A method and system for extracting features of a bitcoin transaction address

Abstract

The application provides a bitcoin transaction address feature extraction method and system, and relates to the technical field of blockchains, comprising: collecting bitcoin original transaction data, designing and extracting transaction behavior features of the original transaction data; performing feature weighting and feature screening on the transaction behavior features of the original transaction data to obtain a screened and weighted transaction behavior feature set; extracting transaction relationship features of the original transaction data by using a network representation learning algorithm to generate a transaction relationship feature set; and performing feature fusion on the screened and weighted transaction behavior feature set and the transaction relationship feature set to obtain a comprehensive feature set. The application constructs a bitcoin transaction behavior feature extraction method, extracts transaction relationship features by using a network representation learning algorithm, combines transaction behavior features and transaction relationship features to obtain comprehensive features, and the comprehensive features play an important role in the identification of the identities of entities behind transaction addresses, and can effectively improve the accuracy of entity identity identification.

Description

Technical Field

[0001] This invention belongs to the field of blockchain technology, and in particular relates to a method and system for extracting Bitcoin transaction address features. Background Technology

[0002] Researchers and regulators have been working to develop technologies and methods to deanonymize Bitcoin transactions, increasing their traceability and compliance. Therefore, designing methods for extracting Bitcoin transaction features for machine learning classifiers to identify the entities behind transaction addresses is crucial for effective regulatory oversight of Bitcoin transactions and for the standardized management of the trading market.

[0003] Existing methods for extracting Bitcoin transaction features include the following:

[0004] (1) Transaction graph features: The Bitcoin transaction network is represented as a graph, where addresses are nodes and transactions are edges. Topological features of the graph are extracted, such as the degree of nodes and clustering coefficients, to capture the relationships between addresses.

[0005] (2) Transaction behavior characteristics: Based on transaction data, extract address-related behavioral characteristics, such as transaction frequency, transaction amount, and transaction time interval.

[0006] (3) Transaction pattern characteristics: Identify common transaction patterns, such as transfer patterns from a specific address to another address, or frequent transaction patterns between multiple addresses.

[0007] (4) Account clustering features: Using clustering algorithms to group similar addresses and using clustering information as features helps to identify addresses under centralized control.

[0008] Existing methods for extracting Bitcoin transaction address features still have some limitations. These include:

[0009] (1) High-dimensionality problem: Bitcoin transaction data is usually high-dimensional, involving a large number of features. High-dimensional data may lead to the curse of dimensionality, increasing the complexity of modeling and analysis.

[0010] (2) Feature correlation problem: Some features may be highly correlated, leading to redundant information. In the feature selection process, it is necessary to identify and remove redundant features to improve the efficiency and accuracy of the model.

[0011] (3) Problem of single feature: For example, the transaction graph feature extraction method only extracts graph-related features and does not involve the transaction behavior features of addresses. The transaction behavior feature extraction method does not involve the transaction relationship features between addresses, which leads to the singleness of features and thus affects the accuracy of recognition.

[0012] Therefore, how to solve the problems of high dimensionality, feature correlation, and feature singularity in feature extraction, and improve the accuracy of feature extraction, has become an urgent problem to be solved. Summary of the Invention

[0013] To address the aforementioned problems, this invention proposes a method and system for extracting Bitcoin transaction address features. By incorporating feature selection, the feature dimensionality and noise can be reduced, and highly relevant features can be selected to remove redundant parts, thereby solving the problems of high dimensionality and feature correlation. Furthermore, by combining transaction graph feature extraction and transaction behavior feature extraction, the extracted comprehensive features not only include the transaction behavior features of the addresses themselves but also the transaction relationship information between addresses, thus solving the problem of feature singularity and improving the accuracy of feature extraction.

[0014] To achieve the above objectives, this invention discloses a method for extracting Bitcoin transaction address features, comprising:

[0015] Collect raw Bitcoin transaction data, and design and extract transaction behavior characteristics from the raw transaction data;

[0016] The transaction behavior characteristics of the original transaction data are weighted and filtered to obtain a set of filtered and weighted transaction behavior characteristics.

