CN114547611A - Intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics - Google Patents
Intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics Download PDFInfo
- Publication number
- CN114547611A CN114547611A CN202210165284.4A CN202210165284A CN114547611A CN 114547611 A CN114547611 A CN 114547611A CN 202210165284 A CN202210165284 A CN 202210165284A CN 114547611 A CN114547611 A CN 114547611A
- Authority
- CN
- China
- Prior art keywords
- intelligent contract
- transaction
- current
- graph
- pompe
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 claims abstract description 18
- 230000007246 mechanism Effects 0.000 claims abstract description 14
- 238000004458 analytical method Methods 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 11
- 238000005516 engineering process Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims description 85
- 230000006870 function Effects 0.000 claims description 70
- 238000000605 extraction Methods 0.000 claims description 19
- 230000006399 behavior Effects 0.000 claims description 15
- 239000002131 composite material Substances 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 230000004048 modification Effects 0.000 claims description 4
- 238000012986 modification Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000005206 flow analysis Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 3
- 239000007787 solid Substances 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 abstract 1
- 238000012163 sequencing technique Methods 0.000 abstract 1
- 230000003068 static effect Effects 0.000 abstract 1
- 238000013459 approach Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/50—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems
- G06F21/55—Detecting local intrusion or implementing counter-measures
- G06F21/56—Computer malware detection or handling, e.g. anti-virus arrangements
- G06F21/562—Static detection
- G06F21/563—Static detection by source code analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
- G06Q20/382—Payment protocols; Details thereof insuring higher security of transaction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
- G06Q20/40—Authorisation, e.g. identification of payer or payee, verification of customer or shop credentials; Review and approval of payers, e.g. check credit lines or negative lists
- G06Q20/401—Transaction verification
- G06Q20/4016—Transaction verification involving fraud or risk level assessment in transaction processing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Accounting & Taxation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Computer Hardware Design (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Finance (AREA)
- Data Mining & Analysis (AREA)
- General Business, Economics & Management (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Virology (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses an intelligent contract Ponconi deception bureau detection method and system based on multi-modal characteristics, wherein the method comprises the steps of firstly, carrying out sequencing representation learning on intelligent contract source codes to obtain global lexical characteristics; then, constructing an intelligent contract transaction attribute graph by utilizing static analysis, and extracting local transaction characteristics through a graph neural network; and (4) fusing global lexical features and local transaction features by using an attention mechanism, and carrying out intelligent contract detection of the Pompe deception on a secondary basis. The invention has the advantages that: the intelligent contract code semantics are fully extracted by combining two modal characteristics of the global lexical method and the local transaction of the intelligent contract, and the detection accuracy is improved; the program slicing technology and the code graph are used for representing that noise codes are removed while preserving the Pompe deception semantic related codes, so that the detection accuracy is improved; the detection method based on the code characteristics and the deep learning does not depend on other data and expert rules, improves the detection application range and efficiency, and reduces the detection cost.
Description
Technical Field
The invention belongs to the field of software security, and relates to an intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics.
Background
A smart contract is a piece of software that runs on a blockchain platform. The system has characteristics of decentralization, incapability of tampering, anonymous execution and the like, and has attracted wide attention in numerous fields such as finance, investment, games and the like. With the widespread application of blockchain technology, a great deal of fraud is emerging on blockchain platforms, the most common of which is a pompe fraud based on intelligent contracts.
The pompe fraud is a common investment fraud that attracts investors through promises of high returns and pays the return of existing investors by approving the funds of new entry investors. The intelligent contract-based pompe frauds have the characteristics of automatic execution, anonymous transaction, difficult tracing and the like. Statistically, current intelligent contracts for pompe frauds have exceeded 2 million victims and cause economic losses of over $ 1.7 billion. How to detect intelligent contract detection of the Pompe fraudster becomes a problem to be solved urgently in block chain supervision.
The existing intelligent contract Pompe fraudster detection method comprises the following steps: 3 methods based on rules, neural network and transaction mode. The rule-based method relies on experts to define rules, and technologies such as symbolic execution or fuzzy test are utilized to search whether suspicious code segments meeting the rules of the experts exist in the codes. But the artificially defined rules are usually simple and lack flexibility, so that all possible scenarios cannot be completely covered, resulting in a high false alarm rate. The neural network-based method inputs all codes into the neural network as sequences, so that structural features and semantic features of the codes are lost, and the learning capability of the model is limited. And the code irrelevant to the Pompe fraudster in the codes is used as noise to interfere with the characteristic learning process of the neural network, so that the detection result is inaccurate. While the transaction pattern-based approach relies on past transaction data of the tested smart contract, this approach is applicable to active smart contracts where there is a large amount of transaction data. For those intelligent contract samples without sufficient transaction data, the detection effect of the method is poor, so that the application range of the method is small.
Disclosure of Invention
The purpose of the invention is as follows: in view of the problems in the prior art, the present invention aims to provide an intelligent contract pointcast cheating bureau detection method and system based on multi-modal characteristics, which have the advantages of more accurate detection, higher execution efficiency and wider application range.
