Pith. sign in

REVIEW 4 major objections 3 minor 62 references

Learning Marked Temporal Point Process Explanations based on Counterfactual and Factual Reasoning

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A rational explanation of a marked temporal point process forecast is the minimal event subset that reproduces the full-history prediction almost as well as the full history and beats its complement; the paper's CFF method finds these subse

desk verdict Interesting but under-specified objective for MTPP explanations; the garbled artifact makes this version unverifiable. read the letter →

arxiv 2508.11943 v1 pith:L7TI2ADH submitted 2025-08-16 cs.LG

Sishun Liu , Ke Deng , Xiuzhen Zhang , Yan Wang This is my paper · ORCID
classification cs.LG
keywords markedtemporalpointprocessexplainabilitycounterfactualreasoningfactualeventsequencesminimalexplanationpost-hocneural
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 studies explanations for neural marked temporal point process (MTPP) models, which predict the next event's time and type from a history of stamped events. Its central claim is that a rational explanation of such a prediction is the minimum subset of historical events that, when the model is run on it, reproduces the full-history prediction to a great extent and performs better than the subset's complement does. The authors argue that framing explanation as purely counterfactual or purely factual leads to irrational subsets, and that the right formulation is a joint counterfactual-and-factual objective. They propose CFF, a method that optimizes this objective with a series of deliberate techniques, and report experiments showing it produces correct explanations faster than baselines. If correct, this gives MTPP users a compact, checkable answer to 'which past events made the forecast?'.

What carries the argument

The load-bearing object is the joint counterfactual–factual scoring objective over subsets of the event history. For a candidate subset $S$, the objective awards fidelity when the MTPP's prediction given $S$ is close to its prediction given the full history, awards contrast when the prediction given the complement of $S$ is worse, and penalizes subset size; the explanation is the subset that optimizes this trade-off. CFF is the optimizer for that objective, built from a series of techniques designed to keep the subset search tractable and accurate in practice.

What would settle it

Run CFF on a synthetic Hawkes process in which exactly one past event is designated as the trigger of the next event and other history events are irrelevant; if CFF's minimal subset ever omits the designated trigger, the fidelity-based rationality criterion has failed. Equivalently, on a degenerate MTPP whose output is constant, the trivial one-event subset matches the full-history prediction, so a rationality standard rooted only in accuracy would accept it.

Watch

Extended reading notes

Core claim

The paper's central discovery is that Explanation for MTPP becomes well-defined only when it is treated as a conjunction, not a disjunction. A candidate subset $S$ is explanatory precisely when the model conditioned on $S$ yields the same prediction as the model conditioned on the full history (factual adequacy), while the model conditioned on the complement of $S$ fails to match the full-history prediction (counterfactual necessity), with $S$ minimal under that joint criterion. Pure counterfactual explanations—subsets whose removal changes the prediction—can be irrational because they can contain events that are incidental to the process; pure factual explanations—subsets that alone reprodu

Load-bearing premise

The load-bearing premise is that 'rational' can be measured by output fidelity: a subset that nearly reproduces the model's full-history prediction and beats its own complement is taken to be an explanation.

Editorial extensions

If this is right

  • Users of MTPP models can inspect a compact set of past events rather than the whole history when deciding whether to trust a forecast.
  • The counterfactual-plus-factual definition rules out some degenerate explanations that pure counterfactual or pure factual definitions accept, such as incidental or redundant subsets.
  • Because CFF is post-hoc and targets prediction fidelity, it can be attached to an already-trained neural MTPP without retraining the underlying model.
  • The reported speed advantage makes the explanation method practical for interactive or high-throughput settings.
  • The experiments indicate the method generalizes across multiple MTPP architectures and event-sequence datasets.

Reading between the lines

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

  • The fidelity standard measures explanation against the model's own output, not against the true generative process; a model that learns spurious correlations can receive 'rational' explanations that omit the real causes. A synthetic-data test with a known generative trigger could separate these.
  • The minimal-subset objective can be viewed as a rate-distortion problem over histories, suggesting that the paper's 'minimal and rational' trade-off could be characterized by a Pareto curve of subset size versus fidelity tolerance, a curve the paper does not draw.
  • The same joint counterfactual–factual subset criterion transfers naturally to any autoregressive sequence predictor, such as language models or session recommenders, where a 'history' is a token or item sequence.
  • Stable, tiny explanation sets across many queries could double as a debugging signal: when the selected events are semantically unrelated to the domain, the model is likely relying on shortcuts.
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