[0017] The network representation learning algorithm is used to extract the transaction relationship features from the original transaction data to generate a transaction relationship feature set;

[0018] The filtered and weighted set of transaction behavior features and the set of transaction relationship features are fused to obtain a comprehensive feature set.

[0019] As a further improvement of the present invention, data preprocessing is performed on the collected raw Bitcoin data, including:

[0020] The required fields are stored in a data table, which is divided into an address table, a transaction relationship table, and a transaction table. The transaction relationship table stores the input and output addresses of each transaction, and the transaction table stores the basic field information of a single transaction.

[0021] As a further improvement of the present invention, the transaction behavior characteristics of the original transaction data are designed, including:

[0022] Based on the transaction relationship table, design the address connectivity pattern characteristics, and based on the transaction table, design the address transaction amount characteristics and transaction time characteristics;

[0023] All designed transaction behavior features are combined into a transaction behavior feature set F = {f1, f2, ..., f...} n}, where n is the number of transaction behavior features.

[0024] As a further improvement of the present invention, feature weighting is applied to the transaction behavior characteristics of the original transaction data, including:

[0025] Calculate the importance scores for all transaction behavior features to obtain the importance score set I = {i1, i2, ..., i} n};

[0026] The transaction behavior feature set is subjected to feature selection using k methods to obtain k feature subsets;

[0027] Count the number of times each transaction behavior feature appears in the k feature subsets of the transaction behavior feature set, and increment the count by 1 to form a set C = {c1, c2, ..., c...} n}, where c i That is, feature f i Increment the number of occurrences by 1;

[0028] The importance score set I and set C are multiplied element by element to obtain the weight set W of the transaction behavior features;

[0029] The weights of each transaction behavior feature are standardized to obtain a standardized weight set W′, which is then multiplied element-wise with the transaction behavior feature set to obtain a weighted feature set.

[0030] As a further improvement of the present invention, feature filtering is performed on the transaction behavior characteristics of the original transaction data, including:

[0031] Based on the importance score of each transaction behavior feature, the features are sorted in descending order, and the top-ranked features are retained according to a preset screening percentage.

[0032] For transaction behavior features that rank lower, determine whether to retain or delete the transaction behavior feature in the weighted feature set based on the number of times the transaction behavior feature appears in the k feature subsets.

[0033] The filtered and weighted transaction behavior feature set Where p is the number of features after filtering.

[0034] As a further improvement of the present invention, a network representation learning algorithm is used to extract transaction relationship features from the original transaction data to generate a transaction relationship feature set; including:

[0035] Based on the original transaction data, an undirected graph G = {V, E} is constructed, with transaction addresses as nodes and transaction relationships between addresses as edges. The node set is V and the edge set is E.

[0036] Multiple random swimwalk samples are performed on the undirected graph of the transaction network to obtain a series of path sequences containing all nodes in the undirected graph of the transaction network.

[0037] A series of path sequences are embedded into a low-dimensional vector space, and the vector representation of each node in the low-dimensional vector space is the transaction relationship feature of the transaction address.

[0038] The transaction relationship features of all nodes are used to construct a transaction relationship feature set.

[0039] As a further improvement of the present invention, the filtered and weighted transaction behavior feature set and the transaction relationship feature set are fused to obtain a comprehensive feature set; the formula is:

[0040]

[0041] in,

[0042] F S This represents the filtered and weighted set of transaction behavior features;

[0043] F D Represents the feature set of transaction relationships;

[0044] P represents F S The number of features in F, where q represents F D The number of features in;

[0045] r = p + q, where r represents the comprehensive feature set F. T The number of features in the data.

[0046] As a further improvement of the present invention, the invention also includes performing feature standardization processing on the comprehensive feature set to obtain a standardized comprehensive feature set.

