1

Blockchain Intelligence: When Blockchain Meets

Artificial Intelligence
Zibin Zheng, Senior Member, IEEE, Hong-Ning Dai, Senior Member, IEEE, Jiajing Wu, Member, IEEE

Abstract—Blockchain is gaining extensive attention due to its overcome shortcomings of AI, such as security and privacy
provision of secure and decentralized resource sharing manner. vulnerabilities [14]), this article mainly concentrates on im-
However, the incumbent blockchain systems also suffer from proving blockchain systems by applying AI technologies. In
a number of challenges in operational maintenance, quality summary, the main contributions of this article are summarized
assurance of smart contracts and malicious behaviour detection as follows:
of blockchain data. The recent advances in artificial intelligence
• We first give an overview of blockchain technology and
arXiv:1912.06485v3 [[Link]] 3 Apr 2020

bring the opportunities in overcoming the above challenges.

The integration of blockchain with artificial intelligence can be point out the challenges in existing blockchain systems.
beneficial to enhance current blockchain systems. This article • We then review the advances in AI and formally in-
presents an introduction of the convergence of blockchain and troduce the convergence of AI and blockchain followed
artificial intelligence (namely blockchain intelligence). This paper by a discussion on opportunities brought by blockchain
also gives a case study to further demonstrate the feasibility of intelligence.
blockchain intelligence and point out the future directions. • We next present a case study to demonstrate the feasi-
Keywords—Blockchain; Artificial Intelligence; Smart Contract; bility of blockchain intelligence.
Machine Learning
II. OVERVIEW OF BLOCKCHAIN TECHNOLOGIES
As a disruptive software technology, blockchain is reshaping
I. I NTRODUCTION diverse business sectors. Blockchain is essentially a chain-
Blockchain has received extensive attention recently due to like data structure storing transactions verified by majority
its provision of secure data sharing services with traceabil- of nodes throughout the whole network as shown in Fig. 1.
ity, immutability and non-repudiation. Despite the merits of Since the committed transactions in the blockchain have been
blockchain, the development of blockchain technologies has stored at every node, they are extremely difficult to be altered
undergone a number of challenges including poor scalability, or falsified. Integrating with digital signature and asymmetric
difficulties in operational maintenance, detecting vulnerable encryption, blockchain data is authenticated and auditable, im-
codes in smart contracts and identifying malicious behaviours plying non-repudiation of the transaction initiator. The length
in blockchain historical data. of blockchain keeps growing with the new validated transac-
The recent advances in artificial intelligence (AI) have tion being appended at end of the chain. Data analytics on
greatly propelled the evolution of diverse business applications. blockchain data can potentially extract valuable information.
The integration of AI with blockchain has the potentials to The development of blockchain technologies has experi-
overcome the limitations of blockchain. We name the intelli- enced two phases: 1) blockchain 1.0 (i.e., symbolized by digital
gent capability bestowed by AI on the blockchain ecosystem currency) and 2) blockchain 2.0 (i.e., symbolized by smart
as blockchain intelligence. In particular, AI approaches such contracts) [1]. In blockchain 1.0, blockchain has been mainly
as machine learning, data mining and data visualization may used for digital currencies like Bitcoin. The appearance of
help to capture the abnormal behaviours in blockchain, identify blockchain has promoted the development of smart contracts.
risks in transactions, detect possible vulnerable program codes A smart contract essentially consists of a number of computer-
in smart contracts, and so on. Consequently, proactive and ized contractual agreements consented by multiple parties [2].
autonomic actions can be made to prevent blockchain from The contractual clauses embedded in smart contracts will be
disruptive or illegal actions, and thus make the blockchain triggered and automatically executed when a certain condition
more intelligent. is satisfied (e.g., who breaches the contract will be automati-
This article aims at reviewing blockchain technologies as cally imposed with a fine).
well as AI technologies, presenting an in-depth analysis on Smart contracts have been implemented on top of blockchain
integrating blockchain with AI, providing implications of as shown in Fig. 1. The approved contractual clauses are
enabling technologies of blockchain intelligence. In contrast converted into executable computer programs. The logical con-
to most of recent studies on the integration of blockchain nections between contractual clauses have also been preserved
and AI (which mainly focuses on applying blockchain to in the form of logical flows in programs (e.g., if-else-if
statement). The execution of each contract statement is
Z. Zheng, J. Wu are with School of Data and Computer Science, recorded as an immutable transaction stored in the blockchain.
Sun Yat-sen University, China (email: zhzibin@[Link], wujia-
jing@[Link]). Meanwhile, smart contracts guarantee appropriate access con-
H.-N. Dai is with Faculty of Information Technology, Macau University of trol and contract enforcement. In particular, developers can
Science and Technology, Macau (email: hndai@[Link]). assign access permissions for each function in the contract.
 2