4 major / 3 minor

Summary. The paper proposes CFF (Counterfactual and Factual Explainer) for explaining predictions of marked temporal point process (MTPP) models. It defines a 'minimal and rational explanation' as the smallest event subset whose predictive accuracy matches the full-history accuracy 'to a great extent' and beats the complement, then combines counterfactual and factual reasoning to find such subsets. The authors claim experiments demonstrate CFF's superiority over baselines in explanation quality and efficiency. However, the submitted manuscript text is heavily garbled and contains a mismatched arXiv identifier stamp, making much of the formalization, algorithm description, and experimental evidence unverifiable. As presented, the central objective is under-specified and the evaluation appears circular.

Significance. The problem is timely: explaining predictions of MTPP models is important for high-stakes event-sequence applications, and the idea of combining counterfactual and factual reasoning is worth exploring. If CFF were precisely defined, with a non-circular evaluation and a clean, verifiable experimental section, it could make a useful contribution. As submitted, however, the contribution is not assessable: no formal definition of 'rational explanation' is provided with the required tolerance, no formal definitions of the counterfactual/factual components are given, and the reported superiority appears to be fitted to the paper's own objective. No reproducible code, machine-checked proofs, or parameter-free derivations are available.

major comments (4)
  1. [Abstract / §1 (Problem Definition)] The central definition is not well-posed. The phrase 'prediction accuracy of MTPP matches that based on full history to a great extent' lacks any tolerance, distance measure, or threshold. It is also not specified whether 'minimum subset' means minimum cardinality or inclusion-minimality, and no argument establishes existence or uniqueness of such a subset. Since CFF is claimed to optimize exactly this criterion, the optimization objective cannot be evaluated from the manuscript. Please supply a formal definition (e.g., a tolerance ε, a metric, and a precise notion of minimality) and prove that the minimum exists.
  2. [Abstract / §1 (Counterfactual vs. Factual Explanations)] The paper asserts that defining explanation purely as counterfactual or purely as factual can yield irrational explanations, and that combining the two solves the problem. However, the manuscript never formally defines what a counterfactual explanation or a factual explanation is for an MTPP, nor does it specify how CFF combines them. Without these definitions, the claimed superiority of the combined approach is not checkable. Formal definitions and a concrete example showing the failure of each single view are needed.
  3. [§5 (Evaluation Metrics)] The explanation-quality evaluation appears to reuse the definitional criterion: a subset is considered a good explanation if its prediction accuracy is close to the full-history accuracy and better than the complement. CFF is designed to optimize exactly this criterion, so comparing it with baselines on this metric is circular: it measures how well CFF optimizes its own objective, not whether the explanations are actually rational or useful. Add a ground-truth-based or human-grounded evaluation, or at least a metric independent of the optimized objective, before claiming superiority.
  4. [§5 (Experiments and Body Text)] The experimental section, and the manuscript as a whole, is not verifiable. The body contains the mismatched stamp 'arXiv:2508.11944v1' inside a paper submitted as arXiv:2508.11943. Most prose is garbled, equations are partially illegible, and the tables appear to contain placeholder or repeated entries rather than numerical results. No dataset description, baseline details, or statistical significance tests can be discerned. As submitted, the claims of correctness, quality, and processing efficiency cannot be checked. A clean, correctly labeled manuscript with readable tables is an absolute prerequisite.
minor comments (3)
  1. [Throughout] The text is severely corrupted (encoding issues, unreadable headers, and placeholder glyphs). This must be fixed before any further review.
  2. [Notation] The acronym CFF is introduced but the expansion is not clearly defined; equations lack consistent numbering, making references impossible.
  3. [Related Work / References] The related-work discussion and reference list are garbled and incomplete; several citations are unreadable and no complete bibliography is visible.

Circularity Check

1 steps flagged · score 5.0 of 10

The 'rational explanation' standard is defined by fidelity to the model's own full-history prediction, and CFF's correctness is assessed against that same standard, making the central explanatory claim self-definitional.

  1. self definitional [Abstract, first paragraph (definition of Explanation for MTPP) and final paragraph (correctness of CFF)]
    "This study focuses on Explanation for MTPP, aiming to identify the minimal and rational explanation, that is, the minimum subset of events in history, based on which the prediction accuracy of MTPP matches that based on full history to a great extent and better than that based on the complement of the subset. ... Experiments demonstrate the correctness and superiority of CFF over baselines regarding explanation quality."