[0047] As a further improvement of this invention, the Z-score standardization method is adopted to perform Z-score standardization on each feature x in the comprehensive feature set, as shown in the formula:

[0048]

[0049] in,

[0050] x′ represents the standardized value of feature x;

[0051] σ represents the mean of all features in the comprehensive feature set, and σ represents the standard deviation of all features in the comprehensive feature set.

[0052] Finally, the standardized comprehensive feature set is obtained.

[0053] The present invention also provides a system for a method of extracting Bitcoin transaction address features, comprising: a transaction behavior feature design and extraction module, a feature weighting and filtering module, a transaction relationship feature extraction module, a feature fusion module, and a feature standardization module;

[0054] The transaction behavior feature design extraction module is used for:

[0055] Collect raw Bitcoin transaction data, and design and extract transaction behavior characteristics from the raw transaction data;

[0056] The feature weighting and filtering module is used for:

[0057] The transaction behavior characteristics of the original transaction data are weighted and filtered to obtain a set of filtered and weighted transaction behavior characteristics.

[0058] The transaction relationship feature extraction module is used for:

[0059] The network representation learning algorithm is used to extract the transaction relationship features from the original transaction data to generate a transaction relationship feature set;

[0060] The feature fusion module is used for:

[0061] The filtered and weighted set of transaction behavior features and the set of transaction relationship features are fused to obtain a comprehensive feature set.

[0062] The feature standardization module is used for:

[0063] The comprehensive feature set is subjected to feature standardization processing to obtain a standardized comprehensive feature set.

[0064] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0065] This invention constructs a method for extracting Bitcoin transaction behavior features through feature weighting and feature filtering, and extracts transaction relationship features through network representation learning algorithm. The transaction behavior features and transaction relationship features are combined to obtain comprehensive features. These comprehensive features play an important role in identifying the entity behind the transaction address and can effectively improve the accuracy of entity identification.

[0066] This invention employs k methods to select features from the transaction behavior feature set, and utilizes the feature selection results in the feature weighting and feature filtering stages. By combining multiple feature selection methods, it reduces information noise and dimensionality. Furthermore, the combination of multiple methods can reduce information loss, increase the reliability of feature selection, and reduce randomness. At the same time, by applying the results of multiple feature selections and assigning feature weights based on feature importance scores, the influence of important features is emphasized, which can further improve the accuracy of subsequent identification results. Attached Figure Description

[0067] Figure 1 This is an overall flowchart of a Bitcoin transaction address feature extraction method disclosed in one embodiment of the present invention;

[0068] Figure 2 This is a flowchart of transaction behavior feature extraction according to one embodiment of the present invention;

[0069] Figure 3 This is a flowchart of feature selection and feature weighting according to an embodiment of the present invention;

[0070] Figure 4 This is a flowchart of network representation learning feature extraction according to an embodiment of the present invention;

[0071] Figure 5 This is a feature fusion flowchart disclosed in one embodiment of the present invention;

[0072] Figure 6 This is a schematic diagram of a Bitcoin transaction address feature extraction system disclosed in one embodiment of the present invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] The present invention will now be described in further detail with reference to the accompanying drawings:

[0075] like Figure 1 As shown, this invention discloses a method for extracting features from Bitcoin transaction addresses. Taking Bitcoin as the research object, it collects raw Bitcoin transaction data, extracts transaction features related to Bitcoin addresses, and performs operations such as filtering and weighting on these features to construct a comprehensive feature set. This set can be used by machine learning algorithms to identify the entity behind the transaction address. The process is roughly divided into four parts: manual feature design of transaction behavior, feature selection and weighting, automatic feature extraction through network representation learning, and overall feature fusion. Specifically, it includes:

[0076] S1, such as Figure 2 As shown, we collect raw Bitcoin transaction data, design and extract transaction behavior characteristics from the raw transaction data;

[0077] in,

[0078] The purpose of this section is to extract transaction behavior characteristics of the address itself. These characteristics include transaction amount, transaction frequency, etc. First, transaction information of the address is collected from the Bitcoin blockchain and preprocessed, including:

[0079] The required fields are stored in the data tables, which are divided into address tables, relationship tables, and transaction tables. The relationship table stores the input and output addresses of each transaction, and the transaction table stores the basic field information of a single transaction.

