pith. sign in

arxiv: 2510.13205 · v3 · submitted 2025-10-15 · 💻 cs.LG · cs.AI

CleverCatch: A Knowledge-Guided Weak Supervision Model for Fraud Detection

Amirhossein Mozafari , Kourosh Hashemi , Erfan Shafagh , Soroush Motamedi , Azar Taheri Tayebi , Mohammad A. Tayebi This is my paper

Pith reviewed 2026-05-18 07:48 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords healthcare fraud detectionweak supervisionknowledge-guided modelanomaly detectionprescription fraudneural embeddingsinterpretabilitydomain expertise
0
0 comments X

The pith

CleverCatch improves fraud detection by integrating expert rules into neural encoders trained jointly on synthetic compliance and violation data.

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

Healthcare fraud detection is hard because of scarce labels, changing tactics, and complex records. The paper presents CleverCatch as a way to use domain knowledge to guide a weak supervision model. It aligns expert rules with data samples in a shared embedding space by training encoders on synthetic examples of both following and breaking the rules. This lets the model learn soft rule embeddings that work on real data while keeping clinical sense. The result is better accuracy and more interpretable detections than pure machine learning or unsupervised methods.

Core claim

By training encoders jointly on synthetic data representing both compliance and violation, CleverCatch learns soft rule embeddings that generalize to complex, real-world datasets. This hybrid design enhances data-driven learning with domain-informed constraints, leading to improved detection accuracy and transparency in healthcare fraud detection.

What carries the argument

Soft rule embeddings that align rules and data samples within a shared embedding space, created by jointly training encoders on synthetic compliance and violation data.

If this is right

  • Outperforms four state-of-the-art anomaly detection baselines with average improvements of 1.3% in AUC and 3.4% in recall.
  • Ablation studies highlight the complementary role of expert rules.
  • Improves both detection accuracy and transparency in the model.
  • Offers an interpretable approach for high-stakes domains such as healthcare.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Approaches like this could extend to other areas with expert knowledge but limited labels, such as detecting financial fraud or insurance claims abuse.
  • The method points to a general way to combine human heuristics with machine learning for better performance in dynamic environments.
  • Testing the model on datasets with newly emerging fraud patterns could show how well the embeddings adapt without full retraining.

Load-bearing premise

Training encoders jointly on synthetic data for compliance and violation enables the soft rule embeddings to generalize to real-world datasets while preserving clinical meaning.

What would settle it

If experiments on additional real-world datasets show no performance gains over baselines or if the learned embeddings fail to maintain alignment with clinical rules, the generalization claim would not hold.

Figures

Figures reproduced from arXiv: 2510.13205 by Amirhossein Mozafari, Azar Taheri Tayebi, Erfan Shafagh, Kourosh Hashemi, Mohammad A. Tayebi, Soroush Motamedi.

Figure 2
Figure 2. Figure 2: (a) Cumulative distribution of drug pair similarities [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Workflow of CLEVERCATCH. The system embeds domain rules and prescription data into a shared latent space via rule and sample encoders. Synthetic compliance/violation samples guide alignment, which generates pseudo-labels to enhance the base anomaly detection model. rules are encoded as ϱ(p, NULL) = MLP([ep; eNULL]), allow￾ing both unary and binary rules to share the same encoding framework. This approach i… view at source ↗
read the original abstract

