Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Large Language Models for Cryptocurrency Transaction Analysis: A Bitcoin Case Study

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Large language models, given a compact graph format and a connectivity-preserving sampler, can analyze real Bitcoin transaction graphs with strong local accuracy and usable few-shot classification.

desk verdict First systematic LLM-on-Bitcoin-graph study with useful artifacts, but an unexplained 20-25 point gap between Table I and Table V undermines the headline accuracy claim. read the letter →

arxiv 2501.18158 v3 pith:YGURD27V submitted 2025-01-30 cs.CR cs.LG

Yuchen Lei , Yuexin Xiang , Qin Wang , Rafael Dowsley , Tsz Hon Yuen , Kim-Kwang Raymond Choo , Jiangshan Yu This is my paper · ORCID
classification cs.CRcs.LG
keywords LLMBitcointransactiongraphrepresentationLLM4TGCETraSfew-shotclassificationcybercrimedetectionexplainableAI
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 tests whether LLMs can make sense of real Bitcoin transaction graphs rather than toy graphs. It proposes LLM4TG, a compact human-readable text format for transaction graphs, and CETraS, a connectivity-preserving sampler, and organizes evaluation into three levels: foundational metrics, characteristic overview, and contextual interpretation. The headline results: node-level attribute recognition exceeds 98.5% accuracy, meaningful subgraph characteristics appear in 95.0% of GPT-4o responses, and few-shot address classification reaches 72.43% top-3 accuracy with explanations. The paper argues this makes LLMs useful for exploratory cybercrime detection in scarce-label settings, while noting exact classification still trails engineered models and explanations need verification.

What carries the argument

Two mechanisms carry the argument. LLM4TG is a text-based graph representation that writes nodes as key-value property lists grouped into address and transaction layers, reducing token use relative to GEXF, GML, and GraphML while keeping node-special-information accuracy near 100% versus 12-23% for standard formats in GPT-4o. CETraS is a sampling algorithm that scores each node by $I_{\text{node}} = (\log(a_{\text{in}}+a_{\text{out}}+1)+\beta\log(d_{\text{in}}+d_{\text{out}}+1))/(L_s+1)$, removes low-importance nodes, and restores shortest paths among retained nodes, keeping sampled subgraphs connected. Together they fit multiple moderately large graphs into one prompt, which is what makes few-shot classification and characteristic overview feasible under token limits.

What would settle it

Run the same few-shot prompts on an independently labeled Bitcoin address dataset where reference subgraphs are selected by a documented random or stratified protocol; if top-3 accuracy falls well below 72.43% or explanations become systematically wrong under neutral selection, the contextual-interpretation claim fails.

Watch

Extended reading notes

Core claim

The central claim is that LLMs can meaningfully analyze real-world cryptocurrency transaction graphs when inputs are represented compactly and connectivity is preserved. On BASD subgraphs, GPT-4 and GPT-4o recover node-level attributes such as in/out degrees and values at 98.5-100% accuracy, while whole-graph comparison metrics fall to 24-58%. For characteristic overview, GPT-4o produces high-quality responses in 82.5% of cases and meaningful characteristics in 95.0%. For few-shot address classification, LLMs reach 72.43% top-3 accuracy with reasons, beating SVM and MLP but trailing tree models and GNNs. The paper reads this as evidence that LLMs are strong for exploratory and explainable analysis, not yet replacements for specialized detectors.

Load-bearing premise

The evaluation assumes that the BASD and BABD subgraphs, their labels, the ten hand-picked features, and the author-selected few-shot reference examples are representative enough that the measured accuracy and explanation quality transfer to real Bitcoin analysis.

Editorial extensions

If this is right

  • If the results hold, LLMs can serve as explainable first-pass screeners for Bitcoin addresses when labeled data is scarce, especially for classes like money laundering and Ponzi schemes.
  • Compact representations like LLM4TG may extend LLM-based graph analysis to other real-world graph domains that currently exceed token limits.
  • The gap between near-perfect node-level accuracy and 24-58% global-comparison accuracy implies that better formats alone will not fix LLM arithmetic and comparison weaknesses.
  • The three-level evaluation scheme offers a reusable template for measuring LLM understanding of transaction graphs beyond Bitcoin.

