A blockchain anonymous tracking method and electronic device based on knowledge graph

Abstract

The present invention discloses a blockchain anonymous tracking method and electronic device based on a knowledge graph. First, known suspicious address accounts are screened according to transaction value to obtain the transaction time and corresponding address account range. The blockchain is then sliced ​​and analyzed based on the transaction flow and value. The graph obtained after the slice analysis is then subjected to dependency syntax tree analysis to extract the implicit logical rules in the graph. The GraphSAGE algorithm is then used to learn and obtain the vector representation of each node in low-dimensional space for use in logical reasoning and analysis tasks. The vector representation of the nodes is then evaluated by a relationship evaluation function to infer the potential logical relationship between the nodes. The K-shell algorithm is then used to evaluate the importance of the nodes in the blockchain network. Finally, a visualization tool is used to represent the importance or influence of the nodes to observe the relationship and influence range between the nodes, infer the hidden relationship between the address accounts, and thus discover the relationship between entities.

Description

Technical Field

[0001] The present invention relates to the fields of blockchain and knowledge graphs, and in particular to digital currency transaction analysis, and is suitable for discovering potential transaction relationships between digital currency addresses. Background Art

[0002] The concept of knowledge graphs was first proposed by Google in 2012. With the rapid development of internet technology and its applications, knowledge graphs, with their strong semantic expression and reasoning capabilities, have attracted widespread attention from researchers and large enterprises, and have been successfully applied in fields such as search engines and intelligent question-answering. Blockchain technology is the core supporting technology for digital cryptocurrency systems, represented by digital currencies. With the rapid development and popularization of digital currencies in recent years, the research and application of blockchain technology has also shown rapid growth, and it is considered one of the most disruptive innovative technologies with the greatest potential to trigger the next industrial revolution.

[0003] Current methods for anonymous blockchain tracing primarily rely on address clustering and transaction flow analysis, typically requiring the development of corresponding transaction network models such as Petri networks, bipartite graphs, and address attribute graphs. However, the emergence and continued development of blockchain coin mixing technology, particularly coin join, has resulted in the inclusion of unlabeled empty addresses and relay addresses in the address space to be clustered, significantly increasing the difficulty of tracking and monitoring anonymous blockchain transactions. While there are detection schemes tailored to different coin mixing methods, these are inefficient and require continuous improvement with each new coin mixing method.

[0004] Knowledge graphs are a typical interdisciplinary field, representing the systematic integration of technologies from multiple fields, including knowledge representation learning, machine learning, natural language processing, and information acquisition. The construction of a knowledge graph involves steps such as knowledge acquisition, entity recognition, relationship extraction, attribute completion, and concept extraction. Knowledge graphs are also a rapidly developing new field, constantly integrating, updating, and advancing with fields such as graph neural networks, federated learning, and machine vision. Because knowledge graphs can describe the objective world and abstract the logical relationships between all things, they possess strong reasoning capabilities in natural language and word meaning. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this paper proposes a knowledge graph-based anonymous blockchain tracking method. This knowledge graph construction, reasoning, and update method can meet the needs of blockchain transaction data, which is constantly updated, and the development of coin mixing technology. Compared with traditional blockchain anonymous tracking methods, it can significantly reduce time and space overhead and improve the efficiency of blockchain anonymous tracking. First, known suspicious addresses and accounts are screened according to transaction value to obtain transaction time and corresponding address account ranges. The blockchain is then sliced ​​and analyzed based on transaction flow and value. Dependency syntax tree analysis is then performed on the graph obtained from the slice analysis to extract the logical rules implicit in the graph. Because previous representations typically require all nodes to be present in the training graph during embedding training, they cannot be naturally generalized to unseen nodes in the graph. Since blockchain networks are constantly updated, simply using an invariant graph for analysis is not feasible. The Graph SAGE algorithm uses node feature information to generate feature representations for nodes that are unseen during training. Therefore, after extracting the logical rules, this paper uses the Graph SAGE algorithm for embedded learning to obtain vector representations of each node in a low-dimensional space for use in logical reasoning and analysis tasks. The vector representation of the node is evaluated by the relationship evaluation function to infer the potential logical relationship between the nodes. Since traditional studies on the discovery of important nodes in the network emphasize the influence of the node's own attributes (such as degree, betweenness, etc.), they ignore the influence condensed by the node's neighbors, which should be a very important influencing factor. The KSN algorithm (the K-shell and Neighborhood Centrality) believes that the influence of a node is determined not only by the node's own K-shell attributes, but also by the K-shell attributes of its second-order neighbor nodes. Therefore, the present invention uses the K-shell algorithm to evaluate the influence or importance of a node in the blockchain network. Finally, a visualization tool is used to represent the importance or influence of the node to observe the potential relationship and influence range between nodes, and finally to infer the hidden relationship between address accounts, thereby discovering the relationship between entities.