Genesis block Block i Block i + 1

Transaction Contract Transaction Contract Transaction Contract

roothash roothash roothash roothash roothash roothash

Smart contract Smart contract Smart contract

TX1 TX2 TXn Smart contract TX1 TX2 TXn Smart contract TX1 TX2 TXn Smart contract
Smart contract Smart contract Smart contract

Fig. 1. Overview of blockchain

Although blockchain and blockchain-enabled smart con- general purpose computer hardware such as graphic processing
tracts are promising in reshaping various industrial sectors, units (GPUs) have greatly driven the development of AI.
the intrinsic limitations of blockchain systems also lead to the Consequently, we have witnessed the proliferation of diverse
following challenges. AI applications such as computer vision, natural language
1) Operational maintenance. Due to the decentralization and processing, speech recognition, sentimental analysis. Big data
heterogeneity of blockchain systems, it is difficult to identify plays a critical role in propelling AI as well as AI applications.
the potential factors affecting the performance of blockchain. For example, deep learning mainly based on deep neural
For example, the transaction throughput bottleneck of Hyper- networks (DNNs) has achieved superior performance thanks
ledger Fabric is different from that of Bitcoin and Ethereum to the availability of massive data so that DNNs can extract
since different consensus algorithms are adopted. Moreover, (or learn) enough features from large volume of data.
like other software systems, smart contracts consist of a num- The appearance of diverse blockchain systems has generated
ber of computer programs, which may suffer from software the enormous volumes of blockchain data, which are publicly
bugs, malicious codes and incompatibility of running environ- available to everyone. Take Bitcoin as an example. As reported
ments. Consequently, it is crucial to achieve the intelligent by Statista ([Link] at the end of the third
and robust operational maintenance of complex blockchain quarter of 2019, Bitcoin contains nearly 242 GB data. Data
systems. analytics on the massive blockchain data cannot only extract
2) Quality assurance of smart contracts. Smart contracts huge business values but also bring great opportunities to
suffer from a number of software vulnerabilities such as re- overcome the aforementioned challenges of blockchain sys-
entrancy vulnerability [1], overcharging issue [3], randomness tems. The recent advances in AI have also greatly driven the
controlling [4] and Decentralized Autonomous Organization development of big data analytics [7]. Thus, the integration
(DAO) attack [5]. In addition, contract correctness is also of AI and blockchain technologies can potentially overcome
crucial to smart contracts since it is nearly impossible to make the aforementioned challenges of blockchain systems, thereby
any revisions once they are deployed on top of blockchains. forming intelligent blockchain systems. We name such inte-
However, like software systems, smart contracts often contain gration of blockchain with AI as blockchain intelligence. It is
programming bugs which may lead to crashes or misbe- worth mentioning that there are several research efforts [14],
haviours while it is challenging to detect and identify these [15] on integrating blockchain with AI while most of them
bugs due to the complexity of smart contracts. are mainly concentrated on exploiting blockchain for AI so as
3) Malicious behaviour detection. Besides legal businesses, to overcome the emerging security and privacy vulnerabilities
blockchain may be exploited for malicious activities which are of AI. In contrast to these studies, our work in this paper is
nevertheless difficult to be detected due to the pseudonymity mainly focused on solving the intrinsic issues of blockchains
of blockchain (i.e., anonymous blockchain addresses). On the by using AI technologies.
other hand, the encrypted blockchain data also leads to the
We summarize the opportunities brought by AI to enhance
difficulty of detecting and identifying malicious behaviours
incumbent blockchain systems as follows (as shown in Fig. 2).
via simply data analytics. Moreover, the massive volume and
heterogeneity of blockchain data as well as diversity of user 1) Intelligent operational maintenance of blockchain.
behaviours make the problem even worse. As a result, con- Blockchain generates huge amount of data in a real-time
ventional classification-based methods (e.g., machine learning manner. Analyzing blockchain data, we can detect the pos-
methods) cannot be directly applied. sible faults, forecast the failures and identify the perfor-
mance bottleneck so as to tune or adjust the performance of
blockchain systems. There are four different levels of data
III. O PPORTUNITIES BROUGHT BY A RTIFICIAL analytics including: descriptive analytics, diagnostic analytics,
I NTELLIGENCE predictive analytics and prescriptive analytics. In particular,
Artificial intelligence (AI) as a broad discipline covering the work of [8] presents a common platform to evaluate
machine learning and cognitive computing is an ability of intel- the performance of three representative blockchain systems:
ligent agents conducting intellectual tasks. The recent advances Ethereum, Parity and Hyperledger Fabric via descriptively
in big data, AI technologies (such as deep neural networks) and analyzing blockchain data. Meanwhile, the descriptive analyt-
 3