Healthcare fraud detection remains a critical challenge due to limited availability of labeled data, constantly evolving fraud tactics, and the high dimensionality of medical records. Traditional supervised methods are challenged by extreme label scarcity, while purely unsupervised approaches often fail to capture clinically meaningful anomalies. In this work, we introduce CleverCatch, a knowledge-guided weak supervision model designed to detect fraudulent prescription behaviors with improved accuracy and interpretability. Our approach integrates structured domain expertise into a neural architecture that aligns rules and data samples within a shared embedding space. By training encoders jointly on synthetic data representing both compliance and violation, CleverCatch learns soft rule embeddings that generalize to complex, real-world datasets. This hybrid design enables data-driven learning to be enhanced by domain-informed constraints, bridging the gap between expert heuristics and machine learning. Experiments on the large-scale real-world dataset demonstrate that CleverCatch outperforms four state-of-the-art anomaly detection baselines, yielding average improvements of 1.3\% in AUC and 3.4\% in recall. Our ablation study further highlights the complementary role of expert rules, confirming the adaptability of the framework. The results suggest that embedding expert rules into the learning process not only improves detection accuracy but also increases transparency, offering an interpretable approach for high-stakes domains such as healthcare fraud detection.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces CleverCatch, a knowledge-guided weak supervision model for healthcare fraud detection in prescription data. It embeds domain expertise by aligning expert rules and data samples in a shared embedding space, with encoders trained jointly on synthetic compliance/violation examples to learn soft rule embeddings intended to generalize to real-world datasets. The central claim is that on a large-scale real-world dataset, CleverCatch outperforms four state-of-the-art anomaly detection baselines with average improvements of 1.3% in AUC and 3.4% in recall, supported by an ablation study confirming the complementary role of the rules.

Significance. If the generalization from synthetic training data to real fraud patterns holds with proper controls, the hybrid architecture could meaningfully advance weak supervision techniques for label-scarce, high-dimensional anomaly detection in regulated domains. The soft rule embedding approach and explicit use of domain knowledge for interpretability represent a constructive direction that, if substantiated, would be of interest to the ML-for-healthcare community.

major comments (2)
  1. [§5] §5 (Experiments): The headline result of 1.3% AUC / 3.4% recall lift is reported as an average without per-baseline scores, statistical significance tests, or details on baseline hyperparameter tuning and implementation, making it impossible to isolate the contribution of the knowledge-guided component from possible differences in model capacity or dataset-specific effects.
  2. [§3.2] §3.2 (Synthetic Data Construction) and §4.3 (Ablation): No quantitative characterization of synthetic data coverage (e.g., distribution shift metrics relative to the real prescription dataset) or sensitivity analysis on synthetic fidelity is supplied, which is load-bearing for the claim that joint encoder training produces soft rule embeddings that preserve clinical meaning and transfer to evolving real-world fraud patterns.
minor comments (2)
  1. [§3] The definition and computation of soft rule embeddings should be formalized with explicit equations in the methods section to clarify alignment and training objectives.
  2. [§5] Figure captions and axis labels in the experimental results should include error bars or confidence intervals to aid interpretation of the reported improvements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. The comments highlight important areas for improving the clarity and rigor of our experimental results and synthetic data analysis. We address each major comment point by point below and have made corresponding revisions to the manuscript.

read point-by-point responses

  1. Referee: [§5] §5 (Experiments): The headline result of 1.3% AUC / 3.4% recall lift is reported as an average without per-baseline scores, statistical significance tests, or details on baseline hyperparameter tuning and implementation, making it impossible to isolate the contribution of the knowledge-guided component from possible differences in model capacity or dataset-specific effects.

    Authors: We agree that the current presentation of results as averages limits interpretability. In the revised manuscript, we have added a new table in Section 5 that reports AUC and recall for each of the four individual baselines, along with the per-baseline improvements from CleverCatch. We have also included results from statistical significance tests (paired t-tests across five random seeds) with p-values to establish that the gains are significant. An appendix now details the hyperparameter tuning procedure for all baselines, including search ranges, selection criteria, and final configurations used. These changes allow readers to better evaluate the specific contribution of the knowledge-guided component. revision: yes

  2. Referee: [§3.2] §3.2 (Synthetic Data Construction) and §4.3 (Ablation): No quantitative characterization of synthetic data coverage (e.g., distribution shift metrics relative to the real prescription dataset) or sensitivity analysis on synthetic fidelity is supplied, which is load-bearing for the claim that joint encoder training produces soft rule embeddings that preserve clinical meaning and transfer to evolving real-world fraud patterns.