[0006] To achieve the above objectives, the specific technical solution of the present invention is as follows: A blockchain anonymous tracking method based on knowledge graph: comprising the following steps:

[0007] 1. Parse the blockchain slice file to obtain the corresponding transaction information, such as amount, address account, timestamp, etc.

[0008] 2. According to the time and amount of the suspicious event, the blockchain slice file corresponding to the time of the suspicious event and the amount involved is filtered to obtain the corresponding address account range. The hidden rule between address accounts is defined as R x , the rule set in the dependency syntax tree rule base is defined as R = {R1,…,Rn}, by adding rule R x The corresponding logical relationship is obtained by matching the rules in the rule base.

[0009] 3. The logical relationship and corresponding address account obtained in step 2 above are used to perform embedded learning using the Graph SAGE algorithm to generate vectorized feature representations between nodes for knowledge reasoning.

[0010] 4. Perform heuristic entity recognition on the blockchain data parsed in step 1 to obtain the blockchain entity corresponding to each account address and identify them in sequence.

[0011] 5. Use machine learning to train the relationships between nodes to analyze and infer the potential relationships between entities from suspicious addresses.

[0012] 6. The KSN algorithm measures the importance of nodes with more comprehensive network topology characteristics.

[0013] 7. Use the Gephi algorithm to color the importance of nodes to different degrees to achieve visualization of transaction tracking.

[0014] Furthermore, in step 1 above, blockchain parsing includes the following steps:

[0015] Step 1.1 Download transaction data through the wallet of the corresponding blockchain.

[0016] Step 1.2: Perform structured parsing of the block data, extract the key fields of the parsed digital currency data and store them in the Key-Value database.

[0017] Furthermore, in step 2 above, screening of suspicious blockchain events includes the following steps:

[0018] Step 2.1 First, record the suspicious address or account as A = {a1,…,a n}, where a i Represents the i-th address or account; the transaction value of the address or account is recorded as m={M1,…,M n}, M i Indicates the start and end conditions of the slice α i The value of the transaction.

[0019] Step 2.2 defines the rule set in the dependency syntax rule base as R = {R1,…,R n}, the hidden rules in the graph are denoted as R x , by adding rule R x Match the rules in the rule base to obtain the corresponding logical relationship.

[0020] Furthermore, in step 3 above, the embedded learning of the Graph SAGE algorithm includes the following steps:

[0021] Step 3.1: Read the corresponding data (such as timestamp, address or account relationship, etc.) from the parsed blockchain database.

[0022] Step 3.2 processes the read data according to certain rules. First, the information of the corresponding address or account (such as timestamp, address activity, number of transactions, and whether it is a malicious address type, etc.) is processed to convert it into a low-dimensional vector of the same dimension.

[0023] Step 3.3 represents the address accounts in the blockchain network as an adjacency matrix A according to the logical relationship, where each address account corresponds to a unique ID, and each ID corresponds to the initial feature representation vector of the address account.

[0024] Step 3.4 uses the adjacency matrix A (representing the logical relationships between addresses and accounts) generated in Step 3.3, along with the initial vectorized representations of each address and account (also known as node features), as input to the Graph SAGE algorithm. Here, each address and account is considered a node in the graph structure. Through repeated iterations of the Graph SAGE algorithm, a vector representation of each node is ultimately generated.

[0025] Furthermore, the execution of Graph SAGE in step 3.4 can be divided into the following steps:

[0026] Step 3.4.1 first uniformly samples a fixed-size neighborhood set for each node, extracting different uniform samples at each iteration k.