The technical scheme is as follows: the invention provides an intelligent contract Pompe fraudster detection method based on multi-modal characteristics, which comprises the following steps of:
(1) constructing an intelligent contract source code data set, and marking a Pompe cheating intelligent contract sample in the data set by utilizing forum data and a manual auditing method;
(2) performing lexical analysis on the intelligent contract to obtain an abstract syntax tree, and developing the abstract syntax tree into a sequence by using a code structured traversal technology to obtain an intelligent contract serialized representation;
(3) program slicing is carried out on intelligent contract source codes by using grammatical features of the Pompe deception intelligent contract as slicing criteria, and sentences related to the Pompe deception semantic meanings are extracted to form a slice set; performing control flow and data flow analysis on the phrase set to construct an intelligent contract transaction attribute graph;
(4) performing feature extraction on the intelligent contract serialization input model by using a transform encoder as a feature extraction model to obtain an intelligent contract global feature vector;
(5) performing feature extraction on the intelligent contract transaction attribute graph by using a graph convolution neural network and graph self-attention pooling operation to obtain a local feature vector for intelligent contract transaction;
(6) and (5) fusing the global lexical feature vector and the local transaction feature vector of the intelligent contract obtained in the step (4) and the step (5) to obtain a composite feature vector of the intelligent contract, and inputting the composite feature vector of the intelligent contract into a multi-layer sensing machine to predict whether the current intelligent contract is a Pompe fraudster.
Further, the step (1) comprises the steps of:
(1.1) manually auditing the collected intelligent contract source codes, confirming whether the intelligent contract sample is a Pompe fraudster intelligent contract or not, and adding the intelligent contract confirmed as the Pompe fraudster intelligent contract into the Setponzi＝{c1,...,cN-wherein N represents the number of pointcast cheat intelligent contracts;
(1.2) collecting intelligent contract addresses through a BigQuery database, obtaining corresponding intelligent contract source codes through an Etherscan platform, manually confirming non-Ponz contract sample sets in the intelligent contract source codes, and forming the non-Ponz contract sample sets through the samplesno_ponzi＝{c1,...,cM-wherein M represents a non-pompe fraud intelligent contract sample number;
(1.2) integrating the intelligent contract samples obtained in the steps (1.1) and (1.2) to obtain an intelligent contract data Set (Set ═ Set)ponzi∪Setno_ponziAnd the intelligent contract sample number in the data set is S-M + N.
Further, the step (2) comprises the steps of:
(2.1) Using the solid compiler to sample the Smart contract ciCompiling the contract into an abstract syntax tree, traversing the abstract syntax tree, extracting a subtree with the root node type of functional definition, and constructing a current intelligent contract sample ciAbstract syntax tree collection of included functions
Wherein K represents the current intelligent contract ciThe number of functions;
(2.2) abstracting the syntax tree of the function with a structured traversal
Expansion into token sequence
Wherein
Representing the length of a token sequence of the current function; merging token sequence sets of all functions to obtain a current intelligent contract sample ciIs shown in a serialized form
(2.3) merging the serialized representations of all the intelligent contract samples to form an intelligent contract corpus
Training on the corpus by using a FastText model to obtain any tokeniCorresponding initial feature vector
Further, the step (3) includes the steps of:
(3.1) against Smart contract sample ciSet of abstract syntax tree of epsilon Set and function thereof
Traversing all function abstract syntax trees using syntax analysis
Judging whether the current function contains transfer operation, if so, adding the function into the sensitive function set corresponding to the current contract sample
The rest of the functions forming a non-sensitive set of functions
(3.2) arbitrary function in the set of sensitivity functions
For its abstract syntax tree
Semantic analysis is carried out to obtain the function program dependence graph
Using 3 transfer transaction interfaces as slicing criteria, according to the principle of graph accessibility
Backward slicing is carried out to obtain a transaction slice attribute graph of the current function
And the node V epsilon V in the graph represents a function fiThe statement with semantic dependency relation with transaction behavior, wherein an edge E belongs to E in the graph, and the statement comprises the following components: control flow, data flow, control dependence and data dependence;
(3.3) traversal
And searching global variables used by all nodes in the current function fiTransaction sensitive global variable set of
Merging the transaction sensitive global variables in all sensitive functions to obtain the current intelligent contract ciTransaction sensitive global variable set of
(3.4) traverse the current intelligent contract sample ciMiddle non-transaction sensitive function
Corresponding to the abstract syntax tree, judging whether the function uses transaction sensitive global variables
If used, the current function is aggregated from the non-transaction sensitive functions
Transition to transaction sensitive function set
And generating a transaction slice attribute graph of the current function by taking the transaction sensitive global variable used by the function as a slice criterion
(3.5) Global variables sensitive to arbitrary transactions
Traversing the current intelligent contract to find the definition statement of the variable to form a global variable definition statement set Vdef(ii) a Merging the transaction slice attribute map of all transaction sensitive functions and the global variable definition statement set VdefGenerating a contract transaction attribute map TPG (V, E) of the current contract sample; wherein the node V belongs to V and represents statements such as transaction sensitive global variable definition, transaction sensitive global variable use, transaction interface calling and the like related to transaction behaviors in the current intelligent contract; the edge E E in the figure comprises: data flow, control flow, data dependency, control dependency, and transaction sensitive variable modification.