    Authors: We acknowledge the value of explicitly quantifying the alignment between synthetic and real data. The revised Sections 3.2 and 4.3 now include distribution shift metrics, specifically KL divergence and Wasserstein distance, computed on key feature distributions between the synthetic compliance/violation examples and the real prescription dataset. We have also added a sensitivity analysis that varies synthetic data parameters (such as violation rate and noise injection) and reports the resulting impact on real-data performance and embedding quality. These additions provide quantitative support for the transfer of soft rule embeddings while preserving clinical relevance. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical results on held-out real data anchor performance claims

full rationale

The paper introduces CleverCatch as a hybrid neural architecture that embeds expert rules via joint training on synthetic compliance/violation data, then reports AUC and recall gains on a large-scale real-world prescription dataset against four baselines plus an ablation. No equations, derivations, or first-principles results are presented that reduce the headline metrics to quantities defined by the model's own fitted parameters or synthetic training distribution. The evaluation uses held-out real data as an external benchmark, and the central claim (improved detection via knowledge-guided embeddings) is supported by direct experimental comparison rather than self-definition, renaming, or load-bearing self-citation. The synthetic-to-real generalization step is an empirical assumption, not a circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim depends on the representativeness of synthetic compliance/violation data and the assumption that expert rules can be embedded without distorting clinical semantics; no free parameters or invented physical entities are explicitly introduced.

axioms (2)
  • domain assumption Synthetic data representing compliance and violation is sufficiently representative for the model to generalize to real-world prescription fraud patterns
    Invoked when stating that joint training on synthetic data enables generalization to complex real-world datasets.
  • domain assumption Expert rules can be aligned with data samples in a shared embedding space while preserving their clinical meaning
    Central premise of the knowledge-guided component described in the approach.
invented entities (1)
  • soft rule embeddings no independent evidence
    purpose: To integrate structured domain expertise into the neural architecture by aligning rules and data samples
    New modeling component introduced to bridge expert heuristics and machine learning; no independent falsifiable evidence provided beyond the reported performance.

pith-pipeline@v0.9.0 · 5785 in / 1506 out tokens · 59913 ms · 2026-05-18T07:48:36.036189+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages · 1 internal anchor

  1. [1]

    Crossing the global quality chasm: improving health care worldwide,

    N. A. of Sciences, Medicine, M. Division, B. on Global Health, and C. on Improving the Quality of Health Care Globally, “Crossing the global quality chasm: improving health care worldwide,” 2018

    work page 2018

  2. [2]

    Waste in the us health care system: Estimated costs and potential for savings,

    W. H. Shrank, T. L. Rogstad, and N. Parekh, “Waste in the us health care system: Estimated costs and potential for savings,”JAMA, vol. 322, no. 15, pp. 1501–1509, 2019

    work page 2019

  3. [3]

    Health care fraud,

    Federal Bureau of Investigation, “Health care fraud,” https://www.fbi. gov/investigate/white-collar-crime/health-care-fraud, accessed: 2025- 05-20

    work page 2025

  4. [4]

    Global health and corruption,

    Transparency International, “Global health and corruption,” https: //www.transparency.org.uk/what-we-do/global-health-and-corruption, 2021, accessed: 2025-05-17

    work page 2021

  5. [5]

    The challenge of health care fraud,

    National Health Care Anti-Fraud Association, “The challenge of health care fraud,” https://www.nhcaa.org/tools-insights/ about-health-care-fraud/the-challenge-of-health-care-fraud/, 2021, accessed: 2025-05-17

    work page 2021

  6. [6]

    Healthcare anti- fraud,

    Canadian Life and Health Insurance Association, “Healthcare anti- fraud,” https://www.clhia.ca/web/CLHIA LP4W LND Webstation.nsf/ page/10D0B370160E723B85257F03005BD980, 2023, accessed: 2025- 05-17

    work page 2023

  7. [7]

    A probabilistic programming approach for outlier detection in healthcare claims,

    R. A. Bauder and T. M. Khoshgoftaar, “A probabilistic programming approach for outlier detection in healthcare claims,” inProceedings of the IEEE 15th International Conference on Machine Learning and Applications (ICMLA). IEEE, 2016, pp. 347–354

    work page 2016

  8. [8]