Reading between the lines

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

  • A natural next experiment is to vary CETraS retention ratios and use a documented random or stratified protocol for selecting few-shot reference subgraphs to see how much of the 72.43% top-3 score depends on the authors' hand-picked examples.
  • Because the representation is chain-agnostic, the same pipeline could be applied to Ethereum or other account-based ledgers, though edge semantics and address models differ.
  • Requiring the model to quote the exact node properties behind each top-3 choice would turn the occasionally inaccurate explanations into checkable citations.
  • The near-perfect preservation of node-special information suggests LLM4TG could become a general serialization for LLM question answering over attributed graphs.
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 / 4 minor

Summary. The paper presents a three-tiered evaluation framework for assessing LLMs' ability to analyze Bitcoin transaction graphs. It introduces LLM4TG, a text-based graph representation format, and CETraS, a connectivity-preserving graph sampling algorithm, and reports experiments with GPT-3.5, GPT-4, GPT-4o, DeepSeek, and LLaMA on the authors' BASD and BABD datasets. The main claims are that node-level foundational metrics exceed 98.50% accuracy, that 95.00% of characteristic-overview responses are meaningful, and that few-shot address classification reaches 72.43% top-3 accuracy with explanatory output.

Significance. If the results are reproducible, the paper would be a useful first systematic study of LLM-based cryptocurrency transaction analysis, with practical contributions in token-efficient graph representation, sampling, and a layered benchmark. The authors make their code, prompts, and results publicly available, and the comparison with traditional ML models (SVM, MLP, DT, RF, CatBoost, GNN) is a valuable reference point. However, the quantitative claims are undermined by an unresolved inconsistency between the main evaluation and the cross-format evaluation, and by the absence of statistical uncertainty measures for the reported percentages.

major comments (4)
  1. [Section V-B (Table I) and Section VI-A (Table V)] The paper reports mutually contradictory node-level accuracies for LLM4TG. Table I lists node in degree 99.25%/99.40% and node out degree 100%/99.80% for GPT-4/GPT-4o, while Table V lists node in degree 73.75%, node out degree 75.83%, node in value 75.42%, and node out value 77.08% for GPT-4o, with similar magnitudes for DeepSeek. The text does not explain whether Table V uses different graphs, prompts, sample sizes, or evaluation protocols. Because the abstract and the Level 1 findings rely explicitly on the high numbers, this discrepancy is load-bearing: the authors must either reconcile the two protocols or revise the headline claim.
  2. [Section V-A through V-D] All evaluation results are reported as single percentages without sample sizes, repetition counts, or confidence intervals. For example, Table I reports 99.25% versus 99.40% across models, and Table II reports 95.00% meaningful responses, but the reader cannot assess the variance of these estimates. The Level 2 'meaningful' rating is a subjective judgment by the authors with no documented rubric, no independent raters, and no inter-rater agreement. The paper should provide the number of test cases per metric, confidence intervals where appropriate, and a reproducible rating procedure for the qualitative assessments.
  3. [Section V-D] The few-shot classification evaluation does not document how the reference subgraphs and reference features were selected. The paper's own Section VI-B identifies 'reference graph selection' as a challenge, yet the manuscript does not state the number of chosen references, the sampling procedure, or the criteria for representativeness. Since the reported top-3 accuracy of 72.43% and the accompanying explanations are highly sensitive to these choices, this is a reproducibility gap for the contextual-interpretation results.
  4. [Section V-A] The evaluation is conducted entirely on the authors' own datasets, BASD [35] and BABD [13], including the ten-feature importance list taken from BABD. There is no external validation on an independent benchmark such as the Elliptic dataset. Because the feature set and the label definitions originate from the same group, the measured accuracies may be partly dataset-specific. The paper should explicitly discuss this as a generalizability limitation and, where possible, include at least one external dataset or clearly justify why the current evaluation supports broad conclusions.
minor comments (4)
  1. [Section II-B] The 'Current Gaps' paragraph contains a broken citation placeholder '[?]'; this reference should be filled in or removed.
  2. [Section IV-C] Algorithm 1 has an unreachable second 'return' on line 21, which should be deleted.
  3. [Figure 3 caption] The bar legend is described only as 'bars 1st-5th'; labeling each bar directly or adding a proper legend would improve readability.
  4. [Section VI-A] The prose states that LLM4TG node-level metrics reach '75–80%' accuracy in Table V, but some values in the table are 73.75%, 73.33%, and 75.42%. The text should be aligned with the exact table values to avoid overstating the results.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the reported accuracies are empirical measurements on public datasets, not outputs derived from fitted equations or from the paper's own construction.

