Blockchain anonymous transaction tracking method based on node influence, electronic device and storage device
Patent Information
- Application Number
- CN202310760079.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-26
- Publication Date
- 2026-11-03
- Estimated Expiration
- 2043-06-26
AI Technical Summary
Existing blockchain transaction tracking methods are inadequate in terms of universality, efficiency, and accuracy, especially in terms of difficulty in regulating and tracking illegal transactions, and they consume too much computing resources.
A blockchain transaction tracking method based on node influence is adopted. It utilizes the Simplified Approximate Personalized Predictive Propagation (SA3P) model and reverse reachability set, combined with simplified graph convolutional networks and random walk techniques. By processing the address clustering and adjacency matrix of the transaction graph, node influence is calculated to achieve efficient tracking of transaction paths.
It achieves efficient and universally applicable transaction tracking, reduces computing resource requirements, improves the accuracy and tracking efficiency of transaction paths, and can effectively identify abnormal nodes and recover funds.
Smart Images
Figure CN116823470B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of blockchain, specifically to a blockchain anonymous transaction tracking method, electronic device, and storage device based on node influence. Background Technology
[0002] Blockchain technology is a distributed database technology that integrates peer-to-peer networks, cryptography, and consensus mechanisms. Its decentralized and distributed structure endows it with the characteristics of immutability, transparency, full traceability, and verifiability. Based on these advantages, blockchain has developed rapidly in fields such as finance, healthcare, and management.
[0003] With the rapid development of blockchain technology, transactions on the blockchain have become more frequent, attracting an increasing number of illegal transactions, such as financial fraud, money laundering, and cryptocurrency theft. Because blockchain transactions are conducted anonymously without identity verification, they offer great convenience and lay the foundation for their popularity. However, the anonymity of blockchain is a double-edged sword, easily exploited by cybercriminals, becoming a breeding ground for illegal activities. Furthermore, the current regulatory landscape for blockchain is relatively chaotic, requiring regulatory agencies to strengthen oversight and standardization to protect investor interests and maintain market order. When illegal activities occur, timely prediction and tracking of transaction fund flows, identification of abnormal nodes, and recovery of funds are crucial. Currently, there are many methods for transaction tracking. Mainstream research methods include rule-based heuristics and pollution / coloring-based tracking analysis methods, but these methods are based on specific scenarios and have limitations in terms of universality, effectiveness, and efficiency. Heuristic clustering methods are insufficient for judging address changes, and research on relationships between users is also scarce. Pollution-based tracking analysis methods are prone to misclassifying "clean money" as "dirty money," thus affecting transaction tracking. Furthermore, existing solutions may require significant computational resources and time to complete the tracking task.
[0004] As mentioned above, existing transaction tracking solutions still have shortcomings. Some methods are not accurate enough, which greatly reduces tracking efficiency, and some methods require a lot of computing resources.
[0005] In light of the above, it is necessary to propose a blockchain anonymous transaction tracking method based on node influence to achieve high efficiency and high universality in blockchain transaction tracking. Summary of the Invention
[0006] To address the aforementioned problems, this invention proposes a blockchain transaction tracking method, electronic device, and storage device based on node influence, achieving efficient transaction tracking. The method includes the following steps:
[0007] Step 1: Process the blockchain transaction graph using any address clustering method based on the transaction graph to obtain a directed graph of account transactions. Extract the transaction amount and number of transactions between accounts and store them in the corresponding dictionary.
[0008] Step 2: Combine the transaction amount and number of transactions obtained in Step 1 according to a certain ratio to obtain the corresponding value, and store the value in the matrix according to the node number.
[0009] Step 3: Process the matrix and node feature matrix obtained in Step 2 using the Simplified and Approximate Personalized Propagation of Prediction (SA3P) model to obtain the final node influence.
[0010] Step 4: Based on the initial node, use the reverse reachability set to track transactions, obtain all possible fund flow paths, and finally find the paths with the greater average influence of the nodes to determine which accounts the target account is most likely to transact with, thereby achieving the purpose of transaction tracking.
[0011] Furthermore, the specific process of step 1 includes:
[0012] Step 1.1: Using a transaction graph-based address clustering method, the transaction graph is clustered to obtain a directed graph of account transactions. .in, , These are the node set and edge set of the graph, respectively. , Indicates the number of nodes. , Indicates the number of sides. Indicates the blockchain account address. This indicates a transaction between two account addresses. Represents a node , Indicates from node To the node Directed edges, i.e., nodes Funds flow to nodes middle.
