pith. machine review for the scientific record. sign in

arxiv: 2604.12329 · v1 · submitted 2026-04-14 · 💻 cs.CR · cs.SI

Recognition: no theorem link

UniDetect: LLM-Driven Universal Fraud Detection across Heterogeneous Blockchains

Shuyi Miao , Wangjie Qiu , Shengda Zhuo , Fei Shen , Dan Lin , Xingtong Yu , Chua Tat-Seng , Zhiming Zheng
Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:57 UTC · model grok-4.3

classification 💻 cs.CR cs.SI
keywords fraud detectionblockchainlarge language modelscross-chain detectioncryptocurrencymultimodal learningzero-shot transfer
0
0 comments X

The pith

LLMs guided by domain knowledge generate general transaction summaries to detect fraud across different blockchains.

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

The paper presents UniDetect, a method that directs large language models with domain knowledge to produce transaction summary texts usable on any blockchain. These summaries act as evidence alongside transaction graph patterns, with a two-stage alternating training process that refines the joint reasoning for fraud identification. The approach targets cross-chain DeFi flows that currently evade single-chain monitoring. Experiments report consistent gains over prior methods on multiple chains, strong zero-shot transfer to unseen chains, and some generalization to non-blockchain datasets.

Core claim

UniDetect is a multi-chain cryptocurrency fraud account detection method based on large language models. Domain knowledge guides the LLM to generate general transaction summary texts applicable to heterogeneous blockchain accounts, which serve as evidence for fraud account detection. A two-stage alternating training strategy continuously and dynamically enhances the multimodal joint reasoning for detecting fraudulent accounts based on both the textual evidence and the transaction graph patterns.

What carries the argument

Domain-knowledge-guided LLM that produces general transaction summary texts, paired with two-stage alternating training that alternates between textual evidence and graph patterns for multimodal fraud classification.

If this is right

  • Outperforms existing methods by 5.57% to 7.58% in Kolmogorov-Smirnov statistic on multiple blockchains.
  • Identifies over 94.58% of fraudulent accounts in cross-chain zero-shot detection scenarios.
  • Delivers a 6.06% F1 improvement when applied to non-blockchain data.
  • Supports monitoring of DeFi protocols that reorganize illicit funds into uniform assets across chains.

Where Pith is reading between the lines

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

  • The summaries could serve as a bridge for integrating fraud signals from blockchains with traditional financial transaction logs.
  • If the alternating training stabilizes, similar LLM-plus-graph pipelines might apply to other heterogeneous network fraud problems such as social media account abuse.
  • Broader deployment would require testing whether performance holds when the underlying LLM is updated or replaced.

Load-bearing premise

Domain-knowledge-guided LLM prompts can reliably produce general, unbiased transaction summary texts that serve as effective evidence for fraud detection across truly heterogeneous blockchains without hallucinations or chain-specific artifacts.

What would settle it

Running the method on a previously unseen blockchain where the generated summaries contain detectable chain-specific phrasing or errors that cause detection accuracy to drop below baseline levels.

Figures

Figures reproduced from arXiv: 2604.12329 by Chua Tat-Seng, Dan Lin, Fei Shen, Shengda Zhuo, Shuyi Miao, Wangjie Qiu, Xingtong Yu, Zhiming Zheng.

Figure 1
Figure 1. Figure 1: A workflow comparison between chain-specific blockchain [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The overall framework of UniDetect. During training, we construct account-centered subgraphs, generate transaction summaries via [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Prompt template of forensics analyst agent. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Prompt template of discriminative and residual summary [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Experimental results of UniDetect’s generalization abil [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Hyperparameter sensitivity analysis on the Ethereum dataset reveals that UniDetect maintains stable and competitive performance [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: CoT prompt template of the forensics analyst agent. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: CoT prompt template of discriminative summary analyst [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: RL reward convergence curve during LLM fine-tuning on [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
read the original abstract

As cross-chain interoperability advances, decentralized finance (DeFi) protocols enable illicit funds to be reorganized into uniform liquid assets that flow throughout the cryptocurrency market. Such operations can bypass monitoring targeted at individual blockchains and thereby weaken current regulatory frameworks. Motivated by these, we introduce UniDetect, a multi-chain cryptocurrency fraud account detection method based on large language models (LLMs). Specifically, we use domain knowledge to guide the LLM to generate general transaction summary texts applicable to heterogeneous blockchain accounts, which serve as evidence for fraud account detection. Furthermore, we introduce a two-stage alternating training strategy to continuously and dynamically enhance the multimodal joint reasoning for detecting fraudulent accounts based on both the textual evidence and the transaction graph patterns. Experiments on multiple blockchains show that UniDetect outperforms existing methods 5.57% to 7.58% in Kolmogorov-Smirnov (KS). For cross-chain zero-shot detection, UniDetect identifies over 94.58% of fraudulent accounts. It also generalizes well to non-blockchain data, delivering a 6.06% improvement in F1 over existing methods. The dataset and source code are available at https://github.com/msy0513/UniDetect.