Blockchain Challenges AI & Big Data Opportunities

1) Intelligent operational maintenance

1 Operational maintenance
of blockchain
Machine learning Data visualization Data mining
Real-time
Monitoring
Anomaly
detection
Intelligent
restoring

2) Intelligent quality assurance of smart

2 Quality assurance of
smart contracts
Anomaly detection Federated ML Data
preprocessing
contracts
Defect auto-
correction
Code auto-
optimization

3 Malicious behaviour 3) AI-based malicious behaviour

detection detection
Fraud Market Criminal gang
Database Storage Computing detection Manipulation detection
management model

Fig. 2. Opportunities brought by artificial intelligence to address challenges of blockchain

ics of blockchain log data can help to monitor the real-time account addresses. The massive blockchain data brings the
performance of blockchain systems and identify the possible opportunities in auditing and detecting malicious behaviours.
faults [6]. In addition to diagnostic analytics on blockchain Big data analytics on massive blockchain data can help to
data, predictive analytics is also necessary to anticipate the per- identify malicious users, recognize behaviour pattern, analyze
formance bottleneck of blockchain systems. Unlike diagnostic market manipulation, detect scams. The work of [12] presents
and predictive analytics, prescriptive analytics can simulate and a cross-graph analysis on Ethereum data and identify several
optimize blockchain systems so as to improve the reliability major activities occurring on Ethereum blockchain platforms.
of blockchain systems. As in [10], a machine learning based approach was proposed to
2) Intelligent quality assurance of smart contracts. Like detect and capture Ponzi schemes that took place in Ethereum.
computer software, smart contracts may contain bugs or faulty The work of [13] analyzes the leaked transaction history of
programming codes, which are vulnerable to crashes and Mt. Gox Bitcoin exchange and identify a number of mar-
malicious attacks. It is crucial to detect and identify bugs ket manipulation patterns via singular value decomposition
in smart contracts so as to achieve the ultimate goal of (SVD) method. Moreover, malicious users may exploit multi-
intelligent quality assurance of smart contracts. The work ple anonymous accounts to form a criminal gang to conduct
of [9] presents a symbolic-execution platform namely Oyente illegal activities on blockchain systems. New machine learning
to detect and recognize potential bugs. Meanwhile, smart approaches as well as association analysis on multiple accounts
contracts are essentially program codes that are sensitive to are expected to address this issue.
execution cost (e.g., the execution of smart contracts is charged The advances in AI, machine learning and big data analytics
by gas in Ethereum). Thus, it is a necessity to identify the gas- bring numerous opportunities to address the aforementioned
costly patterns and correct these vulnerable smart contracts. blockchain challenges. We next present a case study to demon-
