Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Detecting Fraud in Financial Networks: A Semi-Supervised GNN Approach with Granger-Causal Explanations

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read SAGE-FIN shows that a semi-supervised graph neural network can detect fraudulent wallets and transactions in a Bitcoin network and attach a compact causal subgraph explanation to each flag.

desk verdict A useful semi-supervised GNN baseline on Elliptic++ that overclaims causality; the Granger label should be replaced with importance. read the letter →

arxiv 2507.01980 v1 pith:L5U7OE53 submitted 2025-06-25 q-fin.ST cs.LGstat.ML

Linh Nguyen , Marcel Boersma , Erman Acar This is my paper · ORCID
classification q-fin.STcs.LGstat.ML
keywords semi-supervisedlearninggraphneuralnetworkfinancialfrauddetectionbipartiteGrangercausalityexplainableAIElliptic++datasetanomaly
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

Fraud in financial networks is hard to detect because labeled examples are scarce and regulators require explanations, not just predictions. SAGE-FIN is a semi-supervised graph neural network for bipartite node-and-edge-attributed graphs: it learns from the few labeled wallets and transactions while reconstructing graph structure and features from the much larger unlabeled mass. On the Elliptic++ Bitcoin dataset it reports F1 scores of 0.806 for wallets, 0.807 for transactions, and 0.883 for edge prediction. A companion explainer ranks edges by the change in classification loss when each edge is removed, $C_j = L_2 - L_1$, and returns the top $K$ edges around a flagged node as a connected subgraph $S$ for which $p(y|S)$ stays close to $p(y|G)$. The paper's claim is that a semi-supervised GNN can be both practically useful and auditably explainable in a regulated financial setting.

What carries the argument

The central object is SAGE-FIN, a four-layer message-passing architecture built on the GraphBEAN encoder-decoder and extended with a linear node-classification head. Its message-passing rule aggregates, for wallet nodes, transaction nodes, and edges, the average of encoded neighbor representations and the edge representations connecting the two partitions, so both node partitions and edge attributes jointly drive a feature-reconstruction objective, an edge-reconstruction objective, and classification of labeled nodes. The explanation mechanism is the loss-difference quantity $C_j = L_2 - L_1$ (Eq. 5): the change in the node-classification loss when edge $e_j$ is excluded from the original graph, interpreted as the Granger-causal contribution of that edge. The explainer keeps edges whose removal raises the loss, ranks them by $C_j$, and returns the top $K$ as a connected subgraph $S$ containing the target node. Granger causality is used in its forecasting sense, here repurposed: an edge is causally relevant if removing it changes the model's ability to predict the class.

What would settle it

One could settle the causal claim by using the 49 timestamps in Elliptic++ to enforce temporal precedence: a genuinely Granger-causal edge must occur before the flag it explains, so if the top-$K$ subgraph frequently contains edges after the flagged node, or if a random $K$-edge subgraph preserves $p(y|G)$ just as well as the selected one, the causal interpretation would be refuted.

Watch

Extended reading notes

Core claim

SAGE-FIN demonstrates that a semi-supervised message-passing network that jointly reconstructs node features and edges and classifies labeled nodes can detect fraudulent wallets and transactions in a real bipartite Bitcoin network without labels on most of the graph. On Elliptic++, the test F1 scores are 0.806 for wallet addresses and 0.807 for transactions; the edge-reconstruction decoder reaches 0.883, which the paper uses as a proxy for anomalies among unlabeled nodes. The Granger causal explainer, adapted from a distillation method for bipartite graphs, computes for each edge in an $n$-hop neighborhood the loss difference $C_j = L_2 - L_1$ when that edge is removed, and selects the top $K$ edges that form a connected subgraph $S$ containing the target node. In the illustrated cases the selected subgraph preserves the model's verdict: the fraudulent wallet keeps about a 0.89 fraud probability on the 10-edge subgraph versus 0.84 on the full graph, and the non-fraudulent transaction stays near 0.91 versus 0.83. The authors therefore claim that fraud flags can come with compact, graph-structured causal explanations that point an auditor to the neighborhood that drove the model.

Load-bearing premise

The whole causal-explanation claim rests on treating 'removing an edge raises the loss' as evidence that the edge caused the fraud flag, and the paper itself concedes in Section 6.3 that its Granger notion is correlation-based and can be misled by hidden confounders; if that loss change is only an importance score, the explanations are not genuinely causal.