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

3 major / 2 minor

Summary. The paper presents UniDetect, an LLM-driven framework for universal fraud account detection across heterogeneous blockchains. Domain knowledge guides LLMs to produce general transaction summary texts as evidence; these are fused with graph patterns via a two-stage alternating training procedure. Reported results include 5.57–7.58% KS gains over baselines on multiple chains, >94.58% recall in cross-chain zero-shot settings, and 6.06% F1 improvement on non-blockchain data, with code and data released.

Significance. If validated, the approach would offer a practical step toward cross-chain monitoring in DeFi by leveraging textual evidence to bridge heterogeneous transaction graphs. The open release of dataset and code strengthens reproducibility and enables follow-on work on multimodal fraud detection.

major comments (3)
  1. [Section 3] Section 3 (method): the central performance claims rest on LLM-generated transaction summaries serving as reliable, general evidence. No quantitative validation is provided (hallucination rates, inter-chain consistency metrics, human evaluation, or automated checks), nor is there an ablation that isolates the contribution of the text modality versus graph patterns alone. Without these, the reported KS and zero-shot gains cannot be attributed to the claimed universal mechanism rather than chain-specific artifacts or LLM biases.
  2. [Experiments] Experimental section: the abstract and results report concrete KS/F1 improvements and zero-shot recall, yet the manuscript provides insufficient detail on baseline implementations, data splits, statistical significance testing (e.g., paired t-tests or multiple-run variance), and whether post-hoc prompt or hyperparameter choices were tuned on the test sets. This undermines confidence that the 5.57–7.58% margins are robust.
  3. [Section 3.2] Two-stage training: the alternating optimization between textual and graph modalities introduces additional hyperparameters (alternation schedule, fusion weights) whose sensitivity is not analyzed. If these are tuned per chain, the universality claim is weakened.
minor comments (2)
  1. [Section 3.1] Notation for the multimodal fusion module is introduced without a clear equation or diagram showing how textual embeddings are aligned with graph node features.
  2. [Figures 4-6] Figure captions and axis labels in the results section could be expanded to indicate exact dataset sizes and number of runs per bar.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to strengthen validation, experimental details, and analysis of the training procedure.

read point-by-point responses

  1. Referee: [Section 3] Section 3 (method): the central performance claims rest on LLM-generated transaction summaries serving as reliable, general evidence. No quantitative validation is provided (hallucination rates, inter-chain consistency metrics, human evaluation, or automated checks), nor is there an ablation that isolates the contribution of the text modality versus graph patterns alone. Without these, the reported KS and zero-shot gains cannot be attributed to the claimed universal mechanism rather than chain-specific artifacts or LLM biases.

    Authors: We agree that additional quantitative validation would strengthen attribution of the gains to the universal textual mechanism. In the revised manuscript, we will add a human evaluation of a sampled set of LLM-generated summaries, reporting hallucination rates and inter-chain consistency metrics. We will also include an ablation study comparing graph-only performance against the full multimodal model to isolate the text modality's contribution. revision: yes

  2. Referee: [Experiments] Experimental section: the abstract and results report concrete KS/F1 improvements and zero-shot recall, yet the manuscript provides insufficient detail on baseline implementations, data splits, statistical significance testing (e.g., paired t-tests or multiple-run variance), and whether post-hoc prompt or hyperparameter choices were tuned on the test sets. This undermines confidence that the 5.57–7.58% margins are robust.

    Authors: We will expand the experimental section with full details on baseline implementations and their adaptations to our multi-chain setting. Data splits will be explicitly described, and we will report results with means and standard deviations over multiple runs along with paired t-test significance. We confirm and will clarify that all prompt and hyperparameter tuning occurred on held-out validation sets with no test-set access. revision: yes

  3. Referee: [Section 3.2] Two-stage training: the alternating optimization between textual and graph modalities introduces additional hyperparameters (alternation schedule, fusion weights) whose sensitivity is not analyzed. If these are tuned per chain, the universality claim is weakened.

