event relevances for predicting the infection of [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Class definitions in the Attacker dataset. The number next to each [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Explanation by baseline and our proposed methods. The ground-truth attacker subgraphs are marked with circles, and the numbers on the edges denote the [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 3 linked inside Pith

  1. [1]

    C. Gao, X. Wang, X. He, Y . Li, Graph neural networks for recommender system, in: WSDM, ACM, 2022, pp. 1623–1625

    2022

  2. [2]

    S. Deng, H. Rangwala, Y . Ning, Learning dynamic context graphs for predicting social events, in: KDD, ACM, 2019, pp. 1007–1016

    2019

  3. [3]

    Cencetti, G

    G. Cencetti, G. Santin, A. Longa, E. Pigani, A. Barrat, C. Cattuto, S. Lehmann, M. Salathé, B. Lepri, Digital prox- imity tracing on empirical contact networks for pandemic control, Nat. Commun. 12 (1) (2021) 1655

    2021

  4. [4]

    L. Zhao, Y . Song, C. Zhang, Y . Liu, P. Wang, T. Lin, M. Deng, H. Li, T-GCN: A temporal graph convolutional network for traffic prediction, IEEE Trans. Intell. Transp. Syst. 21 (9) (2020) 3848–3858

    2020

  5. [5]

    W. Xia, M. Lai, C. Shan, Y . Zhang, X. Dai, X. Li, D. Li, Explaining temporal graph models through an explorer- navigator framework, in: ICLR, OpenReview.net, 2023

    2023

  6. [6]

    J. Chen, R. Ying, Tempme: Towards the explainability of temporal graph neural networks via motif discovery, in: NeurIPS, 2023

    2023

  7. [7]

    Xiong, T

    P. Xiong, T. Schnake, G. Montavon, K.-R. Müller, S. Naka- jima, Normalized relevance measure as a unifying frame- work to explain neural network latent structures, CoRR abs/2606.00557 (2026)

    arXiv 2026

  8. [8]

    Schnake, O

    T. Schnake, O. Eberle, J. Lederer, S. Nakajima, K. T. Schütt, K.-R. Müller, G. Montavon, Higher-order explana- tions of graph neural networks via relevant walks, IEEE Trans. Pattern Anal. Mach. Intell. 44 (11) (2022) 7581– 7596

    2022

Show all 42 references
  1. [9]

    S. Bach, A. Binder, G. Montavon, F. Klauschen, K.-R. Müller, W. Samek, On pixel-wise explanations for non- linear classifier decisions by layer-wise relevance propaga- tion, PloS one 10 (7) (2015) e0130140

    2015

  2. [10]

    Achtibat, S

    R. Achtibat, S. M. V . Hatefi, M. Dreyer, A. Jain, T. Wie- gand, S. Lapuschkin, W. Samek, Attnlrp: Attention-aware layer-wise relevance propagation for transformers, in: ICML, V ol. 235 of Proceedings of Machine Learning Re- search, PMLR/OpenReview.net, 2024, pp. 135–168

    2024

  3. [11]

    Kauffmann, J

    J. Kauffmann, J. Dippel, L. Ruff, W. Samek, K.-R. Müller, G. Montavon, Explainable ai reveals clever hans effects in unsupervised learning models, Nature Machine Intelli- gence 7 (3) (2025) 412–422

    2025

  4. [12]

    what is relevant in a text document?