Further, the step (4) comprises the steps of:
(4.1) for any Intelligent contract sample ciGenerating (2.3) a serialized representation of the current sample
Dividing the batch into a plurality of batchs, wherein the number of tokens contained in each batch is l';
(4.2) calculating the position encoding vector PE of each token aiming at any batch, wherein the calculation method comprises the following steps:
PE(pos,2i)＝sin(pos/100002i/d)
PE(pos,2i+1)＝cos(pos/100002i/d)
wherein pos represents the position of the current token in batch; d is the dimension of the vector when token position encoding; i represents the index of the position-coding vector;
(4.3) adding the position coding vector of the token and the initial feature vector of the token to obtain the token in the current batchiInitialization vector of
Merging all token initialization vector representations in all batchs to obtain intelligent contract serialization representation
Wherein the content of the first and second substances,
representing the vector representation corresponding to the jth batch; d represents the feature vector dimension of each token; l' is the number of tokens in batch; n is a radical ofbatchA quantity of lots representing a sequence of intelligent contracts;
(4.4) serializing the Intelligent contract representation
Inputting the characteristics into a multi-head self-attention module for feature learning; the multi-head attention mechanism has J attention heads, and the dimension of the characteristic vector output by the multi-head self-attention module is dmultiheadFor each attention headiThe learning characteristics are as follows:
wherein the content of the first and second substances,
representing a matrix formed by feature vectors of all tokens in the jth batch; wi Q、Wi KAnd Wi VAre learnable parameters of the current attention head and have dimensions of
dhead＝dmultiheadJ; combining the results of all attention heads to obtain the output of the current multi-head attention mechanism:
wherein
Is a learnable parameter;
(4.5) outputting a multi-head attention mechanism
And the original input XciResidual error connection is carried out and the residual error is input into a feedforward network through layer normalization operation to obtain the h-th global feature vector representation of the current intelligent contractg：
hffn_in＝LayerNorm(X+hmultihead(X))
hg＝max(0,hffn_inW1 ffn+b1)W2 ffn+b2
Wherein, W1 ffn、W2 ffn、
Parameters may be learned for a feed forward network.
Further, the step (5) includes the steps of:
(5.1) for any Intelligent contract sample ciE Set corresponding transaction attribute graph
Wherein any node represents a statement in a contract; drawing (A)The initial feature vector of any node in the set is represented as: the sum of all token initial vectors in the current node; firstly, carrying out token sequence expansion on any node to obtain v ═ token1,...,tokennFourthly, calculating a node v in the graphiInitial feature vector of
Comprises the following steps:
(5.2) any node v in the graphiWith a neighboring node vje.N (i) passing edge e(i,j)Connection, current node viThe state update formula for the convolution layer in the t-th graph is:
wherein the content of the first and second substances,
cascading current node characteristics, neighbor node characteristics and the mutual relation between the two nodes as the input of state updating so that the model can learn the interaction between the nodes;
and
is a learnable parameter of the current convolutional layer; the final characteristic vector of any node v in the graph obtained by the T-layer graph convolution layer is expressed as
(5.3) utilization of the drawings
Each node V in the set belongs to V initial characteristic vector
Final feature vector
Obtaining local feature vector of current intelligent contract
Where, conv stands for one-dimensional convolution operation,
is the number of nodes in the slice code property graph, σ (-) represents the activation function.
Further, the step (6) comprises the steps of:
(6.1) sample c for the current intelligent contractiCorresponding global features
And local features
Respectively calculating the corresponding attention weights in the following calculation modes:
wherein, Wg、Wl、bgAnd blIs a learnable parameter;
(6.2) weighting the global and local features by using the attention weight to obtain a final feature vector of the current intelligent contract
Represents:
and (6.3) inputting the generated final feature vector of the intelligent contract into a multilayer sensing machine, and judging whether the current intelligent contract is a Pompe deception intelligent contract.
Based on the same inventive concept, the invention also provides an intelligent contract Pompe fraudster detection system based on multi-modal characteristics, which comprises:
the data set construction module is used for constructing an intelligent contract source code data set; collecting intelligent contract source codes by utilizing forum data, and marking whether the intelligent contract sample is a Pompe cheat or not by utilizing a manual auditing method;
the serialization representation module is used for obtaining the intelligent contract serialization representation; firstly, carrying out lexical analysis on intelligent contract codes to obtain an abstract syntax tree of the intelligent contract codes, and then carrying out serialized expansion by using a structured traversal technology to obtain code serialized representation;
the transaction attribute graph extraction module is used for extracting local semantic features related to transaction behaviors in the intelligent contract; the nodes in the transaction attribute graph are composed of statements related to transaction behaviors; the edges in the transaction attribute graph comprise a plurality of semantic types such as a data dependence edge, a data flow edge, a control flow and a control dependence edge;
the intelligent contract global feature extraction module is used for extracting global lexical features of the intelligent contracts; extracting global features from the intelligent contract serialization representation by using a transformer encoder model;
the intelligent contract local transaction feature extraction module is used for extracting transaction related local features from the intelligent contract transaction attribute graph by utilizing a graph convolution neural network and graph self-attention pooling;
the multi-mode feature fusion detection module is used for fusing global features and local transaction features by using a multi-head attention mechanism to obtain a current intelligent contract composite feature representation; and inputting the composite characteristics into a multi-layer sensing machine to judge whether the current intelligent contract is a Pompe fraudster or not.
Has the advantages that: compared with the prior art, the invention has the following beneficial effects:
1. the invention automatically excavates the grammatical and semantic features of various types of Pompe fraudster intelligent contracts through a large number of intelligent contract samples training neural network models to carry out Pompe fraudster detection, and avoids the dependence on manually defined rules, so that the invention can obtain more accurate detection results, wider detection range and lower detection cost; meanwhile, the invention does not depend on any transaction data, so that the invention can detect whether the current investment object is the Pompe fraudster or not before the investor transacts with the intelligent contract of the Pompe fraudster;
2. the method accurately extracts the transaction behavior related code characteristics in the intelligent contract, considers the transaction behavior related code characteristics in and among functions, can more comprehensively reserve the semantic information related to transaction in the code, and simultaneously reduces the interference of other transaction unrelated codes on the model;
3. the invention considers various semantic relation information such as code control flow, data flow, control dependence, data dependence, variable modification and the like when extracting transaction related features, and the feature updating in the model is to input the semantic relation among different nodes into the model, thereby ensuring that the model can learn the features of different semantic relations;
4. the invention combines the global lexical characteristics and the local characteristics of the intelligent contract as the basis for detecting the Pompe frauds, wherein the global lexical characteristics fully utilize the code naturalness in the intelligent contract and improve the capability of a model for understanding the intention of an intelligent contract developer; the local features fully utilize the semantics of related codes of transaction behaviors in the intelligent contract and help the model to more accurately learn the transaction mode in the intelligent contract; by combining the code intention and the code transaction mode semantics, the model is helped to more accurately identify whether the current intelligent contract is a Pompe cheating contract.