    The paper defines rational explanation as a subset whose prediction accuracy matches the full-history accuracy and beats the complement. CFF is introduced to 'solve Explanation for MTPP', i.e., to optimize exactly this subset-fidelity objective. If 'explanation quality' in the experiments is measured by the same match-to-full-history criterion, then CFF's correctness is a consequence of its optimization target rather than independent evidence that the found subsets are rational. The normative term 'rational' is thus loaded into the definition; no external ground-truth causal or human-judgment validation is described in the abstract.

full rationale

The paper's central claim is that CFF finds minimal and rational event subsets. That claim is anchored in a definition in which rationality is equated with prediction-accuracy matching the full history and beating the complement. Because CFF is designed to optimize precisely this criterion, the experimental 'correctness' result is not an independent check of rationality; it restates the objective. The comparison against baselines is not circular—it can show that CFF optimizes the stated objective better than alternatives—but it cannot validate the choice of the objective as a theory of rational explanation. The abstract supplies no tolerance for 'to a great extent' and no causal ground truth, so the 'rational' conclusion is definitional. The garbled body text, including a stray arXiv:2508.11944 header, prevents verification of whether independent validation exists, but the abstract's definitional structure is independently load-bearing. No self-citation chain is present. Score 5 reflects a partial circularity: the optimization and baseline comparison have independent content, but the headline normative claim reduces to the paper's own definition.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on three unverified premises: (1) that prediction fidelity to the full model output is a valid measure of explanation rationality, which is the definitional core of the paper; (2) that the black-box MTPP model can be meaningfully queried on arbitrary subsets and complements of the history; and (3) that the combinatorial subset search is tractable and near-optimal as the abstract's 'deliberately designed techniques' claim. Additionally, the abstract's definition requires a tolerance for 'matches to a great extent' that is not stated, leaving the objective under-specified. These premises are not established in the readable part of the manuscript.

free parameters (2)
  • fidelity tolerance for 'matches to a great extent'
    The definition of a rational explanation requires a numeric tolerance for how closely subset-based prediction accuracy must match full-history accuracy. The abstract does not state this tolerance, and the behavior of the objective depends on it.
  • CFF hyperparameters of the 'deliberately designed techniques'
    The abstract refers to 'a series of deliberately designed techniques' for efficiency and correctness without naming them; any such technique introduces tuning parameters (regularization, search beam size, etc.) that are undisclosed at the abstract level.
assumptions (3)
  • domain assumption Prediction fidelity to the full history is a valid proxy for explanation rationality.
    The definition of Explanation for MTPP equates a rational explanation with a subset whose prediction quality matches the full-history prediction. This is asserted in the abstract's first paragraph and is the load-bearing premise for the entire objective.
  • domain assumption The black-box MTPP model can be evaluated on arbitrary sub-histories and complements.
    CFF must score subsets by querying the model with partial event histories and with the remaining events. Many sequence models are not designed to consume arbitrary deletions of a factual history, so this input-manipulation capability is assumed without discussion in the abstract.
  • ad hoc to paper Subset-selection search (greedy or relaxed) finds the minimum subset sufficiently well.
    Exact subset selection over an event history is combinatorial. The abstract claims efficiency via 'deliberately designed techniques' but asserts, rather than proves, that the search returns the minimal rational subset.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Marked Temporal Point Process Explanations based on Counterfactual and Factual Reasoning." pith.science (2026). https://pith.science/paper/L7TI2ADH

@misc{pith2026250811943,
  author       = {Pith},
  title        = {Pith review of: Learning Marked Temporal Point Process Explanations based on Counterfactual and Factual Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L7TI2ADH}},
  note         = {Machine review of arXiv:2508.11943}
}
read the original abstract

Neural network-based Marked Temporal Point Process (MTPP) models have been widely adopted to model event sequences in high-stakes applications, raising concerns about the trustworthiness of outputs from these models. This study focuses on Explanation for MTPP, aiming to identify the minimal and rational explanation, that is, the minimum subset of events in history, based on which the prediction accuracy of MTPP matches that based on full history to a great extent and better than that based on the complement of the subset. This study finds that directly defining Explanation for MTPP as counterfactual explanation or factual explanation can result in irrational explanations. To address this issue, we define Explanation for MTPP as a combination of counterfactual explanation and factual explanation. This study proposes Counterfactual and Factual Explainer for MTPP (CFF) to solve Explanation for MTPP with a series of deliberately designed techniques. Experiments demonstrate the correctness and superiority of CFF over baselines regarding explanation quality and processing efficiency.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 59 canonical work pages