[0013] Step 1.2, withdraw transaction amounts between accounts and number of transactions Store them in the corresponding dictionary , Among them Indicates from node Transfer to node The transaction amount Indicates from node Transfer to node The transaction amount Indicates from node To the node Number of transactions Similarly.
[0014] Furthermore, the specific process of step 2 includes:
[0015] Step 2.1, use the transaction amount obtained in Step 1. and number of transactions After normalization, we get and :
[0016] (1)
[0017] (2)
[0018] in and for and The result of normalization, , . For dictionary The minimum value in, For dictionary The maximum value in.
[0019] Step 2.2: Calculate the edge weights based on transaction amount and number of transactions. :
[0020] (3)
[0021] Step 2.3, will Form a weighted adjacency matrix , In most cases, it is an asymmetric matrix.
[0022] Furthermore, the specific process of step 3 includes:
[0023] The Simplified Approximate Personalized Prediction Propagation (SA3P) model is a graph convolutional neural network-based prediction model that simplifies graph convolution operations by using adjacency matrix polynomials and adding personalized information through personalized matrices for more accurate predictions. This strategy consists of two parts: a propagation phase and a prediction phase.
[0024] Step 3.1, the propagation phase, uses an Attention-Based Simplified Graph Convolutional Network (ASGCN) to compute the final representation of node features. Introducing the attention mechanism and weight matrix in this step allows the model to better capture node feature information and improve classification accuracy, while effectively addressing overfitting caused by excessive neural network layers.
[0025] Step 3.2, the prediction stage, uses personalized propagation prediction based on random walk to obtain the final predicted value.
[0026] Furthermore, the specific process of step 3.1 includes:
[0027] Step 3.1.1, the formula for calculating the adjacency matrix polynomial attention function is:
[0028] (4)
[0029] in, Given an adjacency matrix with attention weights, consider its own features. , It is an n-dimensional identity matrix. For a weighted normalized adjacency matrix, since Since it is a directed graph, this invention uses and Normalize the weighted adjacency matrix to obtain ,in and They represent The in-degree and out-degree matrices, and Both are diagonal matrices. , , for The k-th power.
[0030] Using a weighted normalized adjacency matrix Various powers, and through an attention mechanism To make flexible adjustments The weights of various powers are used to obtain The feature information of all nodes is skipped to avoid oversmoothing.
[0031] Step 3.1.2, in order to utilize To extract feature information from all nodes by skipping distances, this invention designs a feature propagation mechanism based on attention. It utilizes an adjacency matrix with attention weights. Update the node feature matrix along these neighboring nodes in different neighborhoods. :
[0032] (5)
[0033] This invention uses a series of attention scores to adjust the importance of each node's features in different neighborhoods, making full use of the node's neighborhood information and improving model performance to a certain extent.
[0034] Step 3.1.3: To prevent overfitting, this invention adds a dropout layer after the feature extraction layer. The dropout layer is a regularization technique that randomly sets the output of neurons to 0, thereby reducing dependencies between neurons and preventing overfitting. Here, dropout is used as an effective feature enhancement technique, which can increase the model's generalization ability and reduce its sensitivity to noise and outliers in the training data. During training, Bernoulli sampling is used to randomly set some elements of the input tensor to zero with probability η. In each training iteration, the dropout layer randomly selects some neurons and sets their outputs to 0.
[0035] (6)
[0036] in, Indicates the product of elementalization. Let be a vector of independent Bernoulli random variables, each with a probability of 1. After applying dropout, the final features of the nodes in the simplified graph convolutional neural network are obtained. :
[0037] (7)
[0038] in, It is a non-linear activation function. The feature matrix is a matrix composed of the feature vectors of each node. Typically, node feature vectors need to be extracted manually or generated using a pre-trained model. The weight matrix... It is learned and can be adjusted according to different tasks and datasets.
[0039] Furthermore, the specific process of step 3.2 includes:
[0040] Step 3.2.1, first let:
[0041] (8)
[0042] in, For the process The node representation after the next iteration. This represents the initial node. It is a simplified graph convolutional network (ASGCN) based on the attention mechanism.
