基于图神经网络的账户余额模型区块链地址分类方法

doi:10.11959/j.issn.1000-436x.2023173

通信学报 ›› 2023, Vol. 44 ›› Issue (9): 115-126.doi: 10.11959/j.issn.1000-436x.2023173

• 学术论文 • 上一篇

基于图神经网络的账户余额模型区块链地址分类方法

李致远¹^,²^,³, 徐丙磊¹, 周颖仪¹

¹ 江苏大学计算机科学与通信工程学院，江苏镇江 212013
² 江苏省工业网络安全技术重点实验室，江苏镇江 212013
³ 江苏省泛在数据智能感知与分析应用工程研究中心，江苏镇江 212013

修回日期:2023-08-31 出版日期:2023-09-01 发布日期:2023-09-01
作者简介:李致远（1981- ），男，河南开封人，博士，江苏大学副教授、硕士生导师，主要研究方向为区块链匿名交易追踪、物联网和软件定义网络及安全
徐丙磊（1997- ），男，山东莱阳人，江苏大学硕士生，主要研究方向为区块链匿名可追踪、区块链地址分类
周颖仪（2000- ），女，江苏苏州人，江苏大学硕士生，主要研究方向为区块链交易网络构建与链路预测
基金资助:
国家重点研发计划基金资助项目(2020YFB1005503);江苏省自然科学基金资助项目(BK20201415)

Graph neural network-based address classification method for account balance model blockchain

Zhiyuan LI¹^,²^,³, Binglei XU¹, Yingyi ZHOU¹

¹ Jiangsu University, School of Computer Science and Communication Engineering, Zhenjiang 212013, China
² Jiangsu Provincial Key Laboratory of Industrial Network Security Technology, Zhenjiang 212013, China
³ Jiangsu Province Ubiquitous Data Intelligent Perception and Analysis Application Engineering Research Center, Zhenjiang 212013, China

Revised:2023-08-31 Online:2023-09-01 Published:2023-09-01
Supported by:
The National Key Research and Development Program of China(2020YFB1005503);The Natural Science Foundation of Jiangsu Province(BK20201415)

摘要/Abstract

摘要：

为了监管账户余额模型公链上的交易，有必要对该类区块链上的交易进行地址分类研究。基于此，提出了一种基于图神经网络的账户余额模型区块链地址分类方法（简称 AJKGS-ABCM）以实现区块链地址的分类，为区块链交易追踪提供有效的支持。该方法将区块链交易数据建模为图结构，以地址为节点，交易为边，提出AJK-GraphSAGE 算法学习图的嵌入表示，模型的输入只需要节点及其采样的邻居节点集合。同时，模型引入注意力机制及跳跃知识结合策略，自适应地为不同层的表示分配权重，并在不同层间共享信息，提高了训练速度和泛化能力。最后进行了实验对比，结果表明该模型在准确度、召回率和F1分数上性能优于其他方法。

关键词: 账户余额模型区块链, 地址分类, 图神经网络, 注意力机制, 跳跃知识

Abstract:

To regulate the transactional activities on the public blockchain involving account balance models, it is necessary to conduct research on address classification for transactions on such blockchains.A blockchain address classification method, named AJKGS-ABCM (attention jumping knowledge graph SAGE account-based blockchain classification model), was proposed to categorize blockchain addresses, providing effective support for blockchain transaction tracking.Blockchain transaction data was represented as a graph structure, with addressed as nodes and transactions as edges.The AJK-GraphSAGE algorithm was introduced to learn embedded representations of the graph, where the model’s input required only nodes and their sampled neighboring node sets.Simultaneously, attention mechanisms and skip-connection knowledge integration strategies were incorporated into the model, allowing for adaptive weight allocation across different layers and information sharing between various levels, thereby enhancing training speed and generalization capabilities.Finally, experimental comparisons are conducted, demonstrating superior performance in terms of accuracy, recall, and F1 score compared to other methods.