In [3], Chen et al. developed a tool namely GASPER to strate the feasibility of blockchain intelligence.
identify and locate seven gas-costly patterns via analyzing
bytecodes of Ethereum smart contracts. Machine learning
methods can be used to detect and recognize vulnerable bugs in IV. C ASE STUDY
smart contracts automatically. Moreover, the growing number Big data analytics of blockchain data is beneficial to fraud
of smart contracts also brings the opportunities to automate recognition of transactions and vulnerability detection of smart
the composition of multiple contracts. In particular, we can contracts. However, it is also challenging to conduct big
find and identify contracts that fulfill users’ requirements data analytics of blockchain data. 1) It is extremely time
and composite them together to realize more comprehensive consuming to download the entire blockchain data due to
applications. Different from conventional distributed software the bulky blockchain size, e.g., it took more than one week
systems such as web services [11], smart contracts lack of and over 500 GB storage space to fully synchronize (i.e.,
semantic description and Quality-of-Service (QoS) evaluation download) the entire Ethereum at a newly-joined peer. 2)
metrics. As a result, new AI-based approaches are expected It requires substantial efforts in extracting and processing
to automate labeling semantics of smart contracts and offer blockchain data. First, blockchain data is stored at clients in
data-driven QoS evaluation of smart contracts. heterogeneous and complex data structures, which cannot be
3) Automated malicious behaviour detection. The decentral- directly analyzed. Meanwhile, the underlying blockchain data
ized blockchain systems result in the difficulty in auditing is either binary or encrypted. Thus, it is a necessity to extract
malicious behaviours such as money laundering, phishing, and process binary and encrypted blockchain data so as to
gambling and scams that occurred in blockchain platforms. obtain valuable information while this process is non-trivial as
Blockchain systems have generated massive transaction data, conventional data analytic methods may not work for this type
which are essentially available to everyone, whereas the histor- of data. 3) There is no general data extract tool for blockchain
ical transaction data are pseudonymous through anonymizing data. Although several open source tools for blockchain data
 4

(a) Raw data Block N Blockchain Peer Ethereum Virtual Machine

collection 30 4.0 1e10

Block Header Execute Contract A MIN MIN
Commit ① Create ② Call 28 MAX 3.5 AVG
Transaction 1
Transaction 2 ③ Suicide ④ Reward AVG 3.0
... 26 2.5

log(GasPrice)
Receipt Contract B

GasPrice
24 2.0
22 1.5
Block Receipt Trace
Trace 1.0
20 0.5
(b) Processed 180 100 200 300 400 500 600 700 800 0.00 200 400 600 800 1000
Datasets Per 10000 Block Per 20 Block
Block and Internal Ether Contract Contract ERC20 Token ERC721 Token
Transaction Transaction Info Calling Transaction Transaction (a) Macro view of GasPrice (b) Micro view of GasPrice