[0027] Step 3.4.2 Next, for each node (V is a set of nodes), in its direct neighborhood The representation of the aggregate node is converted into a vector The vector representation when k=0 is defined as the input node feature. This step can be expressed as:

[0028]

[0029] in represents the feature representation of the aggregation field, u represents the neighbors of node v, N(v) represents the node neighborhood composed of neighbors, Represents the current neighbor node feature, and generates the current feature representation of the neighborhood after the aggregation function

[0030] Step 3.4.3 After aggregating adjacent feature vectors, Graph SAGE converts the current representation of the node into and aggregated neighborhood vectors This connection vector is fed through the fully connected layer of the nonlinear activation function σ and converted into the parameters used in the next iteration of the algorithm, representing This step can be expressed as:

[0031]

[0032] in Represents the input feature representation in the next iteration, W k represents the weight matrix of the kth layer, and σ represents the activation function Sigmoid.

[0033] Step 3.4.4 After k iterations of Graph SAGE, normalize the feature representation Finally, the feature representation of each node after k iterations is generated

[0034] The specific process of the Graph SAGE algorithm is as follows Figure 2 shown.

[0035] Furthermore, in step 4 above, the process of identifying the blockchain entity includes the following steps:

[0036] Step 4.1 performs heuristic entity recognition on all blockchain address accounts according to transaction type.

[0037] Furthermore, the heuristic entity recognition in step 4.1 can be divided into the following steps:

[0038] Step 4.1.1 defines the set of all transactions as T, T = {t1, t2, t3, ..., t m}, m represents the total number of transactions.

[0039] Step 4.1.2 defines each transaction as a five-tuple t=(I, O, V I ,V O ,F), I represents the input address account set, I={i1,i2,i3,…,i n}, n is the number of input address accounts; O={o1,o2,o3,…,o k}, where O represents the output address account set, k is the number of output address accounts; V I Represents the total input value of all address accounts, V O Represents the total output value of all address accounts, F represents the transaction fee, where V I , V O , the calculation of F set is shown in formulas 3, 4, and 5:

[0040]

[0041]

[0042] in Represents a single input address account i s The input value, Represents a single input address account o s output value.

[0043] Step 4.2 identifies entities according to different transaction types of the blockchain, thereby clustering different address accounts into different entities.

[0044] Furthermore, in step 5 above, the evaluation of the vector representation relationship between nodes includes the following steps:

[0045] Step 5.1: Each node corresponds to its feature representation one by one, extract the corresponding relations and entities from the network, and retain 30% of the test set.

[0046] Step 5.2 uses DisMult as the scoring function. The definition of the scoring function is as follows:

[0047]

[0048] where e s and e o It is the node feature representation learned using Graph SAGE, R r is the diagonal matrix corresponding to the relation r, s, r, and o are the feature vector representations of the head entity node, relation, and tail entity node, respectively.

[0049] Step 5.3 uses negative sampling to train the machine learning model. For each positive sample, sample ω negative samples (replace s or o with other entities). By optimizing the cross-entropy loss function, the model scores the positive samples higher than the negative samples. The specific definition is as follows:

[0050]

[0051] in is the loss, T is the set of positive and negative samples, l is the sigmoid function, y = 1 represents a positive sample, and y = 0 represents a negative sample.

[0052] Step 5.4 uses the trained model to detect suspicious addresses to determine whether there is a potential relationship.

[0053] The specific process of step 5 node evaluation is as follows Figure 3 shown.

[0054] Furthermore, in step 6 above, the evaluation of node importance includes the following steps:

[0055] Step 6.1 treats the blockchain network as an undirected graph and generates a set of neighbor nodes within the second order for each node o, that is, the set of neighbor nodes that can be reached within 2 steps from node v. This set is denoted as τ(v).

[0056] In step 6.2, the K-shell value of node v is denoted as ks(v). The K-shell value explains the hierarchical characteristics of the network by recursively removing all nodes with degree k from the edge nodes of the network. The nodes at the core have higher influence.