Key words: account balance model blockchain, address classification, graph neural network, attention mechanism, jumping knowledge

中图分类号:

TP301

李致远, 徐丙磊, 周颖仪. 基于图神经网络的账户余额模型区块链地址分类方法[J]. 通信学报, 2023, 44(9): 115-126.

Zhiyuan LI, Binglei XU, Yingyi ZHOU. Graph neural network-based address classification method for account balance model blockchain[J]. Journal on Communications, 2023, 44(9): 115-126.

图/表 13

图1

表1

参数含义"

参数	含义
N	节点的数量
$A$	图的邻接矩阵，其中A_ij 表示节点i和节点 j之间是否有边
$X$	节点特征矩阵，其中 $x_{i}$ 表示节点i的特征向量
k	GraphSAGE的层数，k=1,2,…,K，K为总层数
C	分类类别的数量，其中c=1,2,…,C表示每个类别
Θ	模型参数集合，包括GraphSAGE层的权重矩阵 $W^{1}, W^{2}, \dots, W^{K}$ 、注意力权重矩阵 $W_{a}$ 和分类层权重矩阵 $W_{c}$
η	学习率，用于随机梯度下降算法
$h_{i}^{k}$	节点v_i 在第k层的表示
$N (v_{i})$	节点v_i 的邻居节点集合
$H_{}^{1}, H_{}^{2}, \dots, H_{}^{K}$	经过K个GraphSAGE层的输出表示矩阵
$A_{_{i}}^{k}$	节点v_i 的注意力权重向量，表示第k层GraphSAGE层输出表示的重要性
$H_{i}^{final}$	节点v_i 经过注意力和跳跃知识结合策略处理后的最终表示
O	分类层的输出
$P$	类别概率矩阵，其元素P_i,c表示节点v_i 属于类别c 的概率
$Y$	节点的真实类别标签矩阵，表示每个节点的独热编码类别标签
L	交叉熵损失函数值，用于衡量模型预测与真实类别标签之间的一致性
$\nabla_{Θ} L$	损失函数关于模型参数的梯度

表1

图2

图3

图4

表2

表3

实验评估指标"

指标	计算式	含义
Precision	$P = \frac{TP}{TP+FP}$	被正确分类为某标签的地址占总的被分类为某标签的地址的比例
Recall	$R (TPR) = \frac{TP}{TP+FN}$	所有正例中被正确分类的比例
F1-score	$F1-score = 2 \frac{P R}{P + R}$	兼顾Precision和Recall的指标
micro-F1	$micro-F1 = \frac{\sum_{i = 1}^{C} 2 {TP}_{i}}{\sum_{i = 1}^{C} 2 {TP}_{i} + {FP}_{i} + {FN}_{i}}$	先计算所有类别的总的Precision和Recall，后计算F1值，即micro-F1
macro-F1	$macro-F1 = \frac{1}{C} \sum_{i = 1}^{C} \frac{2 {TP}_{i}}{2 {TP}_{i} + {FP}_{i} + {FN}_{i}}$	将所有类别的Precision和Recall求平均，然后计算F1值，即macro-F1