Drawings
FIG. 1 is a flow chart of an intelligent contract Pompe fraudster detection method based on multi-modal features.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
The invention provides an intelligent contract Pompe fraudster detection method based on multi-modal characteristics, which comprises the following steps of:
and constructing an intelligent contract source code data set, and marking a Pompe cheat intelligent contract sample in the data set by using forum data and a manual auditing method.
(1.1) collecting intelligent contract addresses disclosed as Pompe frauds in the bitcointalk forum and the dapprada forum, and collecting intelligent contract source codes corresponding to the addresses through an Etherscan platform. Manually auditing the collected intelligent contract source codes, confirming whether the intelligent contract sample is a Pompe fraudster intelligent contract or not, and adding the intelligent contract confirmed as the Pompe fraudster into the Setponzi＝{c1,...,cNWhere N represents the number of pointcast cheat intelligent contracts.
(1.2) collecting intelligent contract addresses through a BigQuery database, obtaining corresponding intelligent contract source codes through an Etherscan platform, manually confirming non-Ponz contract sample sets in the intelligent contract source codes, and forming the non-Ponz contract sample sets through the samplesno_ponzi＝{c1,...,cMWhere M represents the number of non-pompe fraud intelligent contract samples.
(1.3) integrating the intelligent contract samples obtained in the steps (1.1) and (1.2) to obtain an intelligent contract data Set (Set ═ Set)ponzi∪Setno_ponziAnd the intelligent contract sample number in the data set is S-M + N.
(1) Constructing an intelligent contract serialization representation; and performing lexical analysis on the intelligent contract to obtain an abstract syntax tree, and developing the abstract syntax tree into a sequence by using a code structured traversal technology to obtain the intelligent contract serialized representation.
(2.1) merging intelligence with a solidity compilerAbout sample ciCompiled into an abstract syntax tree. Traversing the abstract syntax tree, extracting a subtree with the root node type of functional definition, and constructing a current intelligent contract sample ciAbstract syntax tree collection of included functions
Wherein K represents the current intelligent contract ciThe number of functions.
(2.2) abstracting the syntax tree of the function using the structured-based traversal method (SBT)
Expansion into token sequence
Wherein
Representing the length of the sequence of the current function token. Merging token sequence sets of all functions to obtain a current intelligent contract sample ciSerialized representation of
(2.3) merging the serialized representations of all the intelligent contract samples to form an intelligent contract corpus
Training on the corpus by using a FastText model to obtain any tokeniCorresponding initial feature vector
(2) Constructing an intelligent contract transaction attribute map representation; and (3) performing program slicing on the intelligent contract source code by using the grammatical features of the Pompe deception intelligent contract as a slicing criterion, and extracting semantic related sentences with the Pompe deception to form a slice set. And performing control flow and data flow analysis on the phrase set to construct an intelligent contract transaction attribute graph.
(3.1) against Smart contract samples ciSet of abstract syntax tree of epsilon Set and function thereof
Traversing all function abstract syntax trees using syntax analysis
Judging whether the current function contains transfer operation, if so, adding the function into the sensitive function set corresponding to the current contract sample
The rest of the functions forming a non-sensitive set of functions
(3.2) arbitrary function in the set of sensitivity functions
For its abstract syntax tree
Semantic analysis is carried out to obtain the function program dependence graph
Using 3 transfer transaction interfaces as slicing criteria, according to the principle of graph accessibility
Performing backward slicing to obtain the Transaction slice attribute Graph (STPG, Sliced Transaction Performance Graph) of the current function
And the node V epsilon V in the graph represents a function fiThe statement with semantic dependency relation with transaction behavior, wherein an edge E belongs to E in the graph and comprises: four types of control flow, data flow, control dependency and data dependency。
(3.3) traversal
And searching global variables used by all nodes in the current function fiTransaction sensitive global variable set of
Merging the transaction sensitive global variables in all sensitive functions to obtain the current intelligent contract ciTransaction sensitive global variable set of
(3.4) traverse the current intelligent contract sample ciMiddle non-transaction sensitive function
Corresponding to the abstract syntax tree, judging whether the function uses transaction sensitive global variables
If used, the current function is aggregated from the non-transaction sensitive functions
Transitioning to transaction sensitive function set
And generating a transaction slice attribute graph of the current function by taking the transaction sensitive global variable used by the function as a slice criterion
(3.5) Global variables sensitive to arbitrary transactions
Traversing the current intelligent contract to find the definition statement of the variable to form a global variable definition statement set Vdef. Merging the transaction slice attribute map of all transaction sensitive functions and the global variable definition statement set VdefThe contract transaction attribute Graph (containtransactionproperty Graph) TPG of the current contract sample is generated as (V, E). Wherein the node V belongs to V and represents statements such as transaction sensitive global variable definition, transaction sensitive global variable use, transaction interface call and the like related to transaction behaviors in the current intelligent contract. The edge E E in the figure comprises: data flow, control flow, data dependency, control dependency, and transaction sensitive variable modification.
(3) And (3) performing feature extraction on the intelligent contract serialization input model by using a transformerecoder as a feature extraction model to obtain an intelligent contract global feature vector.