    Authors: We will add a sensitivity analysis subsection evaluating performance stability across ranges of the alternation schedule and fusion weights. The hyperparameters follow a single validation-based selection procedure applied uniformly to all chains; they are not tuned individually per chain. The added analysis will demonstrate robustness and support the universality claim. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical performance claims rest on held-out experiments, not self-referential derivations

full rationale

The paper introduces an LLM-guided method for generating transaction summaries followed by two-stage multimodal training, then reports concrete empirical metrics (KS gains of 5.57–7.58%, 94.58% zero-shot recall, 6.06% F1 on external data) evaluated on held-out blockchain and non-blockchain datasets. No equations, fitted parameters, or uniqueness theorems are presented whose outputs reduce by construction to the inputs; the central results are standard supervised-learning performance numbers whose validity depends on data splits and evaluation protocols rather than definitional equivalence or self-citation chains.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The approach rests on the assumption that LLMs can be prompted with domain knowledge to produce generalizable transaction summaries and that the two-stage training will yield robust multimodal reasoning without overfitting to the training chains.

free parameters (2)
  • LLM prompt design and domain knowledge injection
    Prompt templates and the specific domain knowledge used to guide summary generation are chosen by the authors and directly affect the textual evidence quality.
  • Two-stage training hyperparameters and alternation schedule
    Learning rates, batch sizes, and the exact alternation strategy between text and graph stages are tuned and affect the final joint reasoning performance.
axioms (2)
  • domain assumption LLM-generated transaction summaries are accurate, unbiased, and general across heterogeneous blockchains
    This is invoked when the paper states that the summaries serve as evidence for fraud detection.
  • domain assumption Transaction graph patterns provide complementary signal to the textual summaries
    Assumed when the multimodal joint reasoning is presented as the key mechanism.

pith-pipeline@v0.9.0 · 5533 in / 1496 out tokens · 67303 ms · 2026-05-10T15:57:02.421410+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

47 extracted references · 9 canonical work pages · 5 internal anchors

  1. [1]

    Scott Beamer, Krste Asanović, and David Patterson. 2013. Direction-optimizing breadth-first search.Scientific Programming21, 3-4 (2013), 137–148

    2013

  2. [2]

    Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. 2014. Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555(2014)

    work page internal anchor Pith review arXiv 2014

  3. [3]

    Zhihao Ding, Jieming Shi, Qing Li, and Jiannong Cao. 2024. Effective illicit account detection on large cryptocurrency multigraphs. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 457–466

    2024

  4. [4]

    Yingtong Dou, Zhiwei Liu, Li Sun, Yutong Deng, Hao Peng, and Philip S Yu. 2020. Enhancing graph neural network-based fraud detectors against camouflaged fraudsters. InProceedings of the 29th ACM international conference on information & knowledge management. 315–324

    2020

  5. [5]

    Mingjiang Duan, Tongya Zheng, Yang Gao, Gang Wang, Zunlei Feng, and Xinyu Wang. 2024. Dga-gnn: Dynamic grouping aggregation gnn for fraud detection. In Proceedings of the AAAI conference on artificial intelligence, Vol. 38. 11820–11828

    2024

  6. [6]

    Elliptic. 2025. Chain-hopping Emerges as the Defining Money Laundering Method of 2025. https://www.elliptic.co/blog/chain-hopping-defining-money- laundering-method-of-2025

    2025

  7. [7]

    Sam Gilbert. 2022. Crypto, web3, and the Metaverse.Bennett Institute for Public Policy, Cambridge, Policy Brief(2022)

    2022

  8. [8]

    Aditya Grover and Jure Leskovec. 2016. node2vec: Scalable feature learning for networks. InProceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining. 855–864

    2016

  9. [9]

    Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive Representation Learning on Large Graphs. InAdvances in Neural Information Processing Systems, I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.), Vol. 30. Curran Associates, Inc

    2017

  10. [10]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. 2022. Lora: Low-rank adaptation of large language models.ICLR1, 2 (2022), 3

    2022

  11. [11]

    Sihao Hu, Zhen Zhang, Bingqiao Luo, Shengliang Lu, Bingsheng He, and Ling Liu. 2023. Bert4eth: A pre-trained transformer for ethereum fraud detection. In Proceedings of the ACM Web Conference 2023. 2189–2197

    2023

  12. [12]

    Xuanwen Huang, Kaiqiao Han, Yang Yang, Dezheng Bao, Quanjin Tao, Ziwei Chai, and Qi Zhu. 2024. Can gnn be good adapter for llms?. InProceedings of the ACM Web Conference 2024. 893–904

    2024

  13. [13]

    Semi-Supervised Classification with Graph Convolutional Networks

    Thomas N. Kipf and Max Welling. 2016. Semi-Supervised Classification with Graph Convolutional Networks.CoRRabs/1609.02907 (2016). arXiv:1609.02907