表3

表4

表5

表6

表7

图5

图6

参考文献 23

[1]	BENDIAB G , HAMEURLAINE A , GERMANOS G ,et al. Autonomous vehicles security:challenges and solutions using blockchain and artificial intelligence[J]. IEEE Transactions on Intelligent Transportation Systems, 2023,24(4): 3614-3637.
[2]	FENG Q , HE D , ZEADALLY S ,et al. A survey on privacy protection in blockchain system[J]. Journal of Network and Computer Applications, 2019,126: 45-58.
[3]	HATHALIYA J J , MODI H , GUPTA R ,et al. Deep learning and blockchain-based essential and parkinson tremor classification scheme[C]// Proceedings of 2022 IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS). Piscataway:IEEE Press, 2022: 1-6.
[4]	LI Z Y , HE E H . Graph neural network-based bitcoin transaction tracking model[J]. IEEE Access, 2023,11: 62109-62120.
[5]	MONAMO P , MARIVATE V , TWALA B . Unsupervised learning for robust Bitcoin fraud detection[C]// Proceedings of Information Security for South Africa (ISSA). Piscataway:IEEE Press, 2017: 129-134.
[6]	TOYODA K , OHTSUKI T , MATHIOPOULOS P T . Multi-class bitcoin-enabled service identification based on transaction history summarization[C]// Proceedings of IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber,Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData). Piscataway:IEEE Press, 2019: 1153-1160.
[7]	LIN Y J , WU P W , HSU C H ,et al. An evaluation of bitcoin address classification based on transaction history summarization[C]// Proceedings of 2019 IEEE International Conference on Blockchain and Cryptocurrency (ICBC). Piscataway:IEEE Press, 2019: 302-310.
[8]	BARTOLETTI M , PES B , SERUSI S . Data mining for detecting bitcoin ponzi schemes[C]// Proceedings of 2018 Crypto Valley Conference on Blockchain Technology (CVCBT). Piscataway:IEEE Press, 2018: 75-84.
[9]	LI Y , CAI Y , TIAN H ,et al. Identifying illicit addresses in bitcoin network[C]// International Conference on Blockchain and Trustworthy Systems. Singapore:Springer, 2020: 99-111.
[10]	HU T , LIU X , CHEN T ,et al. Transaction-based classification and detection approach for Ethereum smart contract[J]. Information Processing ＆ Management, 2021,58(2): 102462.
[11]	WU J J , YUAN Q , LIN D ,et al. Who are the phishers？ phishing scam detection on ethereum via network embedding[J]. IEEE Transactions on Systems,Man,and Cybernetics:Systems, 2022,52(2): 1156-1166.
[12]	CHEN W , GUO X , CHEN Z ,et al. Phishing scam detection on ethereum:towards financial security for blockchain ecosystem[C]// Proceedings of International Joint Conference on Artificial Intelligence. Piscataway:IEEE Press, 2020: 4456-4462.
[13]	CHEN L , PENG J Y , LIU Y ,et al. Phishing scams detection in ethereum transaction network[J]. ACM Transactions on Internet Technology, 2021,21(1): 1-16.
[14]	KE G , MENG Q , FINLEY T ,et al. Lightgbm:a highly efficient gradient boosting decision tree[C]// Advances in Neural Information Processing Systems. San Francisco:Morgan Kaufmann Press, 2017:30.
[15]	YUAN Q , HUANG B Y , ZHANG J ,et al. Detecting phishing scams on ethereum based on transaction records[C]// Proceedings of 2020 IEEE International Symposium on Circuits and Systems (ISCAS). Piscataway:IEEE Press, 2020: 1-5.
[16]	ZHOU J , HU C , GONG S ,et al. BlockGC:a joint learning framework for account identity inference on blockchain with graph contrast[J]. arXiv Preprint,arXiv:2112.03659, 2021.
[17]	LIU J L , ZHENG J T , WU J J ,et al. FA-GNN:filter and augment graph neural networks for account classification in ethereum[J]. IEEE Transactions on Network Science and Engineering, 2022,9(4): 2579-2588.
[18]	KIPF T N , WELLING M . Semi-supervised classification with graph convolutional networks[J]. arXiv Preprint,arXiv:1609.02907, 2016.
[19]	HAMILTON W L , YING R , LESKOVEC J . Inductive representation learning on large graphs[J]. arXiv Preprint,arXiv:1706.02216, 2017.
[20]	GUO G D , WANG H , BELL D A ,et al. KNN model-based approach in classification[C]// Proceedings of OTM Confederated International Conferences.[S.l.:s.n.], 2003: 986-996.
[21]	PEROZZI B , AL-RFOU R , SKIENA S . DeepWalk:online learning of social representations[C]// Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York:ACM Press, 2014: 701-710.
[22]	GROVER A , LESKOVEC J . Node2Vec:scalable feature learning for networks[J]. KDD:Proceedings International Conference on Knowledge Discovery ＆ Data Mining, 2016,2016: 855-864.
[23]	RIBEIRO L F R , SAVERESE P H P , FIGUEIREDO D R . Struc2Vec:learning node representations from structural identity[C]// Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York:ACM Press, 2017: 385-394.