Editorial extensions

If this is right

  • Financial institutions can train a fraud detector on a small labeled sample while still using the full unlabeled transaction graph, reducing the cost of labeling.
  • Auditors receive a compact, human-readable subgraph of at most ten edges within a four-hop neighborhood for each flag, making the model's decision reviewable.
  • The same message-passing recipe extends to $k$-partite graphs, so the approach can be carried to other financial networks such as audit bookkeeping graphs.
  • Because edge prediction is used as a proxy for anomalies on unlabeled nodes, the model can also flag suspicious wallets and transactions that have no ground-truth label.
  • The paper's main advantage over higher-F1 tree baselines is not raw accuracy but the structural explanations it attaches to each prediction.

Reading between the lines

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

  • Beyond the paper, the loss-difference $C_j$ is an interventional probe on the model, not on the data-generating process; testing it against known laundering chains would show whether the selected subgraphs correspond to real fraud routes.
  • Beyond the paper, because the dataset carries 49 two-week timestamps, a temporal GNN could convert the model's static graph view into an evolving one and likely sharpen both detection and explanation; the paper itself notes the temporal dimension is neglected.
  • Beyond the paper, a null-model benchmark that randomizes edges or removes random edges would quantify how much of $p(y|S)$ staying close to $p(y|G)$ is due to the causal selection rule rather than to the natural redundancy of graph neighborhoods.
  • Beyond the paper, the $k$-partite generalization suggests the same semi-supervised scheme could be applied to audit networks or money-transfer systems where labeled fraud is rare.
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.

Referee Report

4 major / 5 minor

Summary. The paper proposes SAGE-FIN, a semi-supervised graph neural network for fraud detection on bipartite node-and-edge-attributed financial networks. The architecture extends GraphBEAN with a node classification decoder and evaluates on Elliptic++, reporting F1 scores of 0.806 for wallet addresses, 0.807 for transactions, and 0.883 for edges. The paper also introduces a 'Granger causal explainer' that selects a top-K subgraph by measuring the change in classification loss when each edge is removed, claiming that these subgraphs explain the model's fraud predictions.

Significance. The problem is practically important, and using the public Elliptic++ dataset is a strength. If the claims were fully supported, the paper could offer a semi-supervised tool with auditor-oriented explanations for financial fraud detection. However, the causal explanation claim is not supported by the method actually implemented: Eq. (5) defines a removal-based importance score rather than Granger causality. The experimental evidence is also weakened by single-run results, the absence of GNN baselines on the same data, and an internal inconsistency between the reported precision/recall and F1 values in Table 2. The detection component and dataset evaluation are salvageable, but the manuscript needs substantial revision before it can be accepted.