[0057] Further, the calculation of the node K-shell value in step 6.2 can be divided into the following steps:

[0058] Step 6.2.1 First remove the nodes with degree k=1 in the network. If there are still nodes with degree k=1 in the remaining network, continue to remove these nodes.

[0059] Step 6.2.2: The K-shell values ​​of all deleted nodes with degree k = 1 are recorded as ks = 1.

[0060] In step 6.2.3, let k = 2, 3, 4, ... in sequence, thus completing the K-shell decomposition of the network.

[0061] Step 6.3 The KSN (the K-shell and Neighborhood Centrality) value of node v is defined as follows:

[0062]

[0063] The closer the node is to the network core, the greater the KSN value of the node, and the higher its importance. The specific process of the KSN algorithm is as follows: Figure 4 shown.

[0064] Furthermore, in step 7 above, the visualization process of the node importance includes the following steps:

[0065] Step 7.1 normalizes the KSN value of the node, which is defined as follows:

[0066]

[0067] Where KSN(v) represents the KSN value of the current node before calculation, v represents the current node, V represents the node set, and K(v) represents the KSN value of the node after normalization.

[0068] Step 7.2 converts the normalized value into the color depth of the specific model color, which is defined as follows:

[0069] C(v)=K(v)*α*m (10)

[0070] Where α is a hyperparameter used to uniformly map values ​​to RGB, m is the color representation range in the specific model, and C(v) represents the color depth of each node.

[0071] Step 7.3 adds the potential node relationships generated in step 5 to the adjacency matrix created in step 3.3 to form a new blockchain network. Since the number of added relationships is limited, it does not affect the processing of the network.

[0072] Step 7.4 sets the edge weight in the blockchain to 1, the id corresponds to each address account, and the edge type is a directed edge.

[0073] Step 7.5 formats the data generated in step 7.4 into a data file format that Gephi can accept. Source represents the starting node number of the edge. Target represents the target node number of the edge. Type represents whether it is a directed or undirected graph. Weight represents the weight of the edge.

[0074] Step 7.6 imports the formatted data into Gephi to generate a visual relationship diagram.

[0075] Step 7.7: Input the node color depth generated in step 6.2 into Gephi to obtain the color representation of the node. This allows visualization of node importance and transaction tracking.

[0076] The above method can be built into an electronic device, which is a control device or a storage device. The control device is equipped with the above-mentioned blockchain anonymous tracking method based on the knowledge graph, and the storage device stores a program code of a blockchain anonymous tracking method based on the knowledge graph.

[0077] Beneficial effects of the present invention:

[0078] 1. Save resource overhead caused by frequent calculation of node feature vectors due to the increasing amount of blockchain data

[0079] 2. The relationship between suspicious target nodes can be inferred based on the different relationships between blockchain nodes

[0080] 3. Visualize transaction tracking through blockchain node relationships and influence levels

[0081] 4. A general solution to solve the anonymity and untraceability of blockchain is proposed

[0082] 5. Using the topological structure of the node to mine the transaction characteristics of the node is universal. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 Blockchain anonymous tracking flow chart

[0084] Figure 2Graph SAGE operation flow chart

[0085] Figure 3 Relationship Assessment Flowchart

[0086] Figure 4 KSN algorithm flow chart DETAILED DESCRIPTION

[0087] The present invention will be further described below with reference to the accompanying drawings.

[0088] like Figure 1 As shown, the present invention is further explained below. It should be noted that the specific implementation of this example is based on this technology, and a detailed implementation process and implementation steps are given, but the protection scope of the present invention is not limited to this implementation example.

[0089] Assuming that the time T of the malicious event is known, the specific implementation method of the present invention is as follows:

[0090] (1) Step 1: Use the blockchain parsing tool to parse the blockchain slice file to obtain the corresponding transaction information, such as timestamp, address account name, amount, etc.

[0091] (2) According to the time and amount of the suspicious event, the blockchain slice file corresponding to the time of the suspicious event and the amount involved is filtered to obtain the corresponding address account range. The hidden rule between address accounts is defined as R x , the rule set in the dependency syntax tree rule base is defined as R = {R1,…,R n}, by matching the rules with the rules in the rule base, the corresponding logical relationship between address accounts is obtained.