    L. Arras, F. Horn, G. Montavon, K.-R. Müller, W. Samek, "what is relevant in a text document?": An interpretable machine learning approach, PLOS ONE 12 (8) (2017) 1– 23

    2017

  5. [13]

    A. Ali, T. Schnake, O. Eberle, G. Montavon, K.-R. Müller, L. Wolf, XAI for transformers: Better explanations through conservative propagation, in: ICML, Proceedings of Ma- chine Learning Research, PMLR, 2022, pp. 435–451

    2022

  6. [14]

    Esders, T

    M. Esders, T. Schnake, J. Lederer, A. Kabylda, G. Mon- tavon, A. Tkatchenko, K.-R. Müller, Analyzing atomic interactions in molecules as learned by neural networks, J. Chem. Theory Comput. 21 (2) (2025) 714–729

    2025

  7. [15]

    Montavon, S

    G. Montavon, S. Lapuschkin, A. Binder, W. Samek, K.-R. Müller, Explaining nonlinear classification decisions with deep taylor decomposition, Pattern Recognit. 65 (2017) 211–222

    2017

  8. [16]

    Samek, G

    W. Samek, G. Montavon, S. Lapuschkin, C. J. Anders, K.- R. Müller, Explaining deep neural networks and beyond: A review of methods and applications, Proc. IEEE 109 (3) (2021) 247–278

    2021

  9. [17]

    Montavon, A

    G. Montavon, A. Binder, S. Lapuschkin, W. Samek, K.-R. Müller, Layer-wise relevance propagation: An overview, in: Explainable AI, Springer, 2019, pp. 193–209

    2019

  10. [18]

    Arras, J

    L. Arras, J. A. Arjona-Medina, M. Widrich, G. Montavon, M. Gillhofer, K.-R. Müller, S. Hochreiter, W. Samek, Ex- plaining and interpreting LSTMs, in: Explainable AI: In- terpreting, Explaining and Visualizing Deep Learning, V ol. 11700, Springer, 2019, pp. 211–238

    2019

  11. [19]

    Xiong, T

    P. Xiong, T. Schnake, G. Montavon, K.-R. Müller, S. Naka- jima, Efficient computation of higher-order subgraph attri- bution via message passing, in: ICML, PMLR, 2022, pp. 24478–24495

    2022

  12. [20]

    J. Chen, T. Ma, C. Xiao, Fastgcn: Fast learning with graph convolutional networks via importance sampling, in: ICLR (Poster), OpenReview.net, 2018

    2018

  13. [21]

    W. L. Hamilton, Z. Ying, J. Leskovec, Inductive repre- sentation learning on large graphs, in: NIPS, 2017, pp. 1024–1034

    2017

  14. [22]

    T. N. Kipf, M. Welling, Semi-supervised classification with graph convolutional networks, in: 5th International Con- ference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017

    2017

  15. [23]

    K. T. Schütt, H. E. Sauceda, P.-J. Kindermans, A. Tkatchenko, K.-R. Müller, Schnet–a deep learning ar- chitecture for molecules and materials, J. Chem. Phys. 148 (24) (2018) 241722

    2018

  16. [24]

    H. Yuan, H. Yu, S. Gui, S. Ji, Explainability in graph neural networks: A taxonomic survey, IEEE Trans. Pattern Anal. Mach. Intell. 45 (5) (2023) 5782–5799

    2023

  17. [25]

    W. Ju, Z. Fang, Y . Gu, Z. Liu, Q. Long, Z. Qiao, Y . Qin, J. Shen, F. Sun, Z. Xiao, J. Yang, J. Yuan, Y . Zhao, Y . Wang, X. Luo, M. Zhang, A comprehensive survey on deep 14 graph representation learning, Neural Networks 173 (2024) 106207

    2024

  18. [26]

    Monti, D

    F. Monti, D. Boscaini, J. Masci, E. Rodolà, J. Svoboda, M. M. Bronstein, Geometric deep learning on graphs and manifolds using mixture model cnns, in: CVPR, IEEE Computer Society, 2017, pp. 5425–5434

    2017

  19. [27]

    Rossi, B

    E. Rossi, B. Chamberlain, F. Frasca, D. Eynard, F. Monti, M. M. Bronstein, Temporal graph networks for deep learn- ing on dynamic graphs, CoRR abs/2006.10637 (2020)

    2006 arXiv

  20. [28]