major comments (4)
  1. [3.2, Eq. (5)] The quantity C_j = L2 - L1 is a leave-one-edge-out removal importance score, not Granger causality. Granger causality, as defined in Section 2.3 Eq. (4), is a statistical hypothesis test about whether past values of one time series reduce the forecast-error variance of another; it requires lagged variables and a significance test. The algorithm in Section 3.2 instead compares the node-classification loss on the full graph with the loss after deleting an edge, and the authors concede in Section 6.3 that this notion is 'still based on correlation' and can be misled by hidden confounders. This is load-bearing because the paper's differentiating contribution over GraphBEAN is the claim of Granger-causal explanations. The authors should either implement a genuine Granger-causal procedure using the temporal structure of Elliptic++ (which they currently do not use) or reframe the explainer as a fidelity-based importance method and adjust the title, abstract, and conclusions accordingly.
  2. [5.2] The evaluation metric p(y|S) close to p(y|G) is circular. Since S is constructed from edges with the largest positive C_j = L2 - L1, keeping S is precisely the construction that minimizes the change in classification loss. Reporting p(y|S) close to p(y|G) does not provide independent evidence that S is causally relevant. The authors should compare the selected subgraphs against random subgraphs of the same size, against baseline explainers such as GNNExplainer and PGM-Explainer, and should report fidelity metrics over a sample of nodes rather than two illustrative examples.
  3. [Table 2] The reported F1 scores are numerically inconsistent with the reported precision and recall. For wallets, precision 0.802 and recall 0.775 imply F1 = 2*0.802*0.775/(0.802+0.775) ≈ 0.788, not 0.806. For transactions, precision 0.753 and recall 0.792 imply F1 ≈ 0.772, not 0.807. Additionally, all numbers come from a single run, and Section 4.2 states that hyperparameters were selected after experiments on the validation set. The authors should report means and standard deviations over multiple seeds, include significance tests, and correct the inconsistency.
  4. [4.2, Table 2] The comparison is made only against non-graph baselines (LR, RF, MLP, XGB). Since SAGE-FIN is an extension of GraphBEAN, the absence of GraphBEAN or another GNN baseline on the same Elliptic++ split makes it impossible to assess the contribution of the semi-supervised extension. Furthermore, the removal of the 72 aggregated transaction features in Section 4.2 is justified only by analogy to message passing; an ablation or feature-retention experiment is needed to verify that this removal does not discard predictive information.
minor comments (5)
  1. [2.2, Eqs. (1)-(3)] There is a notation mismatch: Eq. (2) writes hv(q)_i while the surrounding text and Eq. (3) use the subscript j for the V-partition nodes; please align the notation.
  2. [4.2] The random 70-15-15 split ignores the 49 timestamps in Elliptic++; even if the current model is static, a random split over temporally ordered data can leak information, and the choice should be justified or replaced with a temporal split.
  3. [6.1] The comparison of edge-prediction F1 with GraphBEAN on the Wikipedia dataset is not evidence of stability on Elliptic++; a direct comparison on the same dataset and split is needed.
  4. [Figures 6-9] The captions say 'circles present addresses' and 'squares present transactions'; 'present' should be 'represent'.
  5. [References] Reference [24] contains a typo ('Jounrla') and the reference list has inconsistent formatting, e.g., reference [14]; the list should be carefully proofread.

Circularity Check

2 steps flagged · score 4.0 of 10

Causal-explanation claim renames leave-one-out importance as Granger causality, and the fidelity validation is self-referential; the fraud-detection results themselves are external and independent.

  1. renaming known result [Section 3.2, Eq. (5); cf. Section 2.3, Eq. (4)]
    "The Granger causal effect of that particular edge on the SAGE-FIN’s output is then assessed by the reduction in the loss function... Formally, Cj = L2 − L1 (5) where, Cj is the Granger causal effect of edge ej... By ordering the edges based on their causal contributions Cj, the top K edges which constitutes a connected subgraph S signify the most influential connections and can be extracted to provide a causal explanation for the SAGE-FIN’s predictions."

    Eq. (4) defines Granger causality as a reduction in forecast-error variance from past values of a time series. Eq. (5) implements 'Granger causal effect' as a leave-one-edge-out change in node-classification loss of a static GNN. There is no time series, no lag, and no forecast-error comparison; C_j is the standard perturbation/importance quantity from the distillation method of [36], relabeled as causal. The claimed derivation of causal explanations therefore reduces to this importance score by definition, not to the stated Granger criterion.

  2. self definitional [Section 2.3 (definition) and Section 5.2 (validation)]
    "The Granger causal explainer also implies that the outcome of the classifier does not change significantly when using the entire graphG or only a subgraph S. ... Within the test set containing labeled data, p(y|S) should be close to p(y|G). A good explainer should be able to generate more compact subgraphs yet maintain the prediction accuracy."

    The explainer constructs S as the top-K edges ranked by C_j = L2 − L1, i.e., edges whose removal most increases classification loss. The validation then checks p(y|S) ≈ p(y|G), the same fidelity objective the selection was designed to approximate. Since the node-classification loss is cross-entropy, C_j is a monotone function of the change in p(y|·), so ranking by loss difference is ranking by prediction shift. The reported close probabilities restate the construction criterion; no independent motif ground truth or counterfactual baseline is provided, so the causal conclusion is not independently tested.