full rationale

The evaluation chain is measurement-based rather than derivation-based. LLM4TG and CETraS are proposed and then tested by direct token counts (Fig. 7, Table IV) and by querying LLMs on graph samples; no parameter is fitted to the reported metrics. The Level 1, 2, and 3 numbers are empirical LLM outputs compared against ground-truth labels from the BASD and BABD datasets, which are public resources even though they originate from the authors' prior work. The BABD top-10 feature list used in Section V-D is an input to the few-shot prompts, not a quantity derived from the LLM outputs, so the classification results are not forced by that list. CETraS's beta is explicitly hand-set (Section IV-C), not calibrated to the evaluation targets. The only self-referential elements are citations to the authors' own datasets and feature list; these are not load-bearing in a circular sense because the datasets provide external ground truth. The apparent inconsistency between Table I and Table V node-level accuracies is a reproducibility concern, not a circularity: it does not involve an equation or fitted parameter reducing to its own inputs. Overall, no step in the claimed derivation chain is circular.

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

The central evaluation introduces no new physical or conceptual entities. The main hand-chosen parameters are in the sampling algorithm, and the main assumptions concern dataset representativeness and the validity of the proposed importance metric. No parameter is fitted to the reported accuracies, so the circularity burden is low.

free parameters (2)
  • beta (CETraS connectivity weight) = 2
    Hand-set in Section IV-C to prioritize connectivity over token amounts; no sensitivity analysis is given.
  • Ntarget (CETraS target node count) = varies by experiment
    The target number of retained nodes is an input to Algorithm 1; the paper does not specify how it is chosen across Level 2 and Level 3 experiments, so the level of compression is a free choice that affects downstream accuracy.
assumptions (5)
  • domain assumption BASD and BABD labels and subgraph structures are correct and representative of Bitcoin addresses.
    All measured accuracies are computed on these datasets (Section V-A); label noise or sampling bias would directly distort the headline percentages.
  • domain assumption The Bitcoin transaction graph built from the 22-month on-chain window is complete and accurate.
    Section V-A constructs the graph from raw blockchain data; node degrees and values used in Level 1 tasks depend on this construction being correct.
  • domain assumption The ten features selected from BABD retain the discriminative signal needed for address classification.
    Feature-based classification (Section V-D) uses only this hand-picked set; if important signals are missing, the reported top-3 accuracy understates or misstates feature-based LLM capability.
  • domain assumption LLM API responses are stable enough for point-estimate evaluation.
    The experiments query API models once per task without reported temperature control or repeated sampling; stochasticity could shift the reported percentages.
  • ad hoc to paper The CETraS importance metric preserves the structure needed for classification.
    The formula Inode = log(ain+aout+1) + beta*log(din+dout+1)/(Ls+1) is proposed in Section IV-C without independent validation; raw-graph classification results depend on its fidelity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Large Language Models for Cryptocurrency Transaction Analysis: A Bitcoin Case Study." pith.science (2026). https://pith.science/paper/YGURD27V

@misc{pith2026250118158,
  author       = {Pith},
  title        = {Pith review of: Large Language Models for Cryptocurrency Transaction Analysis: A Bitcoin Case Study},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YGURD27V}},
  note         = {Machine review of arXiv:2501.18158}
}
read the original abstract