[0080] Furthermore,

[0081] We manually designed the transaction behavior characteristics from the table above, including:

[0082] Design address connectivity features based on the transaction relationship table, such as the number of addresses as transaction inputs, the number of addresses as transaction outputs, and the number of input and output addresses; design address transaction fund quantity features based on the transaction table, such as the average and standard deviation of transaction amounts; and design transaction time features, such as transaction time intervals and transaction frequency.

[0083] All designed transaction behavior features are combined into a transaction behavior feature set F = {f1, f2, ..., f...} n}, where n is the number of transaction behavior features.

[0084] S2, such as Figure 3 As shown, the transaction behavior characteristics of the original transaction data are weighted and filtered to obtain a set of filtered and weighted transaction behavior characteristics.

[0085] in,

[0086] Assigning feature weights to the transaction behavior characteristics of the raw transaction data, including:

[0087] Calculate the importance scores for all transaction behavior features to obtain the importance score set I = {i1, i2, ..., i...} n};

[0088] Feature selection is performed on the transaction behavior feature set F using k methods (such as variance thresholding, information gain, etc.) to obtain k feature subsets F. SUB 1. F SUB 2、…、F SUB k;

[0089] The number of times each transaction behavior feature appears in the set of k feature subsets in the set of transaction behavior features F is counted, and the counts are incremented by 1 to form a set C = {c1, c2, ..., c...} n}, where c i That is, feature f i Increment the number of occurrences by 1;

[0090] By performing element-wise multiplication of the importance score set I and set C, we obtain the weight set W of the transaction behavior features, as shown in the formula:

[0091] W=I⊙C={i1*c1,i2*c2,...,i n *c n} = {w1, w2, ..., w n};

[0092] The weights of each transaction behavior feature are standardized to obtain a standardized weight set W′={w1′,w2′,...,w n Then, multiply the feature set F by the transaction behavior feature set element by element to obtain the weighted feature set, as shown in the formula:

[0093]

[0094] Furthermore,

[0095] Feature filtering is performed on the transaction behavior characteristics of the raw transaction data, including:

[0096] Based on the importance score I of each transaction behavior feature, the transaction behavior features are sorted in descending order, and the top h% of transaction behavior features are retained according to the preset screening percentage (e.g., h%).

[0097] For transaction behavior features ranked in the bottom 1% (where 1% = 1 - h%), the decision to retain or delete the feature from the weighted feature set is based on the frequency of its occurrence in the k feature subsets; specifically, if its frequency in the feature subsets is... (where 1≤u≤k), then from F W Delete this feature; otherwise, retain the feature.

[0098] The filtered and weighted transaction behavior feature set Where p is the number of features after filtering.

[0099] Specifically,

[0100] Calculating the importance scores of all transaction behavior features can be accomplished using tree-based methods, such as decision trees or random forests. After training the features, the importance of each feature can be obtained, and the corresponding importance score can be denoted as I = {i1, i2, ..., i...}. n}

[0101] S3, such as Figure 4As shown, a network representation learning algorithm is used to extract transaction relationship features from the original transaction data and generate a transaction relationship feature set.

[0102] in,

[0103] The relation_table is used to build an undirected graph G = {V, E} based on the original transaction data, with transaction addresses as nodes and the transaction relationships between addresses as edges. The node set is V and the edge set is E. If there is a transaction between address nodes v1 and v2, then there is an edge connecting nodes v1 and v2.

[0104] Multiple random swimwalks are performed in the undirected graph of the transaction network. A random swimwalk is a process of starting from a node and randomly selecting the next neighbor node in the graph. A series of path sequences are obtained through multiple random swimwalks. Each path sequence starts with a starting node and ends with a certain length. The series of path sequences contains all the nodes in the undirected graph of the transaction network.

[0105] Embedding a series of path sequences into a low-dimensional vector space, the vector representation of each node in the low-dimensional vector space is the transaction relationship feature of the transaction address;

[0106] The transaction relationship features of all nodes are used to construct a transaction relationship feature set.