[0043] Step 3.2.2, then use the random transition matrix. The random walk is simulated, and propagation probability is introduced in the process. It is used to control the depth and direction of the random walk.
[0044] (9)
[0045] Step 3.2.3, using a random transition matrix and bias Iterate over the initial node representation ,get:
[0046] (10)
[0047] Step 3.2.4: Finally, a non-linear activation function is used to obtain the final predicted value from the final representation of the node.
[0048] (11)
[0049] (12)
[0050] in, This is the final representation of the node. For the softmax function, This is the final predicted value.
[0051] Step 3.2.5: Process the final predicted values according to each attribute to obtain the node influence level. For nodes The extent of its influence.
[0052] Furthermore, the specific process of step 4 includes:
[0053] Step 4.1: Filter out all nodes with an out-degree of zero and add them to the node set. middle.
[0054] Step 4.2: Using the reverse reachability set approach, perform a reverse search starting from the node with an out-degree of zero, and limit the path length of the search to [value missing]. Output the found path.
[0055] Step 4.3: Calculate the influence of each output path based on the output node. For path The influence of a node is the sum of the influences of all nodes along the path. This is calculated based on the number of nodes in the path. Average influence size of each node ( For path The influence of each node.
[0056] Step 4.4: Select nodes based on their influence along the path using a greedy algorithm. The path with the greatest average influence for each node is used to track transactions.
[0057] The present invention also proposes an electronic device, which includes a control module and a storage module, wherein the control module is capable of executing the program instructions corresponding to the above method.
[0058] The present invention also proposes a storage device that has a built-in program for the method described above.
[0059] The beneficial effects of this invention are:
[0060] 1. The SA3P model used in this invention not only solves the overfitting problem caused by too many layers in the neural network;
[0061] 2. A reverse search method was used to make the search more comprehensive;
[0062] 3. A random transition matrix is also introduced to simulate random walks. It incorporates the transmission probability, which can solve the problem of using global features to replace local features of the root node due to excessive iterations, and realize the tracking of important blockchain transactions. Attached Figure Description
[0063] Figure 1 Transaction tracking methodology framework diagram
[0064] Figure 2 ASGCN Structure Diagram
[0065] Figure 3 Transaction tracking flowchart
[0066] Figure 4 Transaction tracking process functional module diagram Detailed Implementation
[0067] This invention proposes a blockchain transaction tracking method, electronic device, and storage device based on node influence. The method includes: Step 1, processing the blockchain transaction graph using an arbitrary address clustering method based on the transaction graph to obtain a directed graph of account transactions. Transaction amounts and transaction counts between accounts are extracted and stored in corresponding dictionaries; Step 2, combining the transaction amounts and counts obtained in Step 1 according to a certain ratio to obtain corresponding values, and storing these values in a matrix according to node numbers; Step 3, processing the matrix obtained in Step 2 and the node feature matrices using a Simplified and Approximate Personalized Propagation of Prediction (SA3P) model to obtain the final node influence; Step 4, tracking transactions using the reverse reachability set based on the initial nodes to obtain all possible fund flow paths, and finally finding several paths with higher average node influence to determine which accounts the target account is most likely to transact with, thereby achieving the purpose of transaction tracking. This invention is applicable to various transaction tracking scenarios, generally does not require graph preprocessing, and can effectively avoid overfitting, achieving efficient and accurate transaction tracking.
[0068] The following is in conjunction with the appendix Figure 1-4 To further illustrate the present invention, it should be noted that the specific implementation of the present invention is based on this technology, and a detailed implementation process and steps are provided. However, the scope of protection of the present invention is not limited to this implementation example.
[0069] like Figure 1 and Figure 3 The transaction tracking process of this invention is shown below:
[0070] (1) Process the blockchain transaction graph using any address clustering method based on the transaction graph to obtain a directed graph of account transactions. Extract the transaction amount and number of transactions between accounts and store them in the corresponding dictionary.
[0071] (1-1) A transaction graph-based address clustering method is used to cluster the transaction graph to obtain a directed graph of account transactions. .in, , These are the node set and edge set of the graph, respectively. , , Indicates the blockchain account address. This represents a transaction between two account addresses. The set... elements Represents a node , gather elements Indicates from node To the node Directed edges, i.e., nodes Funds flow to nodes middle.