    Y . Ma, Z. Guo, Z. Ren, J. Tang, D. Yin, Streaming graph neural networks, in: SIGIR, ACM, 2020, pp. 719–728

    2020

  21. [29]

    Kumar, X

    S. Kumar, X. Zhang, J. Leskovec, Predicting dynamic embedding trajectory in temporal interaction networks, in: KDD, ACM, 2019, pp. 1269–1278

    2019

  22. [30]

    D. Xu, C. Ruan, E. Körpeoglu, S. Kumar, K. Achan, Induc- tive representation learning on temporal graphs, in: ICLR, OpenReview.net, 2020

    2020

  23. [31]

    R. S. Trivedi, M. Farajtabar, P. Biswal, H. Zha, Dyrep: Learning representations over dynamic graphs, in: ICLR (Poster), OpenReview.net, 2019

    2019

  24. [32]

    Kocsis, C

    L. Kocsis, C. Szepesvári, Bandit based monte-carlo plan- ning, in: ECML, V ol. 4212 of Lecture Notes in Computer Science, Springer, 2006, pp. 282–293

    2006

  25. [33]

    Tishby, F

    N. Tishby, F. C. N. Pereira, W. Bialek, The informa- tion bottleneck method, CoRR physics/0004057 (2000). arXiv:physics/0004057

    2000 arXiv

  26. [34]

    K. Cho, B. van Merrienboer, Ç. Gülçehre, D. Bahdanau, F. Bougares, H. Schwenk, Y . Bengio, Learning phrase representations using RNN encoder-decoder for statistical machine translation, in: EMNLP, ACL, 2014, pp. 1724– 1734

    2014

  27. [35]

    Shrikumar, P

    A. Shrikumar, P. Greenside, A. Kundaje, Learning impor- tant features through propagating activation differences, in: ICML, PMLR, 2017, pp. 3145–3153

    2017

  28. [36]

    Ancona, E

    M. Ancona, E. Ceolini, C. Öztireli, M. Gross, Towards bet- ter understanding of gradient-based attribution methods for deep neural networks, in: ICLR (Poster), OpenReview.net, 2018

    2018

  29. [37]

    Samek, A

    W. Samek, A. Binder, G. Montavon, S. Lapuschkin, K.-R. Müller, Evaluating the visualization of what a deep neural network has learned, IEEE Trans. Neural Netw. Learn. Syst. 28 (11) (2017) 2660–2673

    2017

  30. [38]

    Blücher, J

    S. Blücher, J. Vielhaben, N. Strodthoff, Preddiff: Explana- tions and interactions from conditional expectations, Artif. Intell. 312 (2022) 103774

    2022

  31. [39]

    W. Jin, M. Qu, X. Jin, X. Ren, Recurrent event network: Autoregressive structure inferenceover temporal knowl- edge graphs, in: EMNLP (1), Association for Computa- tional Linguistics, 2020, pp. 6669–6683

    2020

  32. [40]

    Boschee, J

    E. Boschee, J. Lautenschlager, S. O’Brien, S. Shellman, J. Starz, M. Ward, ICEWS Coded Event Data (2015)

    2015

  33. [41]

    Montavon, Gradient-based vs

    G. Montavon, Gradient-based vs. propagation-based ex- planations: An axiomatic comparison, in: Explainable AI: Interpreting, Explaining and Visualizing Deep Learning, Springer International Publishing, Cham, 2019, pp. 253– 265

    2019

  34. [42]

    Shinzo Abe−Express intent to meet or negotiate→North Korea

    L. Arras, G. Montavon, K.-R. Müller, W. Samek, Explain- ing recurrent neural network predictions in sentiment analy- sis, in: W ASSA@EMNLP, Association for Computational Linguistics, 2017, pp. 159–168. Appendix A. Modularized NRM for LSTM Here we give details of example applic...

    2017

Pith tools

Reviewed July 15, 2026 · model on record in the stance chip above.