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  14. [14]

    Johannes Klicpera, Aleksandar Bojchevski, and Stephan Günnemann. 2018. Per- sonalized embedding propagation: Combining neural networks on graphs with personalized pagerank.CoRR, abs/1810.059973 (2018)

    work page arXiv 2018

  15. [15]

    Dan Lin, Yanli Ding, Weipeng Zou, Jiachi Chen, Xiapu Luo, Jiajing Wu, and Zibin Zheng. 2025. RiskTagger: An LLM-based Agent for Automatic Annotation of Web3 Crypto Money Laundering Behaviors.arXiv preprint arXiv:2510.17848 (2025)

    work page arXiv 2025

  16. [16]

    Liheng Ma, Chen Lin, Derek Lim, Adriana Romero-Soriano, Puneet K Dokania, Mark Coates, Philip Torr, and Ser-Nam Lim. 2023. Graph inductive biases in transformers without message passing. InInternational Conference on Machine Learning. PMLR, 23321–23337

    2023

  17. [17]

    Shuyi Miao, Wangjie Qiu, Xiaofan Tu, Yunze Li, Yongxin Wen, and Zhiming Zheng. 2026. Tracing Your Account: A Gradient-Aware Dynamic Window Graph Framework for Ethereum under Privacy-Preserving Services.IEEE Transactions on Information Forensics and Security(2026)

    2026

  18. [18]

    Shuyi Miao, Wangjie Qiu, Hongwei Zheng, Qinnan Zhang, Xiaofan Tu, Xunan Liu, Yang Liu, Jin Dong, and Zhiming Zheng. 2025. Know Your Account: Double Graph Inference-based Account De-anonymization on Ethereum.2025 IEEE 41rd International Conference on Data Engineering (ICDE)(2025)

    2025

  19. [19]

    Dang Nguyen, Tu Dinh Nguyen, Wei Luo, and Svetha Venkatesh. 2018. Trans2vec: learning transaction embedding via items and frequent itemsets. InPacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 361–372

    2018

  20. [20]

    Bryan Perozzi, Rami Al-Rfou, and Steven Skiena. 2014. Deepwalk: Online learning of social representations. InProceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. 701–710

    2014

  21. [21]

    Farimah Poursafaei, Reihaneh Rabbany, and Zeljko Zilic. 2021. Sigtran: signature vectors for detecting illicit activities in blockchain transaction networks. In Pacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 27– 39

    2021

  22. [22]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks.arXiv preprint arXiv:1908.10084(2019)

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  23. [23]

    Jie Shen, Jiajun Zhou, Yunyi Xie, Shanqing Yu, and Qi Xuan. 2021. Identity inference on blockchain using graph neural network. InInternational conference on blockchain and trustworthy systems. Springer, 3–17

    2021

  24. [24]

    Jianheng Tang, Jiajin Li, Ziqi Gao, and Jia Li. 2022. Rethinking graph neural networks for anomaly detection. InInternational conference on machine learning. PMLR, 21076–21089

    2022

  25. [25]

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupati- raju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, et al. 2024. Gemma: Open models based on gemini research and technology. arXiv preprint arXiv:2403.08295(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  26. [26]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

    2017

  27. [27]

    Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, Yoshua Bengio, et al. 2017. Graph attention networks.stat1050, 20 (2017), 10–48550

    2017

  28. [28]

    Jinhuan Wang, Pengtao Chen, Xinyao Xu, Jiajing Wu, Meng Shen, Qi Xuan, and Xiaoniu Yang. 2025. Tsgn: Transaction subgraph networks assisting phishing detection in ethereum.IEEE Transactions on Dependable and Secure Computing (2025)

    2025

  29. [29]

    Ronald J Williams. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning.Machine learning8, 3 (1992), 229–256

    1992

  30. [30]

    Xihan Xiong, Zhipeng Wang, Qin Wang, Endong Liu, Pascal Berrang, and William Knottenbelt. 2025. Talking Transactions: Decentralized Communication through Ethereum Input Data Messages (IDMs). arXiv:2505.24724 [cs.CR] https://arxiv. org/abs/2505.24724

    work page arXiv 2025

  31. [31]

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2018. How Powerful are Graph Neural Networks?CoRRabs/1810.00826 (2018). arXiv:1810.00826

    work page internal anchor Pith review arXiv 2018

  32. [32]

    Chengdong Yang, Hongrui Liu, Daixin Wang, Zhiqiang Zhang, Cheng Yang, and Chuan Shi. 2025. Flag: Fraud detection with llm-enhanced graph neural network. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. 5150–5160. Conference’17, July 2017, Washington, DC, USA Miao et al