[0072] (1-2) Withdraw transaction amounts between accounts and number of transactions Store them in the corresponding dictionary , Among them Indicates from node Transfer to node The transaction amount Indicates from node Transfer to node The transaction amount Indicates from node To the node Number of transactions Indicates from node To the node The number of transactions.
[0073] (2) Transfer the transaction amount obtained in step 1 and number of transactions Combine them according to a certain ratio (as shown in step 2-2) to obtain the corresponding value, and store the value in the matrix according to the node number.
[0074] (2-1) The transaction amount obtained in step 1 and number of transactions After normalization, we get and :
[0075] (1)
[0076] (2)
[0077] in and for and The result of normalization, , . For dictionary The minimum value in, For dictionary The maximum value in. For dictionary The minimum value in, For dictionary The maximum value in.
[0078] (2-2) Calculate the edge weight based on transaction amount and number of transactions. :
[0079] (3)
[0080] in This is a balancing factor used to balance the weights of the number of trades and the transaction amount. When At that time, edge weight Only subject to transaction amount Impact; when At that time, edge weight Only affected by the number of transactions Influence.
[0081] (2-3) will Form a weighted adjacency matrix ,in The row number. The column number is the column number. In most cases, it is an asymmetric matrix.
[0082] (3) The matrix and node feature matrix obtained in step 2 are processed using the Simplified and Approximate Personalized Propagation of Prediction (SA3P) model to obtain the final node influence.
[0083] Among them, the Simplified Approximate Personalized Prediction Propagation (SA3P) model is a graph convolutional neural network-based prediction model that simplifies graph convolution operations by using adjacency matrix polynomials and adds personalized information to the prediction using a personalized matrix, resulting in more accurate predictions. This strategy consists of two parts: a propagation phase and a prediction phase.
[0084] (3.1) Propagation Phase: In this phase, an Attention-Based Simplified Graph Convolutional Network (ASGCN) is used to compute the final representation of node features. The network structure diagram of ASGCN is shown below. Figure 2 As shown. Step 3.1 introduces an attention mechanism and a weight matrix, which enables the model to better capture node feature information and improve classification accuracy, while effectively solving the overfitting problem caused by too many neural network layers. The specific process is as follows:
[0085] (3.1.1) The formula for calculating the adjacency matrix polynomial attention function is as follows:
[0086] (4)
[0087] in, Its adjacency matrix Attention weight values. Given an adjacency matrix with attention weights, consider its own features. , It is an n-dimensional identity matrix. For a weighted normalized adjacency matrix, since Since it is a directed graph, this invention uses and Normalize the weighted adjacency matrix to obtain ,in and They represent The in-degree matrix and the out-degree matrix, and Both are diagonal matrices. , , for The k-th power.
[0088] Using a weighted normalized adjacency matrix Various powers, and through attention mechanisms To make flexible adjustments The weights of various powers are used to obtain The feature information of all nodes is skipped to avoid oversmoothing.
[0089] (3.1.2) In order to utilize To extract feature information from all nodes by skipping distances, this invention designs a feature propagation mechanism based on attention. It utilizes an adjacency matrix with attention weights. Update the node feature matrix along these neighboring nodes in different neighborhoods. The updated result is denoted as :
[0090] (5)
[0091] This invention uses a series of attention scores to adjust the importance of each node's features in different neighborhoods, making full use of the node's neighborhood information and improving model performance to a certain extent.
[0092] (3.1.3) To prevent overfitting, this invention adds a dropout layer after the feature extraction layer. The dropout layer is a regularization technique that randomly sets the output of neurons to 0, thereby reducing dependencies between neurons and preventing overfitting. Here, dropout is used as an effective feature enhancement technique, which can increase the model's generalization ability and reduce its sensitivity to noise and outliers in the training data. During training, Bernoulli sampling is used to randomly set some elements of the input tensor to zero with probability η. In each training iteration, the dropout layer randomly selects some neurons and sets their outputs to 0.
[0093] (6)
[0094] in, Indicates the product of elementalization. Let be a vector of independent Bernoulli random variables, each with a probability of 1. After applying dropout, the final features of the nodes in the simplified graph convolutional neural network are obtained. :
[0095] (7)
[0096] in, It is a non-linear activation function. The feature matrix is a matrix composed of the feature vectors of each node. Typically, node feature vectors need to be extracted manually or generated using a pre-trained model. The weight matrix... It is learned and can be adjusted according to different tasks and datasets.