[0092] (3) Based on the logical relationship obtained in step (2) and the address account range involved, the node (address account) is embedded in the learning process.

[0093] I(v)=GraphSAGE(v) (11)

[0094] Compute the feature representation I(v) of node v.

[0095] (4), first define the set of all transactions as T, T = {t1, t2, t3, ..., t m}, m represents the total number of transactions. For each transaction, it is defined as a five-tuple t=(I, O, V I ,V O ,F), I represents the input address account set, I={i1,i2,i3,…,i n}, n is the number of input address accounts; O={o1,o2,o3,…,o k}, where O represents the output address account set, k is the number of output address accounts; V I Represents the total input value of all address accounts, V O It represents the total output value of all address accounts, and F represents the transaction fee.

[0096] Entities are then identified by transaction type:

[0097] This paper divides digital currency transaction types into three categories: Coinbase coin production transactions, multi-input and output transactions, and change transactions.

[0098] (5) Use machine learning to train the relationship between feature representations and use 30% of the training set for verification. Randomly extract samples from the positive samples and replace s or o with other entities. By optimizing the cross-entropy loss function, the model scores the positive samples higher than the negative samples, thereby achieving the purpose of training the model. The established relationship detection model is used to detect suspicious addresses to determine whether there is a potential transaction relationship between addresses. The relationship between the original data set, verification set, and training set is shown in the formula

[0099]

[0100] Where D represents the original dataset, S represents the training set, and T represents the test set.

[0101] (6) According to step 6.1, the blockchain network is considered as an undirected graph, and for each node, a set of neighbor nodes within the second order is constructed, denoted as τ(v), that is, the set of neighbor nodes that can be reached within 2 steps from node v. The K-shell value of node v is denoted as ks(v). The K-shell value explains the hierarchical characteristics of the network by recursively removing all nodes with degree 1 from the edge of the network. Nodes at the core have higher influence.

[0102] Step 6.2:

[0103] a) Remove nodes with degree 1 from the network edge. If there are still nodes with degree 1 in the remaining network, continue the operation until there are no nodes with degree 1 on the network edge.

[0104] b). The K-shell value of all removed nodes with degree 1 is recorded as ks(v)=1.

[0105] c) By removing nodes with degrees 2, 3, 4, ..., the K-shell decomposition of the blockchain network is completed.

[0106] d). Calculate the KSN value of the node, as shown in the formula:

[0107]

[0108] The closer the node is to the network core, the larger the KSN value of the node is and the higher its importance is.

[0109] (7) Normalize the node KSN value generated in step 6, which is specifically defined as follows:

[0110]

[0111] Where KSN(v) represents the KSN value of the current node before calculation, v represents the current node, V represents the node set, and K(v) represents the KSN value of the node after normalization.

[0112] Convert the normalized value to the color depth of the specific model color, which is defined as follows:

[0113] C(v)=K(v)*α*m (10)

[0114] Where α is a hyperparameter used to evenly map values ​​to RGB, m is the range of colors in the specific model, and C(v) represents the color depth of each node. The potential node relationships generated in step 4 are then added to the adjacency matrix created in step 3.3 to form a new blockchain network. Since the number of added relationships is limited, it does not affect the processing of the network. Since the blockchain network constructed by the present invention does not consider the weight of the edge, the edge weight in the blockchain is set to 1, the id corresponds to each address account, the edge type is a directed edge, and the data generated in step 7.4 is formatted into a data file format received by Gephi, where source represents the starting node number of the edge. target represents the target node number of the edge. type indicates whether it is a directed graph or an undirected graph. weight is the weight of the edge. After processing the data, the formatted data is imported into Gephi to generate a visual relationship diagram. Finally, the node color depth generated in step 6.2 is input into Gephi to obtain the color representation of the node. This achieves visualization of the importance of the node and transaction tracking.

[0115] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent methods or changes that do not deviate from the technology of the present invention should be included in the scope of protection of the present invention.