Fig. 4. Data visualization of Gas Price of Ethereum

Fig. 3. Data processing during Ethereum transaction flow

transactions, respectively. In addition, the size of circle also

extraction are available, most of them can only support to
represents the amount of transactions, i.e., the larger circle
extract partial blockchain data (not the entire data).
means the larger amount of transactions.
Fig. 5(a) is a normal lottery contract while Fig. 5(b) shows
A. Data Extraction from Ethereum a typical Ponzi scam (namely Rubixi). We can observe several
To address the above challenges, we propose a blockchain significant differences between Fig. 5(a) and Fig. 5(b): 1) there
data analytics framework namely XBlock-ETH to analyze are more participants in Fig. 5(b) than Fig. 5(a); 2) there are
Ethereum data. In particular, we extract raw data consisting of more payment transactions in Fig. 5(b) than Fig. 5(a) which
8,100,000 blocks of Ethereum. Fig. 3(a) illustrates the typical exhibits more randomness in the number of transactions. After
Ethereum transaction execution flow from Block N to EVM extracting the key features and applying other data mining and
through blockchain peer. During this procedure, we collect the machine learning methods, we can successfully classify Ponzi
three types of blockchain raw data: Block, Receipt and Trace. scams from other normal activities.
Since the analysis on the raw blockchain data is difficult, we
process and categorize the obtained Ethereum Blockchain data V. C ONCLUSION AND FUTURE DIRECTIONS
into six datasets: (1) Block and Transaction, (2) Internal Ether
This article first reviews the blockchain technologies and
Transaction, (3) Contract Information, (4) Contract Calls, (5)
analyze the challenges in blockchain systems. We then intro-
ERC20 Token Transactions, (6) ERC721 Token Transactions
duce artificial intelligence (AI) as well as opportunities brought
as shown in Fig. 3(b). It is non-trivial to process the raw since
by AI to blockchain systems. We name such integration of
it requires substantial efforts in extracting useful information
blockchain and AI as blockchain intelligence. We mainly
from raw data and associating with six datasets.
discuss that AI bring benefits to blockchain in aspects of
We then conduct statistic analysis on these refined datasets.
intelligent operational maintenance of blockchain, intelligent
In Ethereum, a miner has a higher priority to package the
quality assurance of smart contracts and automated malicious
transactions with higher “gasPrice” into the block. The
behaviour detection. In addition, we also give a case study to
visualization of “gasPrice” is shown in Fig. 4. In a macro
further demonstrate great potentials of blockchain intelligence.
view (as shown in Fig. 4(a)), the “gasPrice” is gradually
We believe that the integration of AI with blockchain
decreasing with the development of the Ethereum community,
technology will further drive the benignant development
except for several peaks caused by extremely frequent transac-
of blockchain systems. We outline the future directions in
tion when the network is congested. In a micro view (as shown
blockchain intelligence as follows.
in Fig. 4(b)), we extract the time from 8,000,000 to 8,020,000
blocks and find that such fluctuations of “gasPrice” can be • Real-time and automated operational maintenance of
observed by the tidal law. This observation implies that the blockchain. Instead of performance monitoring and fault
fluctuations of “gasPrice” can potentially be predicted. detection, the future intelligent operational blockchain
systems are expected to conduct real-time monitoring
on multiple performance metrics of blockchain systems
B. Detection Ponzi Schemes from Ethereum and achieve the automated restoring from crash spots.
Blockchain can also be exploited to conduct illegal activities • Collective intelligence bestowing smart contracts. The
such as scams. For example, in [10], we propose a method current QoS assurance approaches of smart contracts are
to detect Ponzi scams in Ethereum blockchain through ex- mainly based on the analysis on contract patterns (i.e.,
tracting and analyzing key characteristics of user accounts detecting and classifying vulnerable contracts) at a sole
and operation codes in Ethereum contracts. Fig. 5 makes a peer in the blockchain. In contrast to a single intelligent
comparison between a normal contract and a Ponzi scheme agent, collective intelligence can motivate all partici-
contract in terms of Ether Flow Graph, where the horizontal pants to engage in analyzing and reasoning, thereby
axis denotes the time line, the vertical axis represents the contributing their collective knowledge to make better
number of participants in a particular contract, red circles and decision in a global context. In the future, collective
blue circles denote the investment transactions and payment intelligence is expected to integrate with decentralized
 5

● 120th ●

● ●
●
●
●
● ●
●●●
●●●
20th ●
●
●● ● ●
●●
● 105th
●●
●
●
●
●

●● ● ● ● ●●● ● ●
●● ● ●● ●● ● ●
●●
● ● ●●● ●
● ●
● ● ● ●
●
●
●
●●
● ●

●● ● ●●●●
●● ● ●●● ● ● ●● ●●
●● ●
●●
● ● ● ● ●● ● ●
● ● ●●
Amount 90th ● ●●●
●●
●● ●
● ● ● ● ●
● ●
● ● ● ● ●● ● ● ● ●●● ● ●●●●● ● ● ●● ● ●
●
● ●
● ●●
●● ●● ● ●
Direction
●● ● ●
No. of participants