Cryptocurrencies are widely used, yet current methods for analyzing transactions often rely on opaque, black-box models. While these models may achieve high performance, their outputs are usually difficult to interpret and adapt, making it challenging to capture nuanced behavioral patterns. Large language models (LLMs) have the potential to address these gaps, but their capabilities in this area remain largely unexplored, particularly in cybercrime detection. In this paper, we test this hypothesis by applying LLMs to real-world cryptocurrency transaction graphs, with a focus on Bitcoin, one of the most studied and widely adopted blockchain networks. We introduce a three-tiered framework to assess LLM capabilities: foundational metrics, characteristic overview, and contextual interpretation. This includes a new, human-readable graph representation format, LLM4TG, and a connectivity-enhanced transaction graph sampling algorithm, CETraS. Together, they significantly reduce token requirements, transforming the analysis of multiple moderately large-scale transaction graphs with LLMs from nearly impossible to feasible under strict token limits. Experimental results demonstrate that LLMs have outstanding performance on foundational metrics and characteristic overview, where the accuracy of recognizing most basic information at the node level exceeds 98.50% and the proportion of obtaining meaningful characteristics reaches 95.00%. Regarding contextual interpretation, LLMs also demonstrate strong performance in classification tasks, even with very limited labeled data, where top-3 accuracy reaches 72.43% with explanations. While the explanations are not always fully accurate, they highlight the strong potential of LLMs in this domain. At the same time, several limitations persist, which we discuss along with directions for future research.

Figures

Figures reproduced from arXiv: 2501.18158 by the authors.

Figure 1
Figure 1. LLM Evaluation Framework for Bitcoin Transaction [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Examples with features responses make up 82.50% of GPT-4o’s responses, compared with 62.50% for GPT-4. For average-quality responses, GPT￾4o scores 13.75%, with 12.50% flawed and 1.25% irrelevant; While GPT-4 records 26.25%, with 7.50% flawed and 18.75% irrelevant. Low-quality responses are 3.75% for GPT-4o and 11.25% for GPT-4. Overall, GPT-4o produces meaningful outputs in 95.00% of cases versus 70.00% for GPT-4. … view at source ↗
Figure 3
Figure 3. Classification via different LLMs (x for LLM models, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: LLMs’ performance in contextual interpretation using [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Evaluation on different models (x axis for category, y for corresponding rates (%)) [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Token consumption in different graph formats [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Talking Transactions: Decentralized Communication through Ethereum Input Data Messages (IDMs)

    cs.CR 2025-05 conditional novelty 7.0 of 10

    Ethereum's transaction input data field is used as a decentralized messaging medium, with English messages dominated by security warnings and Chinese messages by emotional expression.

Reference graph

Works this paper leans on

52 extracted references · 45 canonical work pages · cited by 1 Pith paper

  1. [35]

    Leveraging subgraph structure for exploration and analysis of Bitcoin address

    Yuexin Xiang, Tiantian Li, and Yuquan Li. Leveraging subgraph structure for exploration and analysis of Bitcoin address. In IEEE International Conference on Big Data (BigData) , pages 1957–1962, 2022

    work page 1957

  2. [13]

    BABD: A Bitcoin address behavior dataset for pattern analysis

    Yuexin Xiang, Yuchen Lei, Ding Bao, Tiantian Li, Qingqing Yang, Wenmao Liu, Wei Ren, and Kim-Kwang Raymond Choo. BABD: A Bitcoin address behavior dataset for pattern analysis. IEEE Transactions on Information Forensics and Security (TIFS) , 19:2171–2185, 2024

    work page 2024

  3. [1]

    A survey on evaluation of large language models

    Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, et al. A survey on evaluation of large language models. ACM Transactions on Intelligent Systems and Technology (TIST) , 15(3):1–45, 2024

    work page 2024

  4. [2]

    Recent advances in natural language processing via large pre- trained language models: A survey

    Bonan Min, Hayley Ross, Elior Sulem, Amir Pouran Ben Veyseh, Thien Huu Nguyen, Oscar Sainz, Eneko Agirre, Ilana Heintz, and Dan Roth. Recent advances in natural language processing via large pre- trained language models: A survey. ACM Computing Surveys (CSUR) , 56(2):1–40, 2023

    work page 2023

  5. [3]

    Fine-tuning large neural language models for biomedical natural language processing