(4.1) for any Intelligent contract sample ciGenerating (2.3) a serialized representation of the current sample
The device is divided into a plurality of batchs, and the number of tokens contained in each batch is l'.
(4.2) calculating the position encoding vector PE of each token aiming at any batch, wherein the calculation method comprises the following steps:
PE(pos,2i)＝sin(pos/100002i/d)
PE(pos,2i+1)＝cos(pos/100002i/d)
where pos represents the location of the current token in batch; d is the dimension of the vector when token position encoding; i represents the index of the position-coding vector.
(4.3) adding the position coding vector of the token and the initial feature vector of the token obtained in the step (2.3) to obtain the token in the current batchiInitialization vector of
Merging all token initialization vector representations in all batchs to obtain intelligent contract serialization representation
Wherein
Representing the vector representation corresponding to the jth batch; d represents the feature vector dimension of each token; l' is the number of tokens in batch; n is a radical ofbatchRepresenting the number of lots of the intelligent contract sequence.
(4.4) serializing the Intelligent contract representation
Inputting the multi-head self-attention module for feature learning. The multi-head attention mechanism has J attention heads, and the dimension of the characteristic vector output by the multi-head self-attention module is dmultiheadFor each attention headiThe learning characteristics are as follows:
wherein
Are learnable parameters.
(4.5) outputting a multi-head attention mechanism
And the original input
Residual error connection is carried out and the residual error is input into a feedforward network through layer normalization operation, and the h-th global feature vector representation of the current intelligent contract is obtainedg：
hffn_in＝LayerNorm(X+hmultihead(X))
hg＝max(0,hffn_inW1 ffn+b1)W2 ffn+b2
Wherein W1 ffn、W2 ffn、
Parameters may be learned for a feed forward network.
(4) And performing feature extraction on the intelligent contract transaction attribute graph by using a graph convolution neural network and graph self-attention pooling operation to obtain a local feature vector for intelligent contract transaction.
(5.1) any intelligent contract samples c generated for step (3.5)iCorresponding trade attribute graph of e Set
Where any node represents a statement in a contract. The initial feature vector of any node in the graph is represented as: the sum of all token initial vectors in the current node. Firstly, carrying out token sequence expansion on any node to obtain v ═ token1,...,tokennAnd (4) reusing the initial feature vector of each token in the corpus generated in the step (2.3)
Calculating a node v in a graphiInitial feature vector of
Comprises the following steps:
(5.2) any node v in the graphiWith a neighboring node vje.N (i) passing edge e(i,j)Connection, current node viThe state update formula for the convolution layer in the t-th graph is:
wherein
Cascading current node characteristics, neighbor node characteristics and the mutual relation between the two nodes as the input of state updating so that the model can learn the interaction between the nodes;
and
is a learnable parameter for the current convolutional layer. The final characteristic vector of any node v in the graph obtained by the T-layer graph convolution layer is expressed as
(5.3) utilization of the drawings
Each node V in the set belongs to V initial characteristic vector
Final feature vector
Get the currentLocal feature vector of intelligent contract
Where conv stands for a one-dimensional convolution operation,
is the number of nodes in the slice code property graph, σ (-) represents the activation function.
(5) And (3) constructing a multi-mode feature fusion detection module based on an attention mechanism, fusing the global feature vector and the local transaction feature vector of the intelligent contract obtained in the steps (4) and (5) to obtain a composite feature vector of the intelligent contract, and inputting the composite feature vector of the intelligent contract into a multi-layer sensing machine to predict whether the current intelligent contract is a Pompe fraudster.
(6.1) sample c for the current intelligent contractiCorresponding global features
And local features
Respectively calculating the corresponding attention weights in the following calculation modes:
wherein Wg、Wl、bgAnd blAre learnable parameters.
(6.2) weighting the global and local features by using the attention weight to obtain a final feature vector of the current intelligent contract
Represents:
and (6.3) inputting the generated final feature vector of the intelligent contract into a multilayer sensing machine, and judging whether the current intelligent contract is a Pompe deception intelligent contract.
Based on the same inventive concept, the invention also provides an intelligent contract Pompe fraudster detection system based on multi-modal characteristics, which comprises:
the data set construction module is used for constructing an intelligent contract source code data set; collecting intelligent contract source codes by utilizing forum data, and marking whether the intelligent contract sample is a Pompe cheat or not by utilizing a manual auditing method;
the serialization representation module is used for obtaining the intelligent contract serialization representation; firstly, carrying out lexical analysis on intelligent contract codes to obtain an abstract syntax tree of the intelligent contract codes, and then carrying out serialized expansion by using a structured traversal technology to obtain code serialized representation;
the transaction attribute graph extraction module is used for extracting local semantic features related to transaction behaviors in the intelligent contract; the nodes in the transaction attribute graph are composed of statements related to transaction behaviors; the edges in the transaction attribute graph comprise a plurality of semantic types such as a data dependence edge, a data flow edge, a control flow and a control dependence edge;
the intelligent contract global feature extraction module is used for extracting global lexical features of the intelligent contracts; extracting global features from the intelligent contract serialization representation by using a transformer encoder model;
the intelligent contract local transaction feature extraction module is used for extracting transaction related local features from the intelligent contract transaction attribute graph by utilizing a graph convolution neural network and graph self-attention pooling;
the multi-mode feature fusion detection module is used for fusing global features and local transaction features by using a multi-head attention mechanism to obtain a current intelligent contract composite feature representation; and inputting the composite characteristics into a multi-layer sensing machine to judge whether the current intelligent contract is a Pompe fraudster or not.