No. of participants
●● ● ● ●
● ●● ●
● ● ●● ● ●
● ●
● ● ●● ● ●●● ● ●
● ● ●
● ● ● ● ●● ● ● ● ● ● ● ● 0.5 ●●
●
●
●● ● ●
●●
●
Investment
● ● ●
15th ●● ● ●● ● ●● ●● ● ● ●
●
●● ● ●● ● ●● ● ●●
● ● ● ● ● ●●
1.0 75th ●●●
●●
● ●
● ●● ● ● ●●● ●● ● ● ● ● ●●●
●●●● ●
● ● ● ●
● ●●
●
● ●●●●
● ● ● ●● ●●

● 1.5 ●●
●●
●
●
●
●
● Payment
●● ●● ●●
● ● ●●
●● ● ● ● ● ●●
●● ●●
● ● ●● ●
●●●
● ● ● ●●● ●●
● ● ● ●●
● ●● ●●
● ● ●● ●●
●
●● ● ●● ●● ●● ●● ● ●● ● ● ●● ● ● ● ●
● ● ● ●● ●
● ● 2.0 60th
●
●
●
●
●●

● ● ●● ● ●
● ●● ● ●● ●
●● ● ● ●● ● ●● ● ●● ●●
●
●● ●
●● ● ●
●●●
● ●● ●
● ● ●● ●●● ●
●●
●● ● ● ●
●● ●● ● Amount
●
10th ●● ● ●●●● ●●●● ●● ●
●
● ● ●● ● ●● ●
● ● ● ●●● ● ●●● ●
●● ● ● ●

Direction 45th
●
●
●●
● 50
● ●● ● ●
● ● ● ● ● ●● ●● ● ● ● ● ●●●● ● ● ●
● ● ● ● ● ●● ●
●●
● ● ●● ● ●● ●●● ●●
●
●
●● ● ● ● ●● ● ● ●
● ● ● ● ● ●● ●● ●●
● ● ●
● ● ●●● ● ● ● ● ●● ●● ● ●
● Investment ●●
●
●
● ● 100
●● ● ●
● ●
●●● ● ●
● ● ●●●
●● ●
● ● ●● ●● ● ●● ● ● ● ● ● ●● ● ●● ●
● Payment 30th ●
●●
●
●
●
●
● 150
● ●●
● ●●
●● ● ●● ● ●
●● ●● ● ●● ● ● ● ●● ● ●● ● ● ●
● ● ●
●● ●●
●
●
5th ●● ●
● ● ● ● ● ●●● ●● ●
● ● ●● ●
● ●● ● ●●● ●● ●● ● ●
● ●●●
● ●● ●● ●● ● ● ● ● ●●●● ●
●
●
●●
●

●●
● ● ●
●● ●
●
●
● ● ● ● ● ●● ● ● ● ● ● ●● ● ●●● ●
●
● ● ● ●● ● ●● ●● ●●●
●●●● ● ●● ● ●● ● ● ● ●
15th ●●
●
●
●●
●
●●●
●
●
●
●
●●●
●
●
●
●
●
●● ● ●●●
●
●
●
● ● ● ●● ●
●● ● ● ● ●
●
●● ● ●● ● ● ● ●● ●●●
●
●● ● ● ●●
●● ●●
● ● ●●
● ●
●
●

●
●●
●●
●
●●
● ●
●●●
●
●●
●
● ●
● ● ● ●●
●● ●
● ●●● ●● ● ● ●● ●
●
●● ● ●
● ●●● ● ● ●●● ● ●● ● ● ●
● ● ●
●
●
●
●
●● ●
●●●● ● ●
●
●● ●● ●
●●●●
●
● ●
● ● ● ●● ● ● ● ● ●
●

1th ● ● 0th

2016−02−18 02−19 02−20 02−22 02−23 02−24 02−25 02−27 2016−03−14 2016−03−20 2016−04−05 2017−08−19
Time Time

(a) LooneyLottery contract (i.e., normal contract) (b) Rubixi contract (i.e., Ponzi scam contract)

Fig. 5. Ether Flow Graph of two smart contracts [10].