    Medicare fraud detection using neural networks,

    J. M. Johnson and T. M. Khoshgoftaar, “Medicare fraud detection using neural networks,”Journal of Big Data, vol. 6, pp. 63–69, 2019

    work page 2019

  9. [9]

    Deep learning for anomaly detection: A review,

    G. Pang, C. Shen, L. Cao, and A. van den Hengel, “Deep learning for anomaly detection: A review,”ACM Computing Surveys, vol. 54, no. 2, pp. 1–38, 2021

    work page 2021

  10. [10]

    Big data fraud detection using multiple medicare data sources,

    M. Herland, T. M. Khoshgoftaar, and R. A. Bauder, “Big data fraud detection using multiple medicare data sources,”Journal of Big Data, vol. 5, pp. 29–35, 2018

    work page 2018

  11. [11]

    Mining anomalies in medicare big data using patient rule induction method,

    S. Sadiq, Y . Tao, Y . Yan, and M.-L. Shyu, “Mining anomalies in medicare big data using patient rule induction method,” inProceedings of the IEEE 3rd International Conference on Multimedia Big Data (BigMM), 2017, pp. 185–192

    work page 2017

  12. [12]

    Procedure code overutilization detection from healthcare claims using unsupervised deep learning methods,

    M. Suesserman, S. Gorny, D. Lasaga, J. Helms, D. Olson, E. Bowen, and S. Bhattacharya, “Procedure code overutilization detection from healthcare claims using unsupervised deep learning methods,”BMC Medical Informatics and Decision Making, vol. 23, no. 1, p. 196, 2023

    work page 2023

  13. [13]

    Unsupervised feature selection and class labeling for credit card fraud,

    R. K. L. Kennedy, R. S. Nkole, and L. N. Mgutshini, “Unsupervised feature selection and class labeling for credit card fraud,”Journal of Big Data, vol. 12, no. 1, p. 75, 2025. [Online]. Available: https://journalofbigdata.springeropen.com/articles/ 10.1186/s40537-025-01154-1

    work page doi:10.1186/s40537-025-01154-1 2025

  14. [14]

    Deep semi-supervised anomaly detection,

    L. Ruff, R. A. Vandermeulen, N. G ¨ornitz, A. Binder, E. M ¨uller, K.-R. M¨uller, and M. Kloft, “Deep semi-supervised anomaly detection,”

    work page

  15. [15]

    Available: https://arxiv.org/abs/1906.02694

    [Online]. Available: https://arxiv.org/abs/1906.02694

    work page arXiv 1906

  16. [16]

    Deep one-class classification,

    L. Ruff, R. Vandermeulen, N. Goernitz, L. Deecke, S. A. Siddiqui, A. Binder, E. M ¨uller, and M. Kloft, “Deep one-class classification,” inProceedings of the 35th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, J. Dy and A. Krause, Eds., vol. 80. PMLR, 10–15 Jul 2018, pp. 4393–4402. [Online]. Available: https:/...

    work page 2018

  17. [17]

    Deep anomaly detection with deviation networks,

    G. Pang, C. Shen, and A. van den Hengel, “Deep anomaly detection with deviation networks,” inProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, 2019, pp. 353–362

    work page 2019

  18. [18]

    Deep weakly- supervised anomaly detection,

    G. Pang, C. Shen, H. Jin, and A. van den Hengel, “Deep weakly- supervised anomaly detection,” inProceedings of the 29th ACM SIGKDD international conference on knowledge discovery & data mining, 2023

    work page 2023

  19. [19]

    Know-gnn: Explainable knowledge-guided graph neural network for fraud detection,

    Y . Raoet al., “Know-gnn: Explainable knowledge-guided graph neural network for fraud detection,” inICONIP 2021, 2021

    work page 2021

  20. [20]

    Knowgraph: Knowledge-enabled anomaly detection via logical reasoning on graph data,

    A. Zhouet al., “Knowgraph: Knowledge-enabled anomaly detection via logical reasoning on graph data,” inProceedings of ACM CCS, 2024

    work page 2024

  21. [21]