[0107] Specifically,

[0108] The DeepWalk algorithm is used to embed path sequences into a low-dimensional vector space. The dimension of the low-dimensional vector is set to q, and the vector representation of each node is generated by learning the contextual information of the nodes in the path sequence. Finally, each Bitcoin address in the graph is represented by a q-dimensional vector, representing the transaction relationship characteristics of the address, denoted as q. Since DeepWalk has already mapped nodes to low-dimensional vectors, these vectors themselves already have good feature representation capabilities, and under normal circumstances, there is no need to perform traditional feature selection and feature weighting.

[0109] S4, such as Figure 5 As shown, the filtered and weighted transaction behavior feature set and transaction relationship feature set are fused to obtain a comprehensive feature set;

[0110] Among them, the transaction behavior feature F after feature weighting and feature filtering S Transaction relationship characteristic F D They represent features in different dimensions, F S F reflects the transaction characteristics of the address itself. D This reflects the transaction relationships between addresses. To obtain more comprehensive features, the two sets are merged to obtain a combined feature set, as shown in the formula:

[0111]

[0112] In the formula,

[0113] F S This represents the filtered and weighted set of transaction behavior features;

[0114] F D Represents the feature set of transaction relationships;

[0115] P represents F S The number of features in F, where q represents F D The number of features in;

[0116] r = p + q, where r represents the comprehensive feature set F. T The number of features in the data.

[0117] S5. Perform feature standardization on the comprehensive feature set to obtain the standardized comprehensive feature set.

[0118] in,

[0119] The Z-score standardization method is used to standardize each feature x in the comprehensive feature set using the Z-score method. The formula is as follows:

[0120]

[0121] in,

[0122] x′ represents the standardized value of feature x;

[0123] σ represents the mean of all features in the comprehensive feature set, and σ represents the standard deviation of all features in the comprehensive feature set.

[0124] Finally, the standardized comprehensive feature set is obtained.

[0125] Furthermore,

[0126] F can be used subsequently ST Used to train supervised machine learning algorithms to identify the entities behind Bitcoin addresses, it can effectively improve the accuracy of entity identification.

[0127] like Figure 6 As shown, the present invention also provides a system for Bitcoin transaction address feature extraction, comprising: a transaction behavior feature design and extraction module, a feature weighting and filtering module, a transaction relationship feature extraction module, a feature fusion module, and a feature standardization module;

[0128] The transaction behavior feature design extraction module is used for:

[0129] Collect raw Bitcoin transaction data and design and extract transaction behavior characteristics from the raw transaction data;

[0130] The feature weighting and filtering module is used for:

[0131] The transaction behavior characteristics of the original transaction data are weighted and filtered to obtain a set of filtered and weighted transaction behavior characteristics.

[0132] The transaction relationship feature extraction module is used for:

[0133] A network representation learning algorithm is used to extract transaction relationship features from the original transaction data and generate a transaction relationship feature set.

[0134] The feature fusion module is used for:

[0135] The filtered and weighted set of transaction behavior features and the set of transaction relationship features are fused to obtain a comprehensive feature set.

[0136] The feature normalization module is used for:

[0137] The comprehensive feature set is subjected to feature standardization processing to obtain the standardized comprehensive feature set.

[0138] Advantages of this invention:

[0139] This invention constructs a method for extracting Bitcoin transaction behavior features through feature weighting and feature filtering, and extracts transaction relationship features through network representation learning algorithm. The transaction behavior features and transaction relationship features are combined to obtain comprehensive features. These comprehensive features play an important role in identifying the entity behind the transaction address and can effectively improve the accuracy of entity identification.

[0140] This invention employs k methods to select features from the transaction behavior feature set, and utilizes the feature selection results in the feature weighting and feature filtering stages. This combines multiple feature selection methods, reducing noise and dimensionality of information. Furthermore, the combination of multiple methods reduces information loss, increases the reliability of feature selection, and reduces randomness. Simultaneously, applying the results of multiple feature selections and assigning feature weights based on feature importance scores emphasizes the influence of important features, further improving the accuracy of subsequent identification results.