    Robert Tinn, Hao Cheng, Yu Gu, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Fine-tuning large neural language models for biomedical natural language processing. Patterns, 4(4), 2023

    work page 2023

  6. [4]

    Visionllm: Large language model is also an open-ended decoder for vision-centric tasks

    Wenhai Wang, Zhe Chen, Xiaokang Chen, Jiannan Wu, Xizhou Zhu, Gang Zeng, Ping Luo, Tong Lu, Jie Zhou, Yu Qiao, et al. Visionllm: Large language model is also an open-ended decoder for vision-centric tasks. Advances in Neural Information Processing Systems (NeurIPS) , 36, 2024

    work page 2024

  7. [5]

    MiniGPT-4: Enhancing vision-language understanding with advanced large language models

    Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed El- hoseiny. MiniGPT-4: Enhancing vision-language understanding with advanced large language models. In International Conference on Learning Representations (ICLR) , 2024

    work page 2024

  8. [6]

    Large language models for code: Security hardening and adversarial testing

    Jingxuan He and Martin Vechev. Large language models for code: Security hardening and adversarial testing. In ACM SIGSAC Conference on Computer and Communications Security (CCS) , pages 1865–1879, 2023

    work page 2023

Show all 52 references
  1. [7]

    Lost at C: A user study on the security implications of large language model code assistants

    Gustavo Sandoval, Hammond Pearce, Teo Nys, Ramesh Karri, Siddharth Garg, and Brendan Dolan-Gavitt. Lost at C: A user study on the security implications of large language model code assistants. In USENIX Security Symposium (USENIX Sec) , pages 2205–2222, 2023

    2023

  2. [8]

    Large language models encode clinical knowledge

    Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, et al. Large language models encode clinical knowledge. Nature, 620(7972):172–180, 2023

    2023

  3. [9]

    GPT4Graph: Can large language models understand graph structured data? an empirical evaluation and benchmarking

    Jiayan Guo, Lun Du, and Hengyu Liu. GPT4Graph: Can large language models understand graph structured data? an empirical evaluation and benchmarking. arXiv preprint arXiv:2305.15066 , 2023

    2023 arXiv

  4. [10]

    Beyond text: A deep dive into large language models’ ability on understanding graph data

    Yuntong Hu, Zheng Zhang, and Liang Zhao. Beyond text: A deep dive into large language models’ ability on understanding graph data. arXiv preprint arXiv:2310.04944, 2023

    2023 arXiv

  5. [11]

    Can language models solve graph problems in natural language? In Advances in Neural Information Processing Systems (NeurIPS), volume 36, pages 30840–30861, 2023

    Heng Wang, Shangbin Feng, Tianxing He, Zhaoxuan Tan, Xiaochuang Han, and Yulia Tsvetkov. Can language models solve graph problems in natural language? In Advances in Neural Information Processing Systems (NeurIPS), volume 36, pages 30840–30861, 2023

    2023

  6. [12]

    LLM4DyG: Can large language models solve spatial- temporal problems on dynamic graphs? In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) , page 4350–4361, 2024

    Zeyang Zhang, Xin Wang, Ziwei Zhang, Haoyang Li, Yijian Qin, and Wenwu Zhu. LLM4DyG: Can large language models solve spatial- temporal problems on dynamic graphs? In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) , page 4350–4361, 2024

    2024

  7. [14]

    Cryptocurrency in the aftermath: Unveiling the impact of the SVB collapse

    Qin Wang, Guangsheng Yu, and Shiping Chen. Cryptocurrency in the aftermath: Unveiling the impact of the SVB collapse. IEEE Transactions on Computational Social Systems (TCSS) , 2024

    2024

  8. [15]

    Miracle or mirage? a measurement study of NFT rug pulls

    Jintao Huang, Ningyu He, Kai Ma, Jiang Xiao, and Haoyu Wang. Miracle or mirage? a measurement study of NFT rug pulls. Proceedings of the ACM on Measurement and Analysis of Computing Systems (SIGMETRICS), 7(3):1–25, 2023

    2023

  9. [16]