Claims

1. A blockchain anonymous tracking method based on knowledge graph, characterized in that: These include: S1. Parse the blockchain slice file to obtain the corresponding transaction information, such as amount, address account, and timestamp; S2. According to the time and amount of the suspicious event, the blockchain slice file corresponding to the time of the suspicious event and the amount involved is screened to obtain the corresponding address account range, and the suspicious address or account is recorded as A = {a1,…,a n }, where a i Represents the i-th address or account; the transaction value of the address or account is recorded as m={M1,…,M n }, M i Indicates the start and end conditions of the slice α i The value of the transaction; define the hidden rule between address accounts as R x , the rule set in the dependency syntax tree rule base is defined as R = {R1,…,R n }, by adding rule R x Match the rules in the rule base to obtain the corresponding logical relationship; S3. Using the Graph SAGE algorithm to perform embedded learning based on the logical relationship and address account obtained in step S2 above; In S3, the Graph SAGE algorithm performs embedded learning as follows: S3.1 Read the corresponding data from the parsed blockchain database, such as timestamps, addresses, or account relationships; S3.2 converts the read data into a low-dimensional vector, where the vector dimension of each address is the same, and initializes the corresponding address or account to convert it into a low-dimensional vector of fixed dimension; S3.3 represents the addresses and accounts in the blockchain network as an adjacency matrix A according to the logical relationship, where each address and account corresponds to a unique ID, and each ID corresponds to the initial feature representation vector of the address and account; S3.4 uses the adjacency matrix A (representing the logical relationships between addresses and accounts) generated in S3.3, as well as the initial vectorized representation of each address and account (also known as node features), as input to the Graph SAGE algorithm. Here, each address and account is considered a node in the graph structure. Through continuous iterations of the Graph SAGE algorithm, a vector representation of each node is ultimately generated. S4. Perform heuristic entity recognition on the blockchain data parsed in step S1 to obtain the blockchain entity corresponding to each account address and identify them in sequence by serial number; S5. Use machine learning to train the relationships between nodes to analyze and infer the potential relationships between entities in suspicious addresses; S6. Use the KSN algorithm to evaluate the importance of nodes with more comprehensive network topology characteristics.

2. The blockchain anonymous tracking method based on knowledge graph according to claim 1 is characterized in that: In S1, blockchain analysis includes the following: S1.1 Download transaction data through the blockchain wallet; S1.2 performs structured parsing of the block data, extracts the key fields of the parsed Bitcoin data and stores them in the Key-Value database.

3. The blockchain anonymous tracking method based on knowledge graph according to claim 1 is characterized in that: In S3.4, the execution of the Graph SAGE algorithm includes: Step 3.4.1 First, uniformly sample a fixed-size neighborhood set for each node, and draw different uniform samples at each iteration k; Step 3.4.2 Next, each node In its direct vicinity The representation of the aggregate node is converted into a vector The vector representation of is defined as the input node feature, expressed as: in represents the feature representation of the aggregation field, u represents the neighbors of node v, N(v) represents the node neighborhood composed of neighbors, Represents the current neighbor node feature, and generates the current feature representation of the neighborhood after the aggregation function Step 3.4.3 After aggregating adjacent feature vectors, Graph SAGE converts the current representation of the node into and aggregated neighborhood vectors This connection vector is fed through the fully connected layer of the nonlinear activation function σ and converted into the parameters used in the next step, expressed as Expressed as: in Represents the input feature representation in the next iteration, W k represents the weight matrix of the kth layer, σ represents the activation function, which is generally Sigmoid; Step 3.4.4 After k iterations of Graph SAGE, normalize the feature representation Finally, the feature representation of each node is generated 4. The blockchain anonymous tracking method based on knowledge graph according to claim 1 is characterized in that: In S4, the process of identifying the blockchain entity includes the following: Step 4.1 performs heuristic entity recognition on all blockchain address accounts according to transaction type, as follows: Step 4.1.1 defines the set of all transactions as T, T = {t1, t2, t3, ..., t m }, m represents the total number of transactions; Step 4.1.2 defines each transaction as a five-tuple t=(I, O, V I ,V O ,F), I represents the input address account set, I={i1,i2,i3,…,i n }, n is the number of input address accounts; O={o1,o2,o3,…,o k }, where O represents the output address account set, k is the number of output address accounts; V I Represents the total input value of all address accounts, V O Represents the total output value of all address accounts, F represents the transaction fee, where V I , V O , the calculation of F set is shown in formulas 3, 4, and 5: in Represents a single input address account i s The input value, Represents a single input address account o s Output value; Step 4.2 identifies entities according to different transaction types of the blockchain, thereby clustering different address accounts into different entities.