  1. [1]

    Abrate and F

    C. Abrate and F. Bonchi. Counterfactual Graphs for Explainable Classification of Brain Networks . In KDD , 2021

    work page 2021

  2. [2]

    Aimeur, S

    E. Aimeur, S. Amri, and G. Brassard. Fake news, disinformation and misinformation in social media: a review. Social Network Analysis and Mining, 2023

    work page 2023

  3. [3]

    Baribi-Bartov, B

    S. Baribi-Bartov, B. Swire-Thompson, and N. Grinberg. Supersharers of fake news on twitter. Science, 2024

    work page 2024

  4. [4]

    Barkan, V

    O. Barkan, V. Bogina, L. Gurevitch, Y. Asher, and N. Koenigstein. A counterfactual framework for learning and evaluating explanations for recommender systems. In WWW , 2024

    work page 2024

  5. [5]

    Bengio, N

    Y. Bengio, N. Leonard, and A. Courville. Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation . In arXiv:1308.3432, 2013

    arXiv 2013

  6. [6]

    Budhathoki, D

    K. Budhathoki, D. Janzing, P. Bloebaum, and H. Ng. Why did the distribution change? In AISTATS, 2021

    work page 2021

  7. [7]

    R. Cai, Y. Zhu, X. Chen, Y. Fang, M. Wu, J. Qiao, and Z. Hao. On the probability of necessity and sufficiency of explaining graph neural networks: A lower bound optimization approach. Neural Networks, 2025

    work page 2025

  8. [8]

    Z. Chen, F. Silvestri, J. Wang, Y. Zhang, Z. Huang, H. Ahn, and G. Tolomei. GREASE : Generate Factual and Counterfactual Explanations for GNN -based Recommendations . In arXiv:2208.04222, 2022

    arXiv 2022

Show all 62 references
  1. [9]

    D. J. Daley and D. Vere-Jones. An Introduction to the Theory of Point Processes Volume I : Elementary Theory and Methods . Springer, 2003

    2003

  2. [10]

    Enguehard, D

    J. Enguehard, D. Busbridge, A. Bozson, C. Woodcock, and N. Hammerla. Neural Temporal Point Processes for Modelling Electronic Health Records . In The Machine Learning for Health NeurIPS Workshop , 2020

    2020

  3. [11]

    Fernandez, J

    G. Fernandez, J. A. Aledo, J. A. Gamez, and J. M. Puerta. Factual and counterfactual explanations in fuzzy classification trees. IEEE Transactions on Fuzzy Systems , 2022

    2022

  4. [12]

    P. J. Ferraro. Counterfactual Thinking and Impact Evaluation in Environmental Policy . New directions for evaluation, 2009

    2009

  5. [13]

    T. Gao, D. Subramanian, D. Bhattacharjya, X. Shou, N. Mattei, and K. Bennett. Causal Inference for Event Pairs in Multivariate Point Processes . In NeurIPS, 2021

    2021

  6. [14]

    H. G. Gauch. Scientific method in practice. Cambridge University Press, 2003

    2003

  7. [15]

    Y. Ge, S. Liu, Z. Fu, J. Tan, Z. Li, S. Xu, Y. Li, Y. Xian, and Y. Zhang. A survey on trustworthy recommender systems. ACM Trans. Recomm. Syst. , 2024

    2024

  8. [16]

    Ghazimatin, O

    A. Ghazimatin, O. Balalau, R. S. Roy, and G. Weikum. PRINCE: Provider -side Interpretability with Counterfactual Explanations in Recommender Systems . In WSDM , 2020

    2020

  9. [17]

    Goyal, Z

    Y. Goyal, Z. Wu, J. Ernst, D. Batra, D. Parikh, and S. Lee. Counterfactual Visual Explanations . In ICML, 2019

    2019

  10. [18]

    Guidotti

    R. Guidotti. Counterfactual Explanations and How to Find Them : Literature Review and Benchmarking . DMKD, 2022

    2022

  11. [19]

    Hizli, S

    C . Hizli, S. John, A. Juuti, T. Saarinen, K. Pietil \" a inen, and P. Marttinen. Temporal Causal Mediation through a Point Process : Direct and Indirect Effects of Healthcare Interventions . In NeurIPS, 2023

    2023

  12. [20]