full rationale

The semi-supervised fraud-detection contribution is not circular: SAGE-FIN is evaluated on the held-out Elliptic++ test set against LR, RF, MLP, and XGB, and the architecture is an adaptation of the external GraphBEAN model. No load-bearing self-citation chain appears; the authors’ own prior works are cited only as background. The circularity is confined to the explanation module. Section 3.2 defines the 'Granger causal effect' C_j as a leave-one-edge-out loss difference, which is neither the time-series variance comparison of Eq. (4) nor a causal identification; it is a known importance-score method relabeled. Section 5.2 then validates the resulting subgraph by p(y|S) ≈ p(y|G), which is the same invariance property used to define and select S, so the validation largely restates the construction criterion. The paper’s own Section 6.3 concession that Granger causality here 'is still based on correlation' and that hidden confounders can misguide further supports the view that the causal language is a label rather than a derived result. A score of 4 reflects partial circularity in the central causal-explanation claim while acknowledging that the fraud-detection results are independent and externally benchmarked.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claims rest on a handful of manually chosen hyperparameters (dim, learning rate, negative sampling, n-hop, K), on the trustworthiness of Elliptic++ labels, on the proxy assumption that edge prediction measures node fraud, and on the equation C_j = L2-L1 being interpreted as causality. No new theoretical entities are postulated.

free parameters (5)
  • hidden dimension = 32
    Chosen after experiments on the validation set (Section 4.2); the model capacity depends on it.
  • learning rate = 0.005
    Adam learning rate selected by validation experiments (Section 4.2).
  • negative edge sampling ratio = 5x
    Negative cases set to five times the number of edges (Section 4.2); this synthetic imbalance shapes the edge decoder.
  • number of hops n = 4
    Chosen qualitatively from degree distributions to keep explanations interpretable (Section 4.2); limits the explainer's search space.
  • number of explanation edges K = 10
    Set to 10, chosen so the subgraph is human-interpretable (Section 4.2); K directly defines the explanation size.
assumptions (4)
  • domain assumption Elliptic++ ground-truth labels are accurate and complete for labeled nodes.
    The model is trained and evaluated against these labels without independent verification (Section 4.1).
  • domain assumption Edge prediction accuracy is a valid proxy for node anomaly detection on unlabeled data.
    Stated in Section 5.1; the paper notes this proxy is imperfect and generally easier than node classification.
  • ad hoc to paper Removal of the 72 aggregated transaction features does not discard information needed for fraud detection.
    The authors drop these features because they duplicate message passing (Section 4.2), but this is a modeling choice with no ablation to justify it.
  • ad hoc to paper The change in node-classification loss when an edge is removed reflects the causal influence of that edge on the prediction.
    This is the core premise of the Granger explainer (Eq. 5, Section 3.2), yet loss difference is at best a correlation-based importance score, as the authors concede in Section 6.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detecting Fraud in Financial Networks: A Semi-Supervised GNN Approach with Granger-Causal Explanations." pith.science (2026). https://pith.science/paper/L5U7OE53

@misc{pith2026250701980,
  author       = {Pith},
  title        = {Pith review of: Detecting Fraud in Financial Networks: A Semi-Supervised GNN Approach with Granger-Causal Explanations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L5U7OE53}},
  note         = {Machine review of arXiv:2507.01980}
}
read the original abstract

Fraudulent activity in the financial industry costs billions annually. Detecting fraud, therefore, is an essential yet technically challenging task that requires carefully analyzing large volumes of data. While machine learning (ML) approaches seem like a viable solution, applying them successfully is not so easy due to two main challenges: (1) the sparsely labeled data, which makes the training of such approaches challenging (with inherent labeling costs), and (2) lack of explainability for the flagged items posed by the opacity of ML models, that is often required by business regulations. This article proposes SAGE-FIN, a semi-supervised graph neural network (GNN) based approach with Granger causal explanations for Financial Interaction Networks. SAGE-FIN learns to flag fraudulent items based on weakly labeled (or unlabelled) data points. To adhere to regulatory requirements, the flagged items are explained by highlighting related items in the network using Granger causality. We empirically validate the favorable performance of SAGE-FIN on a real-world dataset, Bipartite Edge-And-Node Attributed financial network (Elliptic++), with Granger-causal explanations for the identified fraudulent items without any prior assumption on the network structure.

Figures

Figures reproduced from arXiv: 2507.01980 by the authors.