In conclusion, the method utilizes the deep learning model to automatically learn the relevant global lexical features and the local transaction features of the Pompe fraudulent bureau from a large number of intelligent contract samples, and uses the attention mechanism to perform feature fusion, so that the detection model can efficiently and comprehensively learn the code semantics, and the detection effect and the detection efficiency of the Pompe fraudulent bureau are improved. When global lexical features are extracted, an abstract syntax tree is expanded by using a structured traversal technology, and global natural language features of the intelligent contract are fully learned by inputting a multi-head attention mechanism. When local transaction characteristics are extracted, intelligent contract codes are subjected to slicing processing by utilizing Pompe deception semantics to obtain transaction related grammars and semantic characteristics, interference of unrelated codes on a model is reduced, the transaction related codes are represented by a graph structure, and the grammars, semantic information and a context structure can be fully reserved. On the basis of global and local feature extraction, an attention mechanism is utilized to perform feature fusion and Pompe fraudster detection, and the low accuracy and high manual dependence of the traditional Pompe fraudster detection method are improved, so that the application range of detection is widened, and the detection cost is reduced.
Claims (8)
1. An intelligent contract Pompe fraudster detection method based on multi-modal characteristics is characterized by comprising the following steps of:
(1) constructing an intelligent contract source code data set, and marking a Pompe cheating intelligent contract sample in the data set by utilizing forum data and a manual auditing method;
(2) performing lexical analysis on the intelligent contract to obtain an abstract syntax tree, and developing the abstract syntax tree into a sequence by using a code structured traversal technology to obtain an intelligent contract serialized representation;
(3) program slicing is carried out on intelligent contract source codes by using grammatical features of the Pompe deception intelligent contract as slicing criteria, and sentences related to the Pompe deception semantic meanings are extracted to form a slice set; performing control flow and data flow analysis on the phrase set to construct an intelligent contract transaction attribute graph;
(4) performing feature extraction on the intelligent contract serialization input model by using a transform encoder as a feature extraction model to obtain an intelligent contract global feature vector;
(5) performing feature extraction on the intelligent contract transaction attribute graph by using a graph convolution neural network and graph self-attention pooling operation to obtain a local feature vector for intelligent contract transaction;
(6) and (5) fusing the global lexical feature vector and the local transaction feature vector of the intelligent contract obtained in the step (4) and the step (5) to obtain a composite feature vector of the intelligent contract, and inputting the composite feature vector of the intelligent contract into a multi-layer sensing machine to predict whether the current intelligent contract is a Pompe fraudster.
2. The intelligent contract pompe fraud detection method based on multi-modal features of claim 1, wherein said step (1) comprises the steps of:
(1.1) manually auditing the collected intelligent contract source codes, confirming whether the intelligent contract sample is a Pompe fraudster intelligent contract or not, and adding the intelligent contract confirmed as the Pompe fraudster intelligent contract into the Setponzi＝{c1,...,cN-wherein N represents the number of pointcast cheat intelligent contracts;
(1.2) collecting intelligent contract addresses through a BigQuery database, obtaining corresponding intelligent contract source codes through an Etherscan platform, manually confirming non-Ponz contract sample sets in the intelligent contract source codes, and forming the non-Ponz contract sample sets through the samplesno_ponzi＝{c1,...,cM-wherein M represents a non-pompe fraud intelligent contract sample number;
(1.1) integrating the intelligent contract samples obtained in the steps (1.1) and (1.2) to obtain an intelligent contract data Set (Set ═ Set)ponzi∪Setno_ponziAnd the number of intelligent contract samples in the dataset is S-M + N.
3. The intelligent contract pompe fraud detection method based on multi-modal features of claim 1, wherein said step (2) comprises the steps of:
(2.1) Using the solid compiler to sample the Smart contract ciCompiling to abstract syntaxTraversing the abstract syntax tree, extracting the subtree with the root node type of functional definition, and constructing the current intelligent contract sample ciAbstract syntax tree collection of included functions
Wherein K represents the current intelligent contract ciThe number of functions;
(2.2) abstracting the syntax tree of the function with a structured traversal
Expansion into token sequence
Wherein
Representing the length of a token sequence of the current function; merging token sequence sets of all functions to obtain a current intelligent contract sample ciIs shown in a serialized form
(2.3) merging the serialized representations of all the intelligent contract samples to form an intelligent contract corpus
Training on the corpus by using a FastText model to obtain any tokeniCorresponding initial feature vector
4. The intelligent contract pompe fraud detection method based on multi-modal features of claim 1, wherein said step (3) comprises the steps of:
(3.1) against Smart contract samples ciE Set and its function extractionElephant syntax tree set
Traversing all function abstract syntax trees using syntax analysis
Judging whether the current function contains transfer operation, if so, adding the function into the sensitive function set corresponding to the current contract sample
The rest of the functions forming a non-sensitive set of functions
(3.2) arbitrary function in the set of sensitivity functions
For its abstract syntax tree
Semantic analysis is carried out to obtain the function program dependence graph
Using 3 transfer transaction interfaces as slicing criteria, according to the principle of graph accessibility
Backward slicing is carried out to obtain a transaction slice attribute graph of the current function
And the node V epsilon V in the graph represents a function fiThe statement with semantic dependency relation with transaction behavior, wherein an edge E belongs to E in the graph, and the statement comprises the following components: control flow, data flow, control dependence and data dependence;
(3.3) traversal
And searching global variables used by all nodes in the current function fiTransaction sensitive global variable set of
Merging the transaction sensitive global variables in all the sensitive functions to obtain the current intelligent contract ciTransaction sensitive global variable set of
(3.4) traverse the current intelligent contract sample ciMiddle non-transaction sensitive function
Corresponding to the abstract syntax tree, judging whether the function uses transaction sensitive global variables
If used, the current function is aggregated from the non-transaction sensitive functions
Transitioning to transaction sensitive function set
And generating a transaction slice attribute graph of the current function by taking the transaction sensitive global variable used by the function as a slice criterion
(3.5) Global variables sensitive to arbitrary transactions
Traverse the currentThe intelligent contract finds the definition statement of the variable to form a global variable definition statement set Vdef(ii) a Merging the transaction slice attribute map of all transaction sensitive functions and the global variable definition statement set VdefGenerating a contract transaction attribute map TPG (V, E) of the current contract sample; wherein the node V belongs to V and represents statements such as transaction sensitive global variable definition, transaction sensitive global variable use, transaction interface calling and the like related to transaction behaviors in the current intelligent contract; the edge E E in the figure comprises: data flow, control flow, data dependency, control dependency, and transaction sensitive variable modification.