    Huang and K

    J. Huang and K. C.-C. Chang. Towards reasoning in large language models: A survey. In ACL 2023 , 2023

    2023

  13. [21]

    T. Ide, G. Kollias, D. T. Phan, and N. Abe. Cardinality-regularized hawkes-granger model. In NeurIPS, 2021

    2021

  14. [22]

    Karlsson, J

    I. Karlsson, J. Rebane, P. Papapetrou, and A. Gionis. Explainable time series tweaking via irreversible and reversible temporal transformations. In ICDM , 2018

    2018

  15. [23]

    R. M. Karp. Reducibility among Combinatorial Problems . In Complexity of Computer Computations . 1972

    1972

  16. [24]

    M. T. Lash, Q. Lin, N. Street, J. G. Robinson, and J. Ohlmann. Generalized Inverse Classification . In SIAM, 2017

    2017

  17. [25]

    T. Lei, R. Barzilay, and T. Jaakkola. Rationalizing Neural Predictions . In EMNLP, 2016

    2016

  18. [26]

    Leskovec and A

    J. Leskovec and A. Krevl. SNAP Datasets : Stanford Large Network Dataset Collection . http://snap.stanford.edu/data, 2014

    2014

  19. [27]

    S. Li, M. Feng, L. Wang, A. Essofi, Y. Cao, J. Yan, and L. Song. Explaining point processes by learning interpretable temporal logic rules. In ICLR, 2022

    2022

  20. [28]

    Y. Li, H. Chen, S. Xu, Y. Ge, J. Tan, S. Liu, and Y. Zhang. Fairness in recommendation: Foundations, methods, and applications. ACM Trans. Intell. Syst. Technol. , 2023

    2023

  21. [29]

    W. Lin, H. Lan, and B. Li. Generative causal explanations for graph neural networks. In ICML, 2021

    2021

  22. [30]

    Y. Liu, C. Chen, Y. Liu, X. Zhang, and S. Xie. Multi-objective explanations of GNN predictions. In ICDM, 2021

    2021

  23. [31]

    C. J. Maddison, A. Mnih, and Y. W. Teh. The Concrete Distribution : A Continuous Relaxation of Discrete Random Variables . In ICLR , 2017

    2017

  24. [32]

    Martens and F

    D. Martens and F. Provost. Explaining Data -driven Document Classifications . MIS Q., 2014

    2014

  25. [33]

    H. Mei, G. Qin, and J. Eisner. Imputing missing events in continuous-time event streams. In ICML, 2019

    2019

  26. [34]

    R. K. Mothilal, A. Sharma, and C. Tan. Explaining Machine Learning Classifiers through Diverse Counterfactual Explanations . In FAT , 2020

    2020

  27. [35]

    S. Mu, Y. Li, W. X. Zhao, J. Wang, B. Ding, and J.-R. Wen. Alleviating Spurious Correlations in Knowledge -aware Recommendations through Counterfactual Generator . In SIGIR , 2022

    2022

  28. [36]

    Noorbakhsh and M

    K. Noorbakhsh and M. G. Rodriguez. Counterfactual Temporal Point Processes . In NeurIPS, 2022

    2022

  29. [37]

    T. Omi, N. Ueda, and K. Aihara. Fully Neural Network based Model for General Temporal Point Processes . In NeurIPS, 2019

    2019

  30. [38]

    A. Panos. Decomposable transformer point processes. In NeurIPS, 2024

    2024

  31. [39]

    Parmentier and T

    A. Parmentier and T. Vidal. Optimal Counterfactual Explanations in Tree Ensembles . In ICML , 2021

    2021

  32. [40]

    J. Pearl. Causality. Cambridge University Press, 2009

    2009

  33. [41]

    M. A. Prado-Romero, B. Prenkaj, G. Stilo, and F. Giannotti. A survey on graph counterfactual explanations: Definitions, methods, evaluation, and research challenges. ACM Comput. Surv. , 2024

    2024

  34. [42]

    Prosperi, Y

    M. Prosperi, Y. Guo, M. Sperrin, J. S. Koopman, J. S. Min, X. He, S. Rich, M. Wang, I. E. Buchan, and J. Bian. Causal Inference and Counterfactual Prediction in Machine Learning for Actionable Healthcare . Nature Machine Intelligence, 2 0 (7): 0 369--375, 2020

    2020

  35. [43]