[0141] This invention aims to extract features from Bitcoin transaction addresses, which can then be used in machine learning classification algorithms to identify the types of entities behind Bitcoin addresses. This helps monitor and identify addresses that may be involved in illegal activities, while also assisting regulatory agencies in better understanding and monitoring Bitcoin transaction activities. This helps ensure compliance requirements are met by digital asset trading platforms, exchanges, and financial institutions, and provides more effective regulatory means.

[0142] This invention incorporates a transaction network graph embedding method to consider address connectivity, enabling the capture of contextual relationships between Bitcoin addresses. This results in the trained embedding features exhibiting relationships and connectivity between Bitcoin addresses.

[0143] This invention reduces the dimensionality of the feature space by transforming high-dimensional Bitcoin address transaction relationship features into low-dimensional embedding vectors. This low-dimensional representation helps reduce the dimensionality of the feature space, removes redundant information, and better reveals the similarities and differences between Bitcoin addresses.

[0144] In terms of comprehensive feature representation, this invention combines transaction behavior features with graph embedding vectors, integrating the information from both. This comprehensive feature representation not only reveals the transaction relationships between Bitcoin addresses but also captures the transaction behavior characteristics of the addresses themselves. It provides a richer, more informative, and more comprehensive feature representation of Bitcoin addresses, further enhancing the accurate identification and analysis of Bitcoin addresses.

[0145] This invention combines multiple feature selection methods, applying a specific approach for comprehensive feature selection. This allows for the selection of more representative and important features from the feature set, reducing feature dimensionality and noise. Furthermore, combining multiple methods minimizes the possibility of information loss. Subsequent use in a classifier can improve model performance and generalization ability.

[0146] This invention designs feature weights by integrating feature selection methods. By assigning weights to different features, the influence of important features can be emphasized during model training, thereby further improving model performance.

[0147] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting Bitcoin transaction address features, characterized in that, include: Collect raw Bitcoin transaction data, design and extract transaction behavior features from the raw transaction data, and construct a transaction behavior feature set from all designed transaction behavior features. , where n is the number of transaction behavior features; The transaction behavior characteristics of the original transaction data are weighted and filtered to obtain a set of filtered and weighted transaction behavior characteristics. The feature weighting includes calculating importance scores for all transaction behavior features to obtain an importance score set. The transaction behavior feature set is selected using k methods to obtain k feature subsets; the frequency of each transaction behavior feature in the transaction behavior feature set in the k feature subsets is counted, and the frequency is incremented by 1 to form a set. The importance score set I and set C are multiplied element-wise to obtain the weight set W of the transaction behavior features. The weights of each transaction behavior feature are then standardized to obtain the standardized weight set. Then, perform element-wise multiplication with the transaction behavior feature set to obtain a weighted feature set; The feature selection process involves sorting features in descending order of importance score and retaining the top-ranked features by a predetermined percentage. For features ranked lower, if the number of times the feature appears in the feature subset meets a certain threshold... Delete if necessary, otherwise retain, resulting in a filtered and weighted set of transaction behavior features. , where p is the number of features after filtering; The network representation learning algorithm is used to extract the transaction relationship features from the original transaction data to generate a transaction relationship feature set; The filtered and weighted set of transaction behavior features and the set of transaction relationship features are fused to obtain a comprehensive feature set. The comprehensive feature set is Z-score standardized to obtain a standardized comprehensive feature set, thus completing the extraction of Bitcoin transaction address features.

2. The Bitcoin transaction address feature extraction method according to claim 1, characterized in that: Data preprocessing is performed on the collected raw Bitcoin data, including: The required fields are stored in a data table, which is divided into an address table, a transaction relationship table, and a transaction table. The transaction relationship table stores the input and output addresses of each transaction, and the transaction table stores the basic field information of a single transaction.