Figure 1
Figure 1. An example graph of the Bitcoin wallets-transactions network. layers are allocated for the encoder and the remaining half are utilized for the feature decoder. The q-th layer representation for node ui within set U, node vj within set V , and edge ei,j within set E are presented as h u(q) i , h v(q) j , and h e(q) i,j accordingly. According to the GraphBEAN architecture proposed by [29], to compute the new represent… view at source ↗
Figure 2
Figure 2. Illustration of the message passing process for nodes and edges adapted from [29]. passed to each node and edge, the subsequent representations are computed. The messages are passed to a linear operation and then the output of the linear op￾eration is normalized using batch normalization. Subsequently, this normalized output is passed through a ReLU activation function. 2.3 Granger Causality Granger causality, as or… view at source ↗
Figure 3
Figure 3. The SAGE-FIN architecture adds a node classification decoder to the unsuper￾vised GraphBEAN architecture [29] and supplement the neural network with Causal explanation layer after the model training. The new elements contributed by this paper are highlighted in orange [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Degree distribution of wallet address node set 5 Results Our evaluation consists of two parts: anomaly detection and their Granger causal explanations. 5.1 Anomaly detection To evaluate the anomaly detection of the SAGE-FIN model on the Elliptic++ dataset, metrics incl…
Figure 5
Figure 5. Figure 5: Degree distribution of transaction node set data by the SAGE-FIN model and the four state-of-art machine learning models are presented in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Top 10 edges as the causal explanation for a target fraudulent wallet address (circles present addresses while squares present transactions) [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Top 10 edges as the causal explanation for a target non-fraudulent transaction (circles present addresses while squares present transactions) [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Top 6 edges as the causal explanation for a target fraudulent wallet address (circles present addresses while squares present transactions) 6.3 Limitations In this study, the application of the SAGE-FIN neural network for detecting fraudulent activities within a financ…
Figure 9
Figure 9. Figure 9: Top 6 edges as the causal explanation for a target non-fraudulent transaction (circles present addresses while squares present transactions) Methods like structural causal models or causal representation learning could be used to identify confounding variables and enha…
Figure 10
Figure 10. Figure 10: Illustration of Bitcoin wallet addresses-transactions network as a bipartite graph [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 40 canonical work pages

  1. [1]

    Author, F.: Article title. Journal 2(5), 99–110 (2016) Albashrawi, M.: Detecting Financial Fraud Using Data Mining Techniques: A Decade Review from 2004 to 2015, Journal of Data Science, volume 14 (553-570) (2016). SAGE-FIN: A semi-supervised GNN approach with Causal Explanations 21

    work page 2016

  2. [2]

    Ngai, T., Hu, Y., Wong, H., Chen, Y., Sun, X.:The application of data mining tech- niques in financial fraud detection: A classification framework and an academic re- view of literature, Journal of Decision Support System, volume 50 (559-569) (2011)

    work page 2011

  3. [3]

    Chandola, V.,Banerjee, A., Kumar, V.:Anomaly detection: A survey, ACM comput- ing surveys (CSUR), Volume 41-No.3 (page 1-58) (2009)

    work page 2009

  4. [4]

    Thilakarathna, K., Fukuda, K., Seneviratne, A., Hu, Y., Seneviratne, S.: Charac- terizing and detecting money laundering activities on the bitcoin network, preprint arXiv:1912.12060 (2019)

    arXiv 2019

  5. [5]

    IEEE (page 188-194) (2016)

    Monamo, P., Marivate, V., Twala, B.: A multifaceted approach to Bitcoin fraud detection:Globalandlocaloutliers,15thIEEEInternationalConferenceonMachine Learning and Applications (ICMLA). IEEE (page 188-194) (2016)

    work page 2016

  6. [6]

    Springer, (page 350–373) (2024)

    Visbeek, S., Acar, E., Hengst, F.: Explainable Fraud Detection with Deep Symbolic Classification, In World Conference on Explainable Artificial Intelligence. Springer, (page 350–373) (2024)

    work page 2024

  7. [7]

    Azarm, C, Zeelt, M., Acar, E.: On the Potential of Network-Based Features for Fraud Detection (2024)

    work page 2024

  8. [8]

    Boersma, M.:Complex networks in audit: A data-driven modelling approach, Uni- versiteit van Amsterdam (2024)

    work page 2024

Show all 42 references
  1. [9]

    Sourabh, S., Hoogduin, L., Kandhai, D., Boersma, M.: Financial statement net- works: an application of network theory in audit, The Journal of Network Theory in Finance (59-85) (2018)