5. The intelligent contract pompe fraud detection method based on multi-modal features of claim 1, wherein said step (4) comprises the steps of:
(4.1) for any Intelligent contract sample ciGenerating (2.3) a serialized representation of the current sample
Dividing the batch into a plurality of batchs, wherein the number of tokens contained in each batch is l';
(4.2) calculating the position encoding vector PE of each token aiming at any batch, wherein the calculation method comprises the following steps:
PE(pos,2i)＝sin(pos/100002i/d)
PE(pos,2i+1)＝cos(pos/100002i/d)
wherein pos represents the position of the current token in batch; d is the dimension of the vector when token position encoding; i represents the index of the position-coding vector;
(4.3) adding the position coding vector of the token and the initial feature vector of the token to obtain the token in the current batchiInitialization vector of
Merging all token initialization vector representations in all batchs to obtain intelligent contract serialization representation
Wherein the content of the first and second substances,
representing the vector representation corresponding to the jth batch; d represents the feature vector dimension of each token; l' is the number of tokens in batch; n is a radical ofbatchA quantity of lots representing a sequence of intelligent contracts;
(4.4) serializing Intelligent contracts
Inputting the characteristics into a multi-head self-attention module for feature learning; the multi-head attention mechanism has J attention heads, and the dimension of the characteristic vector output by the multi-head self-attention module is dmultiheadFor each attention headiThe learning characteristics are as follows:
wherein the content of the first and second substances,
representing a matrix formed by feature vectors of all tokens in the jth batch; wi Q、Wi KAnd Wi VAre learnable parameters of the current attention head and have dimensions of
dhead＝dmultiheadJ; combining the results of all attention heads to obtain the output of the current multi-head attention mechanism:
wherein
Is a learnable parameter;
(4.5) outputting a multi-head attention mechanism
And the original input
Residual error connection is carried out and the residual error is input into a feedforward network through layer normalization operation to obtain the h-th global feature vector representation of the current intelligent contractg：
hffn_in＝LayerNorm(X+hmultihead(X))
Wherein, W1 ffn、
Parameters may be learned for a feed forward network.
6. The intelligent contract pompe fraud detection method based on multi-modal features of claim 1, wherein said step (5) comprises the steps of:
(5.1) for any Intelligent contract sample ciCorresponding trade attribute graph of e Set
Wherein any node represents a statement in a contract; the initial feature vector of any node in the graph is represented as: the sum of all token initial vectors in the current node; firstly, carrying out token sequence expansion on any node to obtain v ═ token { token1,...,tokennFourthly, calculating a node v in the graphiInitial feature vector of
Comprises the following steps:
(5.2) any node v in the graphiWith a neighboring node vje.N (i) passing edge e(i,j)Connection, current node viThe state update formula for the convolution layer in the t-th graph is:
wherein the content of the first and second substances,
cascading current node characteristics, neighbor node characteristics and the mutual relation between the two nodes as the input of state updating so that the model can learn the interaction between the nodes;
and
is a learnable parameter of the current convolutional layer; the final characteristic vector of any node v in the graph obtained by the T-layer graph convolution layer is expressed as
(5.3) utilization of the drawings
Each node V belongs to V initial characteristic vector
Final feature vector
Obtaining local feature vector of current intelligent contract
Where, conv stands for one-dimensional convolution operation,
is the number of nodes in the slice code property graph, σ (-) represents the activation function.
7. The intelligent contract pompe fraud detection method based on multi-modal features of claim 1, wherein said step (6) comprises the steps of:
(6.1) sample c for the current intelligent contractiCorresponding global features
And local features
Respectively calculating the corresponding attention weights in the following calculation modes:
wherein, Wg、Wl、bgAnd blIs a learnable parameter;
(6.2) weighting the global and local features by using the attention weight to obtain a final feature vector of the current intelligent contract
Represents:
and (6.3) inputting the generated final feature vector of the intelligent contract into a multilayer sensing machine, and judging whether the current intelligent contract is a Pompe deception intelligent contract.