    2025

  33. [33]

    Haibin Zheng, Minying Ma, Haonan Ma, Jinyin Chen, Haiyang Xiong, and Zhijun Yang. 2023. Tegdetector: a phishing detector that knows evolving transaction behaviors.IEEE Transactions on Computational Social Systems11, 3 (2023), 3988– 4000

    2023

  34. [34]

    Jiajun Zhou, Chenkai Hu, Jianlei Chi, Jiajing Wu, Meng Shen, and Qi Xuan. 2022. Behavior-aware account de-anonymization on Ethereum interaction graph.IEEE Transactions on Information Forensics and Security17 (2022), 3433–3448

    2022

  35. [35]

    Wei Zhuo, Zemin Liu, Bryan Hooi, Bingsheng He, Guang Tan, Rizal Fathony, and Jia Chen. 2024. Partitioning message passing for graph fraud detection.arXiv preprint arXiv:2412.00020(2024). UniDetect: LLM-Driven Universal Fraud Detection across Heterogeneous Blockchains Conference’17, July 2017, Washington, DC, USA Supplementary Material Outline.This supplem...

    work page arXiv 2024

  36. [36]

    Task Description You are a blockchain behavior analyst tasked with analyzing a account's transaction of [The specific chain name] (e.g., Ethereum) patterns to produce a professional, engaging, and cohesive narrative

  37. [37]

    Output Constraint The output should be suitable for generating rich text embeddings or supporting classification tasks

  38. [38]

    Chain-of-Thought Instructions (CoT)

  39. [39]

    Analyze Transaction Data:  Value Flow: Assess whether the account primarily acts as a net receiver, net distributor, or remains balanced based on overall incoming and outgoing value. Describe the scale of value transfers (e.g., large if total_value_out > 1000 ETH, small if < 100 ETH);  Partner Ratio: Compare incoming and outgoing counterparties to ident...

  40. [40]

    Generate Narrative: Produce a 3-5 sentence paragraph that weaves together insights on value flow, partner ratios, transaction timing, and Gas expenditure

  41. [41]

    2015-08-07

    Data Preparation Provided Data: [Transactions] Prompt for Forensics Analyst Agent Figure 7: CoT prompt template of the forensics analyst agent. Ethereum.We obtain the fraud labels from the Etherscan label cloud6, XLabelCloud7 and CryptoScamDB8. These labels cover ma- jor illegal account types such as phishing, Upbit vulnerability at- tacks, gambling, and ...

    2015

  42. [42]

    Task Description For each transaction summary, identify the MOST distinguishing causal pattern that indicates whether the address is normal or fraudulent (e.g., phishing, scam)

  43. [43]

    • Fund Flows: Aggregation (many sources to one address) or Dispersion (one address to many targets), layered transfers via intermediates

    Analysis Framework: • Transaction Patterns: High-frequency/large-value transfers, amounts near regulatory thresholds(e.g., < $50k if the reporting limit is $50k), round-number transfers (e.g., 1,000/10,000 units) without business logic, or self-transfers/reversals. • Fund Flows: Aggregation (many sources to one address) or Dispersion (one address to many ...

  44. [44]

    Output Generate ONE sentence per address. Prompt for Discriminative Summary Analyst Agent Role definition You are provided with a list of [The specific chain name] (e.g., Ethereum) accounts' historical transaction summaries. Each account is classified as either normal or fraudulent (e.g., associated with phishing, scams, or other illicit activities) based...

  45. [45]

    These sentences must present objective, raw data points, without any assessment of normal, fraudulent, or significance

    Task Description For each transaction summary, Write ONE factual sentence. These sentences must present objective, raw data points, without any assessment of normal, fraudulent, or significance

  46. [46]

    Analysis Framework • Noise Extraction: Exclude unsupported or unverifiable statements (e.g., speculative risk assertions and fabricated associations), as well as task- irrelevant generation noise such as prompt echoing and redundant content. • Allowed Factual Data Points (Strictly Quantitative and Objective): Retain only descriptive, verifiable statements...

  47. [47]

    Prompt for Residual Summary Analyst Agent Figure 8: CoT prompt template of discriminative summary analyst agent and residual summary analyst agent

    Output Generate ONE sentence per address. Prompt for Residual Summary Analyst Agent Figure 8: CoT prompt template of discriminative summary analyst agent and residual summary analyst agent. edges, to enable a more comprehensive evaluation of our model. (1)Bitcoin-Mcontains the first 1.5 million transactions before June 2015. For each transaction edge in B...

    2015