[0097] (3.2) In the prediction stage, personalized propagation prediction based on random walk was used to obtain the final predicted value.
[0098] The specific process is as follows:
[0099] (3.2.1) First, let:
[0100] (8)
[0101] in, For the process The node representation after the next iteration. This represents the initial node. It is a simplified graph convolutional network (ASGCN) based on the attention mechanism.
[0102] (3.2.2) Next, a random transition matrix is used. The random walk is simulated, and propagation probability is introduced in the process. It is used to control the depth and direction of the random walk.
[0103] (9)
[0104] (3.2.3) Using a random transition matrix and bias Iterate over the initial node representation ,get:
[0105] (10)
[0106] (3.2.4) Finally, a non-linear activation function is used to obtain the final predicted value for the node's final representation:
[0107] (11)
[0108] (12)
[0109] in, This is the final representation of the node. For the softmax function, This is the final predicted value.
[0110] (3.2.5) The final predicted value is processed according to each attribute to obtain the node influence size. For nodes The extent of its influence.
[0111] (4) Based on the initial node, transaction tracking is performed using the reverse reachable set to obtain all possible fund flow paths. Finally, the paths with the highest average node influence are identified to determine which accounts the target account is most likely to transact with, thereby achieving the purpose of transaction tracking. The specific process is as follows:
[0112] (4.1) Filter out all nodes with an out-degree of zero and add them to the node set. middle.
[0113] (4.2) Use the reverse reachability set approach, starting from the node with an out-degree of zero, to perform a reverse search, and limit the path length of the search to... Output the found path.
[0114] (4.3) Calculate the influence of each output path based on the output node. For path The influence of a node is the sum of the influences of all nodes along the path. This is calculated based on the number of nodes in the path. Average influence size of each node ( For path The influence of each node.
[0115] (4.4) Select the nodes according to their influence on the path using a greedy algorithm. The path with the greatest average influence for each node is used to track transactions.
[0116] This invention also proposes an electronic device, which includes a control module and a storage module, such as... Figure 4 As shown, the control module can execute the program instructions corresponding to the above methods; for example, computer devices such as network servers.
[0117] The present invention also proposes a storage device that has the program of the above method built into it; for example, an optical disc.
[0118] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent methods or modifications that do not depart from the technology of the present invention should be included within the scope of protection of the present invention.
Claims
1. A blockchain anonymous transaction tracking method based on node influence, characterized in that, Includes the following steps: Step 1: Process the blockchain transaction graph using any address clustering method based on the transaction graph to obtain a directed graph of account transactions, extract the transaction amount and number of transactions between accounts, and store them in the corresponding dictionary; Step 2: Combine the transaction amount and number of transactions obtained in Step 1 according to a certain ratio to obtain the corresponding value, and store the value in the matrix according to the node number; Step 3: Process the matrix and node feature matrix obtained in Step 2 using the Simplified and Approximate Personalized Propagation of Prediction (SA3P) model to obtain the final node influence; this includes the following steps: Step 3.1, the propagation phase, uses an Attention-Based Simplified Graph Convolutional Network (ASGCN) to compute the final representation of node features. This step introduces an attention mechanism and weight matrix, enabling the model to better capture node feature information, improving classification accuracy, and effectively addressing overfitting caused by excessive neural network layers. The specific process includes: Step 3.1.1, the formula for calculating the adjacency matrix polynomial attention function is: (4) in, Given an adjacency matrix with attention weights, consider its own features. , It is an n-dimensional identity matrix. For a weighted normalized adjacency matrix, since Since it is a directed graph, we use and Normalize the weighted adjacency matrix to obtain ,in and They represent The in-degree and out-degree matrices, and Both are diagonal matrices. , , for k-th power; Using a weighted normalized adjacency matrix Various powers, and through attention mechanisms To adjust The weights of various powers are used to obtain By skipping the feature information of all nodes, the oversmoothing problem can be avoided; Step 3.1.2, in order to utilize To extract feature information from all nodes by skipping distances, we design a feature propagation mechanism based on attention, utilizing an adjacency matrix with attention weights. Update the node feature matrix along these neighboring nodes in different neighborhoods. : (5) A series of attention scores are used to adjust the importance of each node's features in different neighborhoods, making full use of the node's