    Explainable fraud detection with deep symbolic classification,

    S. Visbeeket al., “Explainable fraud detection with deep symbolic classification,” inXAI-FIN Workshop, 2023

    work page 2023

  22. [22]

    Sefraud: Graph-based self-explainable fraud detection via interpretative mask learning,

    K. Liet al., “Sefraud: Graph-based self-explainable fraud detection via interpretative mask learning,” inKDD, 2024

    work page 2024

  23. [23]

    Huge: Heterophily-guided unsupervised graph fraud detection,

    J. Panet al., “Huge: Heterophily-guided unsupervised graph fraud detection,” inAAAI, 2025

    work page 2025

  24. [24]

    Domain adaptation for customs fraud detection via prototype sharing,

    S. Parket al., “Domain adaptation for customs fraud detection via prototype sharing,” inAAAI, 2022

    work page 2022

  25. [25]

    Network-based prediction of drug combinations,

    F. Cheng, I. A. Kov ´acs, and A.-L. Barab ´asi, “Network-based prediction of drug combinations,”Nature Communications, vol. 10, no. 1197,

    work page

  26. [26]

    Available: https://doi.org/10.1038/s41467-019-09186-x

    [Online]. Available: https://doi.org/10.1038/s41467-019-09186-x

    work page doi:10.1038/s41467-019-09186-x

  27. [27]

    Topic modelling for medical prescription fraud and abuse detection,

    B. Zafari and T. Ekin, “Topic modelling for medical prescription fraud and abuse detection,”Journal of the Royal Statistical Society Series C: Applied Statistics, vol. 68, no. 3, pp. 751–769, 2019

    work page 2019

  28. [28]

    Vital signs: Changes in opioid prescribing in the united states, 2006–2015,

    G. P. Guy, K. Zhang, M. K. Bohm, J. Losby, B. Lewis, R. Young, L. B. Murphy, and D. Dowell, “Vital signs: Changes in opioid prescribing in the united states, 2006–2015,”MMWR. Morbidity and Mortality Weekly Report, vol. 66, no. 26, p. 697, 2017

    work page 2006

  29. [29]

    Medicare part d prescribers by provider and drug,

    Centers for Medicare & Medicaid Services, “Medicare part d prescribers by provider and drug,” https://data.cms.gov/ provider-summary-by-type-of-service/medicare-part-d-prescribers/ medicare-part-d-prescribers-by-provider-and-drug, 2025, accessed: 2025-05-22

    work page 2025

  30. [30]

    List of excluded individuals/entities (leie),

    U.S. Department of Health and Human Services, Office of Inspector General, “List of excluded individuals/entities (leie),” https://oig.hhs. gov/exclusions/exclusions list.asp, 2025, accessed: 2025-05-22

    work page 2025

  31. [31]

    Medical knowledge graph to enhance fraud, waste, and abuse detection on claim data: Model development and performance evaluation,

    H. Sun, J. Xiao, W. Zhu, Y . He, S. Zhang, X. Xu, L. Hou, J. Li, Y . Ni, and G. Xie, “Medical knowledge graph to enhance fraud, waste, and abuse detection on claim data: Model development and performance evaluation,”JMIR Medical Informatics, vol. 8, no. 7, p. e17653, 2020

    work page 2020

  32. [32]

    Improving fraud and abuse detection in general physician claims: A data mining study,

    H. Zare, R. Ghasemi, M. Ghazanfari, F. Roshani, and M. Roshani, “Improving fraud and abuse detection in general physician claims: A data mining study,”JMIR Medical Informatics, vol. 4, no. 1, p. e2, 2016

    work page 2016

  33. [33]

    Predicting new molecular targets for known drugs,

    M. J. Keiser, V . Setola, J. J. Irwin, C. Laggner, A. I. Abbas, S. J. Hufeisen, N. H. Jensen, M. B. Kuijer, R. C. Matos, T. B. Tran, R. Whaley, R. A. Glennon, J. Hert, K. L. H. Thomas, D. D. Edwards, B. K. Shoichet, and B. L. Roth, “Predicting new molecular targets for known drugs,”Nature, vol. 462, no. 7270, pp. 175–181, 2009. [Online]. Available: https:...