    2018

  2. [10]

    Bussmann, N., Giudici, P., Marinelli, D., Papenbrock, J.: Explainable AI in Fintech Risk Management. Front. Artif. Intell. 3 (2020)

    2020

  3. [11]

    Gomber, P., Kauffman, RJ., Parker, C., Weber, BW.: On the fintech revolution: Interpreting the forces of innovation, disruption, and transformation in financial services. J. Manag. Inf. Syst. 35, 1 (page 220–265) (2018)

    2018

  4. [12]

    Weber, P., Carl, KV., Hinz, O.: Applications of Explainable Artificial Intelligence in Finance—a systematic review of Finance, Information Systems, and Computer Science literature. Manag. Rev. Q. 74, 2 (page 867–907) (2024)

    2024

  5. [13]

    Philosophical Issues 13, (page 316–328) (2003)

    Yablo, S.: Causal relevance. Philosophical Issues 13, (page 316–328) (2003)

    2003

  6. [14]

    Motie, S., Raahemi, B.: Financial fraud detection using graph neural networks: A systematic review, Journal of Expert Systems With Applications, volume 240 (2024)

    2024

  7. [15]

    Elmougy, Y., Liu, L.: Demystifying Fraudulent Transactions and Illicit Nodes in the Bitcoin Network for Financial Forensics, Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (2023)

    2023

  8. [16]

    ArXiv abs/1908.02591 (2019)

    Weber, M., Domeniconi, G., Chen, J., Weidele, D., Bellei, C., Robinson, T., Leiser- son, C: Anti-money laundering in Bitcoin: Experimenting with Graph Convolutional Networks for financial forensics. ArXiv abs/1908.02591 (2019)

    2019 arXiv

  9. [17]

    arXiv [cs.LG] (2016)

    Kipf, TN., Welling, M.: Semi-supervised classification with graph convolutional networks. arXiv [cs.LG] (2016)

    2016

  10. [18]

    Zhang, S., Tong, H., Xu, J., Maciejewski, R.: Graph convolutional networks: a comprehensive review. Comput. Soc. Netw. 6, 1 (Nov. 2019),

    2019

  11. [19]

    arXiv [cs.LG] (2018)

    Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C., Sun, M.: Graph Neural Networks: A Review of Methods and Applications. arXiv [cs.LG] (2018)

    2018

  12. [20]

    In 4th ACM International Conference on AI in Finance

    Ghosh, S., Anand, R., Bhowmik, T., Chandrashekhar, S.: GoSage: Heterogeneous graph neural network using hierarchical attention for collusion fraud detection. In 4th ACM International Conference on AI in Finance. ACM, New York, NY, USA (2023). 22 Nguyen et al

    2023

  13. [21]

    In 4th ACM International Conference on AI in Finance

    Jiang, S., Uddin, A., Wei, Z, Yu, D.: The network of mutual funds: A dynamic heterogeneous graph neural network for estimating mutual funds performance. In 4th ACM International Conference on AI in Finance. ACM, New York, NY, USA (2023)

    2023

  14. [22]

    In 4th ACM International Confer- ence on AI in Finance

    Cao, Z., Chen, Z., Mishra, P, Amini, H., Feinstein, Z.: Modeling inverse demand function with explainable dual neural networks. In 4th ACM International Confer- ence on AI in Finance. ACM, New York, NY, USA (2023)

    2023

  15. [23]

    Giudici,P.,Spelta,A.:GraphicalNetworkModelsforInternationalFinancialFlows, Journal of Business and Economic Statistics, Volume 34, (page 128-138), (2016)

    2016

  16. [24]

    Billio, M., Getmansky, M., Lo, A., Pelizzon, L.: Econometric measures of connect- edness and systemic risk in the finance and insurance sectors, Jounrla of Financial Economics, Volume 104, Issue 3, (page 535-559), (2012)

    2012

  17. [25]

    Battiston, S., Puliga, M, Kaushik, R., Tasca, P., Caldarelli, G.: DebtRank: Too Central to Fail? Financial Networks, the FED and Systemic Risk, Nature Scientific reports 2, Article number 541 (2012)

    2012

  18. [26]

    Akoglu, L., Tong, H., Koutra, D.: Graph based anomaly detection and description: a survey, DMKD Journal (2015)