    Ramakrishnan, Y

    G. Ramakrishnan, Y. C. Lee, and A. Albarghouthi. Synthesizing Action Sequences for Modifying Model Decisions . In AAAI , 2020

    2020

  36. [44]

    Sahoh and A

    B. Sahoh and A. Choksuriwong. The Role of Explainable Artificial Intelligence in High -stakes Decision -making Systems : a Systematic Review . J. Ambient Intell. Humanized Comput., 2022

    2022

  37. [45]

    Schulam and S

    P. Schulam and S. Saria. Reliable Decision Support using Counterfactual Models . NeurIPS, 2017

    2017

  38. [46]

    Shchur, A

    O. Shchur, A. C. Türkmen, T. Januschowski, and S. Günnemann. Neural temporal point processes: A review. In IJCAI, 2021

    2021

  39. [47]

    Z. Song, C. Yang, C. Wang, B. An, and S. Li. Latent logic tree extraction for event sequence explanation from LLMs . In arXiv :2406.01124 , 2024

    2024 arXiv

  40. [48]

    J. Tan, S. Xu, Y. Ge, Y. Li, X. Chen, and Y. Zhang. Counterfactual Explainable Recommendation . In CIKM, 2021

    2021

  41. [49]

    J. Tan, S. Geng, Z. Fu, Y. Ge, S. Xu, Y. Li, and Y. Zhang. Learning and Evaluating Graph Neural Network Explanations based on Counterfactual and Factual Reasoning . In WWW , 2022

    2022

  42. [50]

    K. H. Tran, A. Ghazimatin, and R. Saha Roy. Counterfactual Explanations for Neural Recommenders . In SIGIR , 2021

    2021

  43. [51]

    Verma, V

    S. Verma, V. Boonsanong, M. Hoang, K. Hines, J. Dickerson, and C. Shah. Counterfactual Explanations and Algorithmic Recourses for Machine Learning : A Review . ACM Computing Surveys, 2020

    2020

  44. [52]

    D. Wu, T. Ide, G. Kollias, J. Navratil, A. Lozano, N. Abe, Y. Ma, and R. Yu. Learning granger causality from instance-wise self-attentive hawkes processes. In AISTATS, 2024

    2024

  45. [53]

    R. Xu, Y. Yu, C. Zhang, M. K. Ali, J. C. Ho, and C. Yang. Counterfactual and Factual Reasoning over Hypergraphs for Interpretable Clinical Predictions on EHR . In MLHS, 2022

    2022

  46. [54]

    Z. Xu, H. Lamba, Q. Ai, J. Tetreault, and A. Jaimes. CFE 2: Counterfactual editing for search result explanation. In SIGIR, 2024

    2024

  47. [55]

    Y. Yang, C. Yang, B. Li, Y. Fu, and S. Li. Neuro-symbolic temporal point processes, 2024

    2024

  48. [56]

    Zhang, A

    Q. Zhang, A. Lipani, \" O . Kirnap, and E. Yilmaz. Self-attentive Hawkes Process . In ICML , 2020

    2020

  49. [57]

    Zhang, A

    Q. Zhang, A. Lipani, and E. Yilmaz. Learning neural point processes with latent graphs. In WWW, 2021 a

    2021

  50. [58]

    Zhang, J

    S. Zhang, J. Zhang, X. Song, S. Adeshina, D. Zheng, C. Faloutsos, and Y. Sun. PaGE - Link : Path -based Graph Neural Network Explanation for Heterogeneous Link Prediction . In WWW , 2023

    2023

  51. [59]

    Zhang, K

    Y. Zhang, K. Sharma, and Y. Liu. VigDet : Knowledge Informed Neural Temporal Point Process for Coordination Detection on Social Media . In NeurIPS, 2021 b

    2021

  52. [60]

    Zhang, D

    Y. Zhang, D. Cao, and Y. Liu. Counterfactual Neural Temporal Point Process for Estimating Causal Influence of Misinformation on Social Media . In NeurIPS, 2022

    2022

  53. [61]

    H. Zhao, H. Chen, F. Yang, N. Liu, H. Deng, H. Cai, S. Wang, D. Yin, and M. Du. Explainability for large language models: A survey. ACM Trans. Intell. Syst. Technol. , 2024

    2024

  54. [62]

    Q. Zhao, M. A. Erdogdu, H. Y. He, A. Rajaraman, and J. Leskovec. SEISMIC : A Self - Exciting Point Process Model for Predicting Tweet Popularity . In SIGKDD , 2015

    2015

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.