    An empirical study on snapshot DAOs

    Qin Wang, Guangsheng Yu, Yilin Sai, Caijun Sun, Lam Duc Nguyen, Sherry Xu, and Shiping Chen. An empirical study on snapshot DAOs. arXiv preprint arXiv:2211.15993 , 2022

    2022 arXiv

  10. [17]

    Understanding Ethereum via graph analysis

    Ting Chen, Zihao Li, Yuxiao Zhu, Jiachi Chen, Xiapu Luo, John Chi-Shing Lui, Xiaodong Lin, and Xiaosong Zhang. Understanding Ethereum via graph analysis. ACM Transactions on Internet Technology (TOIT), 20(2):1–32, 2020

    2020

  11. [18]

    Tracking counterfeit cryptocurrency end-to-end

    Bingyu Gao, Haoyu Wang, Pengcheng Xia, Siwei Wu, Yajin Zhou, Xiapu Luo, and Gareth Tyson. Tracking counterfeit cryptocurrency end-to-end. Proceedings of the ACM on Measurement and Analysis of Computing Systems (SIGMETRICS) , 4(3):1–28, 2020

    2020

  12. [19]

    An empirical analysis of pool hopping behavior in the Bitcoin blockchain

    Natkamon Tovanich, Nicolas Souli ´e, Nicolas Heulot, and Petra Isen- berg. An empirical analysis of pool hopping behavior in the Bitcoin blockchain. In IEEE International Conference on Blockchain and Cryptocurrency (ICBC), 2021

    2021

  13. [20]

    A study on nine years of Bitcoin transactions: Understanding real-world behaviors of bitcoin miners and users

    Binbing Hou and Feng Chen. A study on nine years of Bitcoin transactions: Understanding real-world behaviors of bitcoin miners and users. In IEEE International Conference on Distributed Computing Systems (ICDCS), 2020

    2020

  14. [21]

    Double and nothing: Understanding and detecting cryptocurrency giveaway scams

    Xigao Li, Anurag Yepuri, and Nick Nikiforakis. Double and nothing: Understanding and detecting cryptocurrency giveaway scams. In Net- work and Distributed Systems Security (NDSS) Symposium , 2023

    2023

  15. [22]

    TxPhishScope: Towards detecting and understanding transaction-based phishing on Ethereum

    Bowen He, Yuan Chen, Zhuo Chen, Xiaohui Hu, Yufeng Hu, Lei Wu, Rui Chang, Haoyu Wang, and Yajin Zhou. TxPhishScope: Towards detecting and understanding transaction-based phishing on Ethereum. In ACM SIGSAC Conference on Computer and Communications Security (CCS), pages 120–134, 2023

    2023

  16. [23]

    Watch your back: Identifying cybercrime financial relationships in Bitcoin through back-and-forth exploration

    Gibran Gomez, Pedro Moreno-Sanchez, and Juan Caballero. Watch your back: Identifying cybercrime financial relationships in Bitcoin through back-and-forth exploration. In ACM SIGSAC Conference on Computer and Communications Security (CCS) , 2022

    2022

  17. [24]

    Toward understanding asset flows in crypto money laundering through the lenses of ethereum heists

    Jiajing Wu, Dan Lin, Qishuang Fu, Shuo Yang, Ting Chen, Zibin Zheng, and Bowen Song. Toward understanding asset flows in crypto money laundering through the lenses of ethereum heists. IEEE Transactions on Information Forensics and Security (TIFS) , 2024

    2024

  18. [25]

    Do cryptocurrency exchanges fake trading volumes? an empirical analysis of wash trading based on data mining

    Jialan Chen, Dan Lin, and Jiajing Wu. Do cryptocurrency exchanges fake trading volumes? an empirical analysis of wash trading based on data mining. Physica A: Statistical Mechanics and its Applications , 586:126405, 2022

    2022

  19. [26]

    Dissecting Bitcoin blockchain: Empirical analysis of Bitcoin network (2009–2020)

    Pranav Nerurkar, Dhiren Patel, Yann Busnel, et al. Dissecting Bitcoin blockchain: Empirical analysis of Bitcoin network (2009–2020). Journal of Network and Computer Applications , 2021

    2009

  20. [27]