3. The Bitcoin transaction address feature extraction method according to claim 2, characterized in that: The design of the transaction behavior characteristics of the original transaction data includes: Based on the transaction relationship table, design the address connectivity pattern characteristics, and based on the transaction table, design the address transaction amount characteristics and transaction time characteristics.

4. The Bitcoin transaction address feature extraction method according to claim 1, characterized in that: In assigning feature weights to the transaction behavior characteristics of the original transaction data, the set middle That is, characteristics The number of occurrences is incremented by 1.

5. The Bitcoin transaction address feature extraction method according to claim 1, characterized in that: The network representation learning algorithm is used to extract transaction relationship features from the original transaction data, generating a transaction relationship feature set; including: Based on the original transaction data, an undirected graph of the transaction network is constructed, with transaction addresses as nodes and the transaction relationships between addresses as edges. The node set is edge set is ; Multiple random walks are performed on the undirected graph of the transaction network to obtain a series of path sequences containing all nodes in the undirected graph of the transaction network. A series of path sequences are embedded into a low-dimensional vector space, and the vector representation of each node in the low-dimensional vector space is the transaction relationship feature of the transaction address. Construct a transaction relationship feature set from the transaction relationship features of all nodes. .

6. The Bitcoin transaction address feature extraction method according to claim 1, characterized in that: The filtered and weighted transaction behavior feature set and the transaction relationship feature set are fused to obtain a comprehensive feature set; the formula is: ； in, This represents the filtered and weighted set of transaction behavior features; Represents the feature set of transaction relationships; P represents The number of features in the data, q represents The number of features in; r represents the comprehensive feature set The number of features in.

7. The Bitcoin transaction address feature extraction method according to claim 1, characterized in that: The Z-score normalization method is used to normalize the features in the comprehensive feature set. Z-score standardization is performed one by one, using the following formula: ； in, Representation of features The standardized value; This represents the mean of all features in the comprehensive feature set. This represents the standard deviation of all features in the comprehensive feature set; Finally, the standardized comprehensive feature set is obtained. .

8. A system for implementing the Bitcoin transaction address feature extraction method according to any one of claims 1 to 7, characterized in that, include: The module includes a transaction behavior feature design and extraction module, a feature weighting and filtering module, a transaction relationship feature extraction module, a feature fusion module, and a feature standardization module. The transaction behavior feature design extraction module is used for: Collect raw Bitcoin transaction data, design and extract transaction behavior features from the raw transaction data, and construct a transaction behavior feature set from all designed transaction behavior features. , where n is the number of transaction behavior features; The feature weighting and filtering module is used for: The transaction behavior characteristics of the original transaction data are weighted and filtered to obtain a set of filtered and weighted transaction behavior characteristics. The feature weighting includes calculating importance scores for all transaction behavior features to obtain an importance score set. The transaction behavior feature set is selected using k methods to obtain k feature subsets; the frequency of each transaction behavior feature in the transaction behavior feature set in the k feature subsets is counted, and the frequency is incremented by 1 to form a set. The importance score set I and set C are multiplied element-wise to obtain the weight set W of the transaction behavior features. The weights of each transaction behavior feature are then standardized to obtain the standardized weight set. Then, perform element-wise multiplication with the transaction behavior feature set to obtain a weighted feature set; The feature selection process involves sorting features in descending order of importance score and retaining the top-ranked features by a predetermined percentage. For features ranked lower, if the number of times the feature appears in the feature subset meets a certain threshold... Delete if necessary, otherwise retain, resulting in a filtered and weighted set of transaction behavior features. , where p is the number of features after filtering; The transaction relationship feature extraction module is used for: The network representation learning algorithm is used to extract the transaction relationship features from the original transaction data to generate a transaction relationship feature set; The feature fusion module is used for: The filtered and weighted set of transaction behavior features and the set of transaction relationship features are fused to obtain a comprehensive feature set. The feature standardization module is used for: The comprehensive feature set is subjected to Z-score feature standardization to obtain the standardized comprehensive feature set, thus completing the extraction of Bitcoin transaction address features.

Images