neighborhood information and improving model performance. Step 3.1.3: To prevent overfitting, a dropout layer is added after the feature extraction layer. This randomly sets the output of each neuron to 0, reducing the dependency between neurons and preventing overfitting. During training, Bernoulli sampling is used to randomly set certain elements of the input tensor to zero with probability η. In each training iteration, the dropout layer randomly selects some neurons and sets their output to 0. (6) in, Indicates the product of elementalization. Let be a vector of independent Bernoulli random variables, each with a probability of 1. After applying dropout, the final features of the nodes in the simplified graph convolutional neural network are obtained. : (7) in, It is a non-linear activation function. The feature matrix and the weight matrix are... It is learned and can be adjusted according to different tasks and datasets; Step 3.2, the prediction phase, uses personalized propagation prediction based on random walks to obtain the final predicted value; the specific process includes: Step 3.2.1, first let: (8) in, For the process The node representation after the next iteration. This represents the initial node. It is a simplified graph convolutional network based on the attention mechanism (ASGCN); Step 3.2.2, then use the random transition matrix. The random walk is simulated, and propagation probability is introduced in the process. Used to control the depth and direction of the random walk; (9) Step 3.2.3, using a random transition matrix and bias Iterate over the initial node representation ,get: (10) Step 3.2.4: Finally, a non-linear activation function is used to obtain the final predicted value from the final representation of the node. (11) (12) in, This is the final representation of the node. For the softmax function, This is the final predicted value; Step 3.2.5: Process the final predicted values according to each attribute to obtain the node influence level. For nodes The extent of their influence; Step 4: Based on the initial node, use the reverse reachability set to track transactions, obtain all possible fund flow paths, and finally find the paths with the greater average influence of the nodes to determine which accounts the target account is most likely to transact with, thereby achieving the purpose of transaction tracking.
2. The blockchain anonymous transaction tracking method based on node influence according to claim 1, characterized in that, The specific process of step 1 includes: Step 1.1: Use the address clustering method based on the transaction graph to cluster the transaction graph to obtain a directed graph of account transactions. ,in, , These are the node set and edge set of the graph, respectively. , Indicates the number of nodes. , Indicates the number of sides. Indicates the blockchain account address. This indicates a transaction between two account addresses, where, Represents a node , Indicates from node To the node Directed edges, i.e., nodes Funds flow to nodes middle; Step 1.2, withdraw transaction amounts between accounts and number of transactions Store them in the corresponding dictionary , Among them Indicates from node Transfer to node The transaction amount Indicates from node Transfer to node The transaction amount Indicates from node To the node Number of transactions Similarly.
3. The blockchain anonymous transaction tracking method based on node influence according to claim 1, characterized in that, Step 2 includes the following specific steps: Step 2.1, use the transaction amount obtained in Step 1. and number of transactions After normalization, we get and : (1) (2) in and for and The result of normalization, , , For dictionary The minimum value in, For dictionary The maximum value in; Step 2.2: Calculate the edge weights based on transaction amount and number of transactions. : (3) in As a balancing factor, it is used to balance the weights of the number of transactions and the transaction amount. At that time, edge weight Only subject to transaction amount Impact; when At that time, edge weight Only affected by the number of transactions Influence; Step 2.3, will Form a weighted adjacency matrix , It is usually an asymmetric matrix.
4. The blockchain anonymous transaction tracking method based on node influence according to claim 1, characterized in that, Step 4 includes the following specific steps: Step 4.1: Filter out all nodes with an out-degree of zero and add them to the node set. middle; Step 4.2: Using the reverse reachability set approach, perform a reverse search starting from the node with an out-degree of zero, and limit the path length of the search to [value missing]. The found path will be output. Step 4.3: Calculate the influence of each output path based on the output node. For path The influence is the sum of the influences of all nodes on the path; then the path is calculated according to the number of nodes. The average influence of each node For path The influence of each node; Step 4.4: Select nodes based on their influence along the path using a greedy algorithm. The path with the greatest average influence for each node is used to track transactions.
5. An electronic device, characterized in that, The electronic device is capable of executing program instructions corresponding to the method described in any one of claims 1-4.
6. A storage device, characterized in that, The storage device contains a program for the method described in any one of claims 1-4.