    Cryptocurrencies activity as a complex network: Analysis of transactions graphs

    Luca Serena, Stefano Ferretti, and Gabriele D’Angelo. Cryptocurrencies activity as a complex network: Analysis of transactions graphs. Peer- to-Peer Networking and Applications , 2022

    2022

  21. [28]

    Complex network analysis of the Bitcoin transaction network

    Bishenghui Tao, Hong-Ning Dai, Jiajing Wu, Ivan Wang-Hei Ho, Zibin Zheng, and Chak Fong Cheang. Complex network analysis of the Bitcoin transaction network. IEEE Transactions on Circuits and Systems II: Express Briefs, 2021

    2021

  22. [29]

    Graph structure and statistical properties of Ethereum transaction relationships

    Dongchao Guo, Jiaqing Dong, and Kai Wang. Graph structure and statistical properties of Ethereum transaction relationships. Information Sciences, 2019

    2019

  23. [30]

    DenseFlow: Spotting cryptocurrency money laundering in Ethereum transaction graphs

    Dan Lin, Jiajing Wu, Yunmei Yu, Qishuang Fu, Zibin Zheng, and Changlin Yang. DenseFlow: Spotting cryptocurrency money laundering in Ethereum transaction graphs. In Proceedings of the ACM on Web Conference (WWW), 2024

    2024

  24. [31]

    To- wards malicious address identification in Bitcoin

    Deepesh Chaudhari, Rachit Agarwal, and Sandeep Kumar Shukla. To- wards malicious address identification in Bitcoin. In IEEE International Conference on Blockchain (Blockchain) , 2021

    2021

  25. [32]

    Mixers detection in Bitcoin network: a step towards detecting money laundering 13 in crypto-currencies

    M Mazhar Rathore, Sushil Chaurasia, and Dhirendra Shukla. Mixers detection in Bitcoin network: a step towards detecting money laundering 13 in crypto-currencies. In IEEE International Conference on Big Data (BigData), 2022

    2022

  26. [33]

    Improving cryptocurrency crime detection: Coinjoin community detec- tion approach

    Anton Wahrst ¨atter, Jor ˜ao Gomes, Sajjad Khan, and Davor Svetinovic. Improving cryptocurrency crime detection: Coinjoin community detec- tion approach. IEEE Transactions on Dependable and Secure Computing (TDSC), 2023

    2023

  27. [34]

    Anti-money laundering in Bitcoin: Experimenting with graph convolutional networks for financial forensics

    Mark Weber, Giacomo Domeniconi, Jie Chen, Daniel Karl I Weidele, Claudio Bellei, Tom Robinson, and Charles E Leiserson. Anti-money laundering in Bitcoin: Experimenting with graph convolutional networks for financial forensics. arXiv preprint arXiv:1908.02591 , 2019

    1908 arXiv

  28. [36]

    Blockchain large language models

    Yu Gai, Liyi Zhou, Kaihua Qin, Dawn Song, and Arthur Gervais. Blockchain large language models. arXiv preprint arXiv:2304.12749 , 2023

    2023 arXiv

  29. [37]

    A survey of graph meets large language model: Progress and future directions

    Yuhan Li, Zhixun Li, Peisong Wang, Jia Li, Xiangguo Sun, Hong Cheng, and Jeffrey Xu Yu. A survey of graph meets large language model: Progress and future directions. In International Joint Conference on Artificial Intelligence (IJCAI) , 2024

    2024

  30. [38]

    Predicting nft classification with GNN: A recommender system for web3 assets

    Guangsheng Yu, Qin Wang, Tanzeela Altaf, Xu Wang, Xiwei Xu, and Shiping Chen. Predicting nft classification with GNN: A recommender system for web3 assets. In IEEE International Conference on Blockchain and Cryptocurrency (ICBC) , 2023

    2023

  31. [39]

    Grapharena: Evaluating and exploring large language models on graph computation

    Jianheng Tang, Qifan Zhang, Yuhan Li, Nuo Chen, and Jia Li. Grapharena: Evaluating and exploring large language models on graph computation. In The Thirteenth International Conference on Learning Representations (ICLR), 2025