blockchain system so as to offer a trustworthy and [8] T. T. A. Dinh, R. Liu, M. Zhang, G. Chen, B. C. Ooi and J. Wang, “Un-
intelligent provision of blockchain services. tangling Blockchain: A Data Processing View of Blockchain Systems,”
IEEE Transactions on Knowledge and Data Engineering, vol. 30, no. 7,
• Integrating multiple machine learning approaches to pp. 1366-1385, 1 July 2018
monitor and supervise blockchain data. Decentralization [9] L. Luu, D.-H. Chu, H. Olickel, P. Saxena, and A. Hobor, “Making Smart
of blockchain brings the difficulty in monitoring and Contracts Smarter,”Proceedings of the 2016 ACM SIGSAC Conference
supervising transactions in blockchain platforms, con- on Computer and Communications Security (CCS ’16), 2016
sequently resulting in a number of illegal behaviours. [10] W. Chen, Z. Zheng, J. Cui, E. Ngai, P. Zheng, and Y. Zhou, “Detecting
Meanwhile, both heterogeneity and pseudonymity of Ponzi Schemes on Ethereum: Towards Healthier Blockchain Technol-
blockchain data make this situation even worse. In the ogy,” Proceedings of the 2018 World Wide Web Conference (WWW ’18),
1409-1418.
future, multiple machine learning approaches should be
[11] P. Rodriguez-Mier, C. Pedrinaci, M. Lama and M. Mucientes, “An Inte-
integrated together to extract key features from different grated Semantic Web Service Discovery and Composition Framework,”
types of blockchain data. In addition, a dynamic graph IEEE Transactions on Services Computing, vol. 9, no. 4, pp. 537-550, 1
(or network) of transactions is also expected to identify July-Aug. 2016.
the association between different accounts so as to [12] T. Chen et al., “Understanding Ethereum via Graph Analysis,”IEEE
recognize the malicious behaviours. INFOCOM 2018 - IEEE Conference on Computer Communications,
2018, pp. 1484-1492.
[13] W. Chen et al., “Market Manipulation of Bitcoin: Evidence from Min-
R EFERENCES ing the Mt. Gox Transaction Network”,IEEE Conference on Computer
Communications, 2019, pp. 964-972.
[1] X. Li, P. Jiang, T. Chen, X. Luo, Q. Wen, “A survey on the security [14] T. N. Dinh and M. T. Thai, “AI and Blockchain: A Disruptive Integra-
of blockchain systems”, Future Generation Computer Systems, 2017, tion”, Computer, vol. 51, no. 9, pp. 48-53, September 2018.
[Link] [15] Nassar, M, Salah, K, ur Rehman, MH, Svetinovic, D., “Blockchain for
[2] N. Szabo, “The idea of smart contracts”, Nick Szabo’s Papers explainable and trustworthy artificial intelligence”, WIREs Data Mining
and Concise Tutorials. [Link] Knowl Discov, 2020; 10:e1340. [Link]
InformationInSpeech/CDROM/Literature/LOTwinterschool2006/szabo.
[Link]/smart contracts [Link]
[3] T. Chen, X. Li, X. Luo, X. Zhang, “Under-optimized smart contracts
devour your money”, Proceedings of 24th International Conference on
Software Analysis, Evolution and Reengineering (SANER), 2017, pp.
442–446.
[4] J. Bonneau, J. Clark, S. Goldfeder, “On bitcoin as a public randomness
source”, IACR Cryptology ePrint Archive, 1015, 2015
[5] The DAO, The Hack, The Soft Fork and The Hard
Fork (2017). [Link]
the-dao-the-hack-the-soft-fork-and-the-hard-fork/
[6] P. Zheng, Z. Zheng, X. Luo, X. Chen, X. Liu, “A detailed and real-
time performance monitoring framework for blockchain systems,” 2018
IEEE/ACM 40th International Conference on Software Engineering:
Software Engineering in Practice Track, pp. 134–143, 2018. (conference
proceedings)
[7] H.-N. Dai et al., “Big data analytics for large-scale wireless networks:
Challenges and opportunities,”, ACM Computing Surveys, vol. 52, no. 5,
2019