8. An intelligent contract pompe fraud detection system based on multi-modal features using the method of any of claims 1-7, comprising:
the data set construction module is used for constructing an intelligent contract source code data set; collecting intelligent contract source codes by utilizing forum data, and marking whether the intelligent contract sample is a Pompe cheat or not by utilizing a manual auditing method;
the serialization representation module is used for obtaining the intelligent contract serialization representation; firstly, carrying out lexical analysis on intelligent contract codes to obtain an abstract syntax tree of the intelligent contract codes, and then carrying out serialized expansion by using a structured traversal technology to obtain code serialized representation;
the transaction attribute graph extraction module is used for extracting local semantic features related to transaction behaviors in the intelligent contract; the nodes in the transaction attribute graph are composed of statements related to transaction behaviors; the edges in the transaction attribute graph comprise a plurality of semantic types such as a data dependence edge, a data flow edge, a control flow and a control dependence edge;
the intelligent contract global feature extraction module is used for extracting global lexical features of the intelligent contracts; extracting global features from the intelligent contract serialization representation by using a transformer encoder model;
the intelligent contract local transaction feature extraction module is used for extracting transaction related local features from the intelligent contract transaction attribute graph by utilizing a graph convolution neural network and graph self-attention pooling;
the multi-mode feature fusion detection module is used for fusing global features and local transaction features by using a multi-head attention mechanism to obtain a current intelligent contract composite feature representation; and inputting the composite characteristics into a multi-layer sensing machine to judge whether the current intelligent contract is a Pompe fraudster or not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210165284.4A CN114547611A (en) | 2022-02-18 | 2022-02-18 | Intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210165284.4A CN114547611A (en) | 2022-02-18 | 2022-02-18 | Intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114547611A true CN114547611A (en) | 2022-05-27 |
Family
ID=81677213
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210165284.4A Withdrawn CN114547611A (en) | 2022-02-18 | 2022-02-18 | Intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114547611A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115268994A (en) * | 2022-07-26 | 2022-11-01 | 中国海洋大学 | Code feature extraction method based on TBCNN and multi-head self-attention mechanism |
| CN117473170A (en) * | 2023-12-27 | 2024-01-30 | 布比（北京）网络技术有限公司 | Intelligent contract template recommendation method and device based on code characterization and electronic equipment |
| CN117521065A (en) * | 2023-11-02 | 2024-02-06 | 海南大学 | Block chain decentralization finance safety detection method and device |
| CN117574214A (en) * | 2024-01-15 | 2024-02-20 | 中科链安(北京)科技有限公司 | Intelligent contract classification model training method, intelligent contract classification method and device |
-
2022
- 2022-02-18 CN CN202210165284.4A patent/CN114547611A/en not_active Withdrawn
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115268994A (en) * | 2022-07-26 | 2022-11-01 | 中国海洋大学 | Code feature extraction method based on TBCNN and multi-head self-attention mechanism |
| CN117521065A (en) * | 2023-11-02 | 2024-02-06 | 海南大学 | Block chain decentralization finance safety detection method and device |
| CN117473170A (en) * | 2023-12-27 | 2024-01-30 | 布比（北京）网络技术有限公司 | Intelligent contract template recommendation method and device based on code characterization and electronic equipment |
| CN117473170B (en) * | 2023-12-27 | 2024-04-09 | 布比（北京）网络技术有限公司 | Intelligent contract template recommendation method and device based on code characterization and electronic equipment |
| CN117574214A (en) * | 2024-01-15 | 2024-02-20 | 中科链安(北京)科技有限公司 | Intelligent contract classification model training method, intelligent contract classification method and device |
| CN117574214B (en) * | 2024-01-15 | 2024-04-12 | 中科链安(北京)科技有限公司 | Intelligent contract classification model training method, intelligent contract classification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Neural symbolic reader: Scalable integration of distributed and symbolic representations for reading comprehension | |
| CN111428044B (en) | Method, device, equipment and storage medium for acquiring supervision and identification results in multiple modes | |
| Nafi et al. | Clcdsa: cross language code clone detection using syntactical features and api documentation | |
| CN111639344B (en) | Vulnerability detection method and device based on neural network | |
| CN114547611A (en) | Intelligent contract Pompe fraudster detection method and system based on multi-modal characteristics | |
| CN113360915A (en) | Intelligent contract multi-vulnerability detection method and system based on source code graph representation learning | |
| CN112199512B (en) | Scientific and technological service-oriented case map construction method, device, equipment and storage medium | |
| CN114547619A (en) | Vulnerability repairing system and method based on tree | |
| CN113904844B (en) | Intelligent contract vulnerability detection method based on cross-mode teacher-student network | |
| CN113127933B (en) | Intelligent contract Pompe fraudster detection method and system based on graph matching network | |
| Hao et al. | A novel method using LSTM-RNN to generate smart contracts code templates for improved usability | |
| CN112990443A (en) | Neural network evaluation method and device, electronic device, and storage medium | |
| Rajbahadur et al. | Pitfalls analyzer: quality control for model-driven data science pipelines | |
| CN116702157A (en) | Intelligent contract vulnerability detection method based on neural network | |
| CN116383832A (en) | Intelligent contract vulnerability detection method based on graph neural network | |
| CN116662991A (en) | Intelligent contract intention detection method based on artificial intelligence | |
| Xia et al. | Source Code Vulnerability Detection Based On SAR-GIN | |
| CN117556425B (en) | Intelligent contract vulnerability detection method, system and equipment based on graph neural network | |
| Ouyang et al. | Binary vulnerability mining based on long short-term memory network | |
| CN116578989B (en) | Intelligent contract vulnerability detection system and method based on deep pre-training neural network | |
| CN117435246B (en) | Code clone detection method based on Markov chain model | |
| CN114444509B (en) | Method, device and equipment for testing named entity recognition model and storage medium | |
| CN116305119A (en) | APT malicious software classification method and device based on predictive guidance prototype | |
| CN117725592A (en) | Intelligent contract vulnerability detection method based on directed graph annotation network | |
| CN116502231A (en) | Intelligent contract vulnerability detection method based on multistage target critical path |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220527 |
|
| WW01 | Invention patent application withdrawn after publication |