标签名称	标签符号	标签描述
交易所	Exchange	该地址属于加密货币交易所，可能涉及加密货币的买卖交易和其他金融服务
矿池/矿工	Mining_pool/Mining	该地址是由一个或多个矿工汇集成的矿池地址，可能涉及加密货币挖矿等活动
博彩	Gambling	该地址可能是在线博彩网站的地址，可能涉及加密货币博彩活动
合约代币	Token	该地址属于一个以太坊代币合约，可能用于代币发行、代币交易等活动
暗网交易	Darknet	该地址可能是一个用于在暗网进行加密货币交易的地址，可能涉及非法交易
钱包	Wallet	该地址可能属于一个以太坊钱包，可能用于存储、管理和转移加密货币
欺诈	Scam	该地址可能是用于进行欺诈活动的地址，可能涉及虚假交易、社工钓鱼、伪造代币等活动
筹资活动	Ico	该地址可能是用于Ico或其他筹资活动的地址，可能涉及投资、筹款等活动
黑客	Hacker	该地址可能是用于黑客盗币活动的地址，可能涉及攻击以太坊网络、其他区块链网络或其他网络的活动
货币清洗	Mixer	该地址可能是用于混合加密货币交易的地址，可能涉及加密货币的匿名化处理
普通用户	User	该地址属于一般的以太坊用户，可能仅用于普通的加密货币交易等活动

方法	Exchange	Mining	Gambling	Token	Darknet	Wallet	Scam	Ico	Hacker	Mixer	User
KNN	0.683	0.675	0.651	0.643	0.648	0.656	0.651	0.624	0.638	0.671	0.661
Deepwalk	0.722	0.715	0.708	0.707	0.701	0.725	0.695	0.727	0.711	0.733	0.728
Node2Vec	0.735	0.731	0.715	0.722	0.719	0.731	0.718	0.734	0.724	0.740	0.723
Struc2Vec	0.741	0.735	0.724	0.731	0.730	0.745	0.729	0.738	0.733	0.744	0.739
I2BGNN	0.831	0.812	0.819	0.826	0.822	0.842	0.833	0.840	0.832	0.829	0.841
GCN	0.821	0.812	0.810	0.809	0.791	0.822	0.811	0.819	0.801	0.820	0.815
GraphSAGE	0.843	0.821	0.819	0.826	0.815	0.833	0.825	0.838	0.841	0.845	0.835
AJKGS-ABCM	0.875	0.856	0.845	0.861	0.855	0.871	0.854	0.870	0.873	0.881	0.859

方法	Exchange	Mining	Gambling	Token	Darknet	Wallet	Scam	Ico	Hacker	Mixer	User
KNN	0.665	0.661	0.642	0.635	0.638	0.649	0.638	0.619	0.626	0.655	0.648
Deepwalk	0.715	0.720	0.695	0.698	0.688	0.714	0.681	0.709	0.701	0.716	0.715
Node2Vec	0.712	0.720	0.705	0.709	0.706	0.722	0.703	0.718	0.709	0.729	0.716
Struc2Vec	0.733	0.718	0.715	0.722	0.719	0.737	0.721	0.726	0.712	0.729	0.715
I2BGNN	0.823	0.809	0.811	0.815	0.806	0.818	0.825	0.836	0.814	0.811	0.829
GCN	0.811	0.806	0.801	0.795	0.785	0.807	0.803	0.812	0.795	0.808	0.802
GraphSAGE	0.831	0.809	0.812	0.811	0.803	0.818	0.814	0.818	0.825	0.831	0.817
AJKGS-ABCM	0.855	0.841	0.838	0.852	0.836	0.862	0.846	0.857	0.861	0.870	0.849