    2025

  32. [40]

    Table meets LLM: Can large language models understand structured table data? a benchmark and empirical study

    Yuan Sui, Mengyu Zhou, Mingjie Zhou, Shi Han, and Dongmei Zhang. Table meets LLM: Can large language models understand structured table data? a benchmark and empirical study. In ACM International Conference on Web Search and Data Mining (WSDM) , pages 645–654, 2024

    2024

  33. [41]

    StructGPT: A general framework for large language model to reason over structured data

    Jinhao Jiang, Kun Zhou, zican Dong, KeMing Ye, Xin Zhao, and Ji- Rong Wen. StructGPT: A general framework for large language model to reason over structured data. In The Conference on Empirical Methods in Natural Language Processing (EMNLP) , 2023

    2023

  34. [42]

    Which modality should I use - text, motif, or image? : Understanding graphs with large language models

    Debarati Das, Ishaan Gupta, Jaideep Srivastava, and Dongyeop Kang. Which modality should I use - text, motif, or image? : Understanding graphs with large language models. In Findings of the Association for Computational Linguistics (NAACL) , pages 503–519, 2024

    2024

  35. [43]

    Exploring the potential of large language models (LLMs) in learning on graphs

    Zhikai Chen, Haitao Mao, Hang Li, Wei Jin, Hongzhi Wen, Xiaochi Wei, Shuaiqiang Wang, Dawei Yin, Wenqi Fan, Hui Liu, et al. Exploring the potential of large language models (LLMs) in learning on graphs. ACM SIGKDD Explorations Newsletter , 25(2):42–61, 2024

    2024

  36. [44]

    Label-free node classification on graphs with large language models (LLMs)

    Zhikai Chen, Haitao Mao, Hongzhi Wen, Haoyu Han, Wei Jin, Haiyang Zhang, Hui Liu, and Jiliang Tang. Label-free node classification on graphs with large language models (LLMs). In The Twelfth International Conference on Learning Representations (ICLR) , 2024

    2024

  37. [45]

    Head-to-tail: How knowledgeable are large language models (LLMs)? a.k.a

    Kai Sun, Yifan Ethan Xu, Hanwen Zha, Yue Liu, and Xin Luna Dong. Head-to-tail: How knowledgeable are large language models (LLMs)? a.k.a. will LLMs replace knowledge graphs? In Proceedings of the NAACL-HLT, 2024

    2024

  38. [46]

    GPT-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. GPT-4 technical report. arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  39. [47]

    Large language models on graphs: A comprehensive survey

    Bowen Jin, Gang Liu, Chi Han, Meng Jiang, Heng Ji, and Jiawei Han. Large language models on graphs: A comprehensive survey. IEEE Transactions on Knowledge and Data Engineering (TKDE) , 2024

    2024

  40. [48]

    Graph summarization with controlled utility loss

    Mahdi Hajiabadi, Jasbir Singh, Venkatesh Srinivasan, and Alex Thomo. Graph summarization with controlled utility loss. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) , pages 536–546, 2021

    2021

  41. [49]

    Personalized graph summarization: formulation, scalable algorithms, and applications

    Shinhwan Kang, Kyuhan Lee, and Kijung Shin. Personalized graph summarization: formulation, scalable algorithms, and applications. In IEEE International Conference on Data Engineering (ICDE) , pages 2319–2332, 2022

    2022

  42. [50]

    Ssumm: Sparse summarization of massive graphs

    Kyuhan Lee, Hyeonsoo Jo, Jihoon Ko, Sungsu Lim, and Kijung Shin. Ssumm: Sparse summarization of massive graphs. In ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), pages 144–154, 2020

    2020

  43. [51]

    An optimized lossless graph summarization for large-scale graphs

    Meiquan Lai, Yaqi Huang, Zhidan Liu, and Kaishun Wu. An optimized lossless graph summarization for large-scale graphs. In IEEE Interna- tional Conference on Parallel and Distributed Systems (ICPADS), pages 355–362. IEEE, 2023

    2023

  44. [52]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems (NeurIPS) , volume 35, pages 24824– 24837, 2022

    2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.