5. The blockchain anonymous tracking method based on knowledge graph according to claim 1 is characterized in that: In S5, the evaluation of the node vector representation relationship includes the following steps: S5.1 Each node is represented in a one-to-one correspondence with its features, and the corresponding relationships and entities are extracted from the network, and a 30% test set is retained; S5.2 uses DisMult as the scoring function, and the function definition is as follows: where e s and e o It is the node feature representation learned using Graph SAGE, R r is the diagonal matrix corresponding to the relation r; S5.3 uses negative sampling to train the machine learning model. For each positive sample, ω negative samples are sampled, that is, s or o is replaced with other entities. By optimizing the cross-entropy loss function, the model gives a higher score to the positive samples than to the negative samples. The specific definition is as follows: Where T is a set of positive and negative samples, l is a sigmoid function, y = 1 represents a positive sample, and y = 0 represents a negative sample; Step 5.4 uses the trained model to detect suspicious addresses to determine whether there is a potential relationship.

6. The blockchain anonymous tracking method based on knowledge graph according to claim 1 is characterized in that: In S6, the evaluation of the importance of the node includes the following steps: S6.1 treats the blockchain network as an undirected graph and generates a set of second-order neighbor nodes for each node, that is, the set of neighbor nodes that can be reached within 2 steps from node v. This set is denoted as τ(v). S6.2 The K-shell value of node v is denoted as ks(v). The K-shell value explains the hierarchical characteristics of the network by recursively removing all nodes with degree k from the edge nodes of the network. The nodes at the core have higher influence. S6.3 The KSN (the K-shell and Neighborhood Centrality) value of node v is defined as follows: The closer the node is to the network core, the larger the KSN value of the node is, and the higher its importance is.

7. The blockchain anonymous tracking method based on knowledge graph according to claim 6 is characterized in that: The S6.2 node K-shell value is calculated as follows: S6.2.1 First remove nodes with degree k = 1 from the network. If there are still nodes with degree k = 1 in the remaining network, continue to remove these nodes. S6.2.2Record the K-shell value of all deleted nodes with degree k = 1 as ks = 1; S6.2.3 Set k = 2, 3, 4, ... in sequence to complete the K-shell decomposition of the network.

8. A blockchain anonymous tracking method based on knowledge graph according to any one of claims 1 to 7, characterized in that: It also includes S7, using the Gephi algorithm to color the importance of nodes to different degrees to achieve visualization of transaction tracking; the details are as follows: Step 7.1 normalizes the KSN value of the node, which is defined as follows: Where KSN(v) represents the KSN value of the current node before calculation, v represents the current node, V represents the node set, and K(v) represents the KSN value of the node after normalization; Step 7.2 converts the normalized value into the color depth of the specific model color, which is defined as follows: C(v)=K(v)*α*m(10) where α is a hyperparameter used to uniformly map values ​​to RGB, m is the color representation range in the specific model, and C(v) represents the color depth of each node; Step 7.3: Add the generated potential node relationships to the adjacency matrix to form a new blockchain network. Step 7.4 sets the edge weight in the blockchain to 1, the id corresponds to each address account, and the edge type is a directed edge; Step 7.5 formats the data generated in step 7.4 into the data file format accepted by Gephi, where source represents the starting node number of the edge, target represents the target node number of the edge, type represents whether it is a directed graph or an undirected graph, and weight is the weight of the edge; Step 7.6 Import the formatted data into Gephi to generate a visual relationship diagram; Step 7.7 Input the node color depth generated in step 6.2 into Gephi to obtain the color representation of the node, thereby realizing the visualization of the importance of the node and transaction tracking.

9. An electronic device, characterized in that: The electronic device is a control device or a storage device, and the control device is provided with a blockchain anonymous tracking method based on a knowledge graph as described in any one of claims 1-8, and the storage device stores the program code of the blockchain anonymous tracking method based on a knowledge graph as described in any one of claims 1-8.

Images