方法	Exchange	Mining	Gambling	Token	Darknet	Wallet	Scam	Ico	Hacker	Mixer	User
KNN	0.674	0.668	0.646	0.639	0.643	0.652	0.644	0.622	0.632	0.663	0.654
Deepwalk	0.718	0.710	0.702	0.703	0.694	0.719	0.688	0.718	0.706	0.724	0.721
Node2Vec	0.723	0.725	0.710	0.716	0.712	0.726	0.710	0.726	0.716	0.734	0.719
Struc2Vec	0.737	0.726	0.719	0.726	0.724	0.741	0.725	0.732	0.722	0.736	0.727
I2BGNN	0.827	0.816	0.815	0.820	0.814	0.830	0.829	0.838	0.823	0.820	0.835
GCN	0.816	0.809	0.805	0.802	0.788	0.814	0.807	0.816	0.798	0.814	0.808
GraphSAGE	0.837	0.815	0.816	0.818	0.825	0.825	0.819	0.828	0.833	0.838	0.826
AJKGS-ABCM	0.865	0.849	0.841	0.856	0.866	0.866	0.850	0.864	0.867	0.875	0.854

方法	Precision	Recall	micro-F1	macro-F1
KNN	0.655	0.643	0.645	0.648
DeepWalk	0.716	0.703	0.703	0.709
Node2Vec	0.725	0.713	0.715	0.719
Struc2Vec	0.734	0.722	0.725	0.727
I2BGNN	0.83	0.819	0.825	0.824
GCN	0.811	0.802	0.801	0.807
GraphSAGE	0.828	0.819	0.818	0.823
AJKGS-ABCM	0.862	0.852	0.849	0.856

基于图神经网络的账户余额模型区块链地址分类方法

Graph neural network-based address classification method for account balance model blockchain

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 23

相关文章 13

Metrics

推荐阅读 0

[1]	陈晋音, 熊海洋, 马浩男, 郑雅羽. 基于对比学习的图神经网络后门攻击防御方法[J]. 通信学报, 2023, 44(4): 154-166.
[2]	冷涛, 蔡利君, 于爱民, 朱子元, 马建刚, 李超飞, 牛瑞丞, 孟丹. 基于系统溯源图的威胁发现与取证分析综述[J]. 通信学报, 2022, 43(7): 172-188.
[3]	何世文, 袁军, 安振宇, 张敏, 黄永明, 张尧学. 基于图神经网络的联合用户调度与波束成形优化算法[J]. 通信学报, 2022, 43(7): 73-84.
[4]	冯海林, 张潇, 刘同存. 融合评论文本特征和评分图卷积表示的推荐模型[J]. 通信学报, 2022, 43(3): 164-171.
[5]	陈卓, 朱淼, 杜军威. 基于多视角图神经网络的欺诈检测算法[J]. 通信学报, 2022, 43(11): 225-232.
[6]	仲美玉, 吴培良, 窦燕, 刘毅, 孔令富. 基于中文语义-音韵信息的语音识别文本校对模型[J]. 通信学报, 2022, 43(11): 65-79.
[7]	王洪雁, 袁海. 基于骨骼及表观特征融合的动作识别方法[J]. 通信学报, 2022, 43(1): 138-148.
[8]	吴翼腾, 刘伟, 于洪涛. 图神经网络的标签翻转对抗攻击[J]. 通信学报, 2021, 42(9): 65-74.
[9]	刘晨曦, 王东, 陈慧玲, 李仁发. 多源异构数据融合的城市私家车流量预测研究[J]. 通信学报, 2021, 42(3): 54-64.
[10]	赵晓娟, 贾焰, 李爱平, 陈恺. 基于层级注意力机制的链接预测模型研究[J]. 通信学报, 2021, 42(3): 36-44.
[11]	郭璠, 张泳祥, 唐琎, 李伟清. YOLOv3-A：基于注意力机制的交通标志检测网络[J]. 通信学报, 2021, 42(1): 87-99.
[12]	李琳辉,周彬,连静,周雅夫. 基于社会注意力机制的行人轨迹预测方法研究[J]. 通信学报, 2020, 41(6): 175-183.
[13]	王竹荣, 薛伟, 牛亚邦, 崔颖安, 孙钦东, 黑新宏. 基于注意力机制的泊位占有率预测模型研究[J]. 通信学报, 2020, 41(12): 182-192.