    2015

  19. [27]

    Jing, R., Zheng, X., Tian, H., Zhang, X., Chen, W., Wu, D., Zeng, D.: A Graph- Based Semi-Supervised Fraud Detection Framework, 4th IEEE International Con- ference on Cybernetics (Cybconf) (page 1-5) (2019)

    2019

  20. [28]

    Rao, Y., Ren, X., Duan, C., Mi, X., Cheng, J., Chen, Y., Wei, X.: Knowledgeguided fraud detection using semi-supervised graph neural network, Web Information Sys- temsEngineering–WISE2021:22ndInternationalConference(page385-393)(2021)

    2021

  21. [29]

    Fathony, R., Ng, J., Chen, J.: Interaction-Focused Anomaly Detection on Bipar- tite Node-and-Edge-Attributed Graphs, International Joint Conference on Neural Networks (IJCNN) (2023)

    2023

  22. [30]

    IEEE Trans

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

    2023

  23. [31]

    Yuan, H., Tang, J., Hu, X., Ji, S.: XGNN: Towards Model-Level Explanations of Graph Neural Networks, Proceedings SIGKDD ACM (2020)

    2020

  24. [32]

    Giudici, P.: Learning in graphical gaussian models, Bayesian statistics, Volume 5, (page 621-628) (1996)

    1996

  25. [33]

    Ahelegbey, A., Giudici, P.: NetVIX - a network volatility index for financial mar- kets, Physica A: statistical mechanics and its applications, Volume 594 (2022)

    2022

  26. [34]

    Ying, Z., Bourgeois, D., You, J., Zitnik, M., Leskovec, J.: Generating Explana- tions for Graph Neural Networks, Proceedings of Advances in Neural Information Processing Systems conference (page 9244–9255) (2019)

    2019

  27. [35]

    Thai, T: Probabilistic Graphical Model Explanations for Graph Neural Networks, Proceedings of Advances in Neural Information Processing Systems con- ference (2020)

    Vu, N. Thai, T: Probabilistic Graphical Model Explanations for Graph Neural Networks, Proceedings of Advances in Neural Information Processing Systems con- ference (2020)

    2020

  28. [36]

    Li,B.,Lin,W.,Lan,H.:GenerativeCausalExplanationsforGraphNeuralNetwork (2021)

    2021

  29. [37]

    Granger, C.: Investigating Causal Relations by Econometric Models and Cross- SpectralMethods, Econometrica:Journalof theEconometricSociety (page 424-438) (1969)

    1969

  30. [38]

    SAGE-FIN: A semi-supervised GNN approach with Causal Explanations 23

    Tank, A., Covert, I., Foti, N., Shojaie, A., Fox, E.: Neural Granger causality, IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(8), (page 4267-4279) (2021). SAGE-FIN: A semi-supervised GNN approach with Causal Explanations 23

    2021

  31. [39]

    Caruana, R., Lou, Y., Gehrke, J., Koch, P., Sturm, M., Elhadad, N.: Intelligible Models for Healthcare: Predicting Pneumonia Risk and Hospital 30-day Readmis- sion, Proceedings of the 21th ACM SIGKDD International Conference on Knowl- edge Discovery and Data Mining, (page 1721...

    2015

  32. [40]

    M., Lee, S.I.: A Unified Approach to Interpreting Model Predictions, Advances in Neural Information Processing Systems (NeurIPS), (page 4765–4774) (2017)

    Lundberg, S. M., Lee, S.I.: A Unified Approach to Interpreting Model Predictions, Advances in Neural Information Processing Systems (NeurIPS), (page 4765–4774) (2017)

    2017

  33. [41]

    John Wiley and Sons (2016)

    Pearl, J., Glymour, M., Jewell, NP.: Causal inference in statistics: A primer. John Wiley and Sons (2016)

    2016

  34. [42]

    Explainable Artificial Intelligence for crypto asset allocation, Finance Research Letters, Volume 47 (2022)

    Babaei, G., Giudici, P., Raffinetti, E. Explainable Artificial Intelligence for crypto asset allocation, Finance Research Letters, Volume 47 (2022). A Appendix A The Elliptic++ dataset presents a bipartite address-transaction graph that il- lustrates the flow of Bitcoin across...

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.