    work page 2009

  34. [34]

    Network predicting drug’s anatomical therapeutic chemical code,

    Y . Wang, S. Chen, N. Deng, and Y . Wang, “Network predicting drug’s anatomical therapeutic chemical code,”Bioinformatics, vol. 29, no. 10, pp. 1317–1324, 2013. [Online]. Available: https://academic.oup.com/ bioinformatics/article/29/10/1317/260431

    work page 2013

  35. [35]

    Drug target identification using side-effect similarity,

    M. Campillos, M. Kuhn, A. Gavin, L. J. Jensen, and P. Bork, “Drug target identification using side-effect similarity,”Science, vol. 321, no. 5886, pp. 263–266, 2008. [Online]. Available: https://pubmed.ncbi.nlm.nih.gov/18621671/

    work page arXiv 2008

  36. [36]

    Drugbank 6.0: the drugbank knowledgebase for 2024,

    C. Knox, M. Wilson, C. M. Klinger, and et al., “Drugbank 6.0: the drugbank knowledgebase for 2024,”Nucleic Acids Research, vol. 52, no. D1, p. D1265–D1275, 2024

    work page 2024

  37. [37]

    Medicaid program: Improper payments for brand name drugs,

    Office of the New York State Comptroller, “Medicaid program: Improper payments for brand name drugs,” Division of State Government Accountability, Office of the New York State Comptroller, Albany, NY , Tech. Rep. Report 2020-S-62, Dec. 2022. [Online]. Available: https: //www.osc.ny.gov/files/state-agencies/audits/pdf/sga-2023-20s62.pdf

    work page 2020

  38. [38]

    Ruling increases pharmacy false claims act risks,

    M. A. Dowell, “Ruling increases pharmacy false claims act risks,”U.S. Pharmacist, vol. 48, no. 9, pp. 7–12, Sep

    work page

  39. [39]

    Available: https://www.uspharmacist.com/article/ ruling-increases-pharmacy-false-claims-act-risks

    [Online]. Available: https://www.uspharmacist.com/article/ ruling-increases-pharmacy-false-claims-act-risks

    work page

  40. [40]

    Healthcare payer strategies to reduce the harms of opioids: White paper,

    C. for Medicare & Medicaid Services, H. F. P. Partnership, and N. at the University of Chicago, “Healthcare payer strategies to reduce the harms of opioids: White paper,” U.S. Department of Health & Human Services, Tech. Rep., 2017. [Online]. Available: https://www.cms.gov/ files/document/download-reducing-harms-opioids-white-paper.pdf

    work page 2017

  41. [41]

    The sinkhorn–knopp algorithm: convergence and appli- cations,

    P. A. Knight, “The sinkhorn–knopp algorithm: convergence and appli- cations,”SIAM Journal on Matrix Analysis and Applications, vol. 30, no. 1, pp. 261–275, 2008

    work page 2008

  42. [42]

    Learning repre- sentations by back-propagating errors,

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning repre- sentations by back-propagating errors,”Nature, vol. 323, no. 6088, pp. 533–536, 1986

    work page 1986

  43. [43]

    Weakly supervised anomaly detection via knowledge-data alignment,

    H. Zhao, C. Zi, Y . Liu, C. Zhang, Y . Zhou, and J. Li, “Weakly supervised anomaly detection via knowledge-data alignment,” inProceedings of the ACM Web Conference 2024, 2024, pp. 4083–4094

    work page 2024

  44. [44]

    A knowledge compilation map,

    A. Darwiche and P. Marquis, “A knowledge compilation map,”Journal of Artificial Intelligence Research, vol. 17, pp. 229–264, 2002

    work page 2002

  45. [45]

    Semi-Supervised Classification with Graph Convolutional Networks

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,”arXiv preprint arXiv:1609.02907, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016