Blockchain address attribute classification method and system based on integrated machine learning
By integrating machine learning methods, combining decision trees and graph neural networks, a full node of the UTXO blockchain is constructed to acquire and preprocess transaction data, and to build address and entity transaction features. This solves the problem of inaccurate identification in the deanonymization of virtual currency addresses, and achieves more accurate address classification and entity information acquisition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-16
- Publication Date
- 2026-03-17
AI Technical Summary
Existing machine learning methods for deanonymizing cryptocurrency addresses have shortcomings in feature construction and model selection, resulting in inaccurate identification of the actual owners behind cryptocurrency addresses and an inability to effectively capture transaction relationships between addresses.
This paper adopts an ensemble machine learning approach, combining decision tree classifiers and graph neural network subgraph classifiers. By constructing a UTXO-based blockchain full node, transaction data is acquired and preprocessed to construct address and entity transaction features. The XGBoost and GCN models are used for training, and the model results are combined through a weighted average strategy to achieve accurate classification of blockchain addresses.
It improves the accuracy and robustness of blockchain address classification, can more effectively identify addresses of the same entity, provides rich entity information, and solves the problems of information loss and sample imbalance in existing methods.
Smart Images
Figure CN116578904B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of blockchain application technology, and specifically to a blockchain address attribute classification method and system based on integrated machine learning. Background Technology
[0002] Blockchain is a decentralized public ledger that records the history of all blockchain transactions and the current cryptocurrency balances of all holders. Decentralized cryptocurrencies, as a preferred payment method for some, offer high levels of transaction anonymity and privacy protection, making them ideal for those who value personal privacy. However, due to the anonymity and privacy of cryptocurrency transactions, some cryptocurrency users can evade regulation. Therefore, it is crucial for regulatory agencies to identify the actual owners behind cryptocurrency addresses and to track and regulate them.
[0003] The main machine learning methods for deanonymizing cryptocurrency addresses include the following:
[0004] Graph Analysis Methods: Cryptocurrency transactions can be represented as a directed graph, where each cryptocurrency address is a node and each transaction is a directed edge. Graph analysis methods can identify transaction clusters by analyzing the topology of the transaction network, inferring relationships between addresses, and ultimately deducing the actual owners behind the addresses. Clustering Methods: Cryptocurrency addresses can be clustered into different groups, such as wallets, exchanges, and mixing services. Clustering methods can group similar addresses together by comparing features such as transaction patterns, transaction volume, and transaction time, thereby inferring the actual owners of these addresses. Classification Methods: Machine learning models can be trained on cryptocurrency transaction data to predict transaction relationships between addresses, thus inferring the actual owners behind the addresses. For example, random forests and neural networks can be used to predict the identity of address owners.
[0005] While current machine learning-based methods have achieved some success, these three main analytical approaches face the following challenges: Graph analysis methods: Current graph analysis methods analyze data based on detailed address transaction data to obtain transaction subgraphs, and then classify these subgraphs. Since UTXO-based cryptocurrency addresses can typically be clustered (the entities in the address cluster are generally controlled by a single user), obtaining address transaction subgraphs by address leads to missing transaction information for the entity corresponding to the target address (the address is just one address of a user entity, making it impossible to obtain the overall information of the user entity from the address alone, thus resulting in missing information); Clustering methods: Current clustering methods are unsupervised, requiring a pre-set number of clusters, which needs to be selected based on the actual data. However, this selection process is often subjective and prone to errors; Classification methods: Current classification methods use a joint vector of address features and user entity features as input data. While considering entity features, most current classification methods employ decision trees, random forests, and neural networks. These methods can fully learn the transaction behavior of the current target address, but they cannot capture information related to the transaction relationships between addresses. Summary of the Invention
[0006] To address the technical problems of existing machine learning methods for deanonymizing cryptocurrency addresses, this invention provides a blockchain address attribute classification method and system based on ensemble machine learning. This method utilizes an ensemble model combining decision tree classifiers and graph neural network subgraph classifiers to effectively and accurately identify the categories of UTXO-based blockchain addresses, thereby achieving deanonymization.
[0007] This invention is achieved through the following technical solution:
[0008] This solution provides a blockchain address attribute classification method based on ensemble machine learning, including:
[0009] Step 1: Set up a full blockchain node based on UTXO to obtain and preprocess detailed transaction data from the blockchain;
[0010] Step 2: Feature construction based on preprocessed detailed transaction data: Address analysis is performed on the preprocessed detailed transaction data to obtain address transaction features, entity clustering analysis is performed to obtain entity transaction features, and an entity transaction subgraph is constructed.
[0011] Step 3: Input the address transaction features, entity transaction features, and entity transaction subgraph into the XGBoost model and GCN model for training, and integrate the trained XGBoost model and GCN model to obtain the final classification model;
[0012] Step 4: Predict the classification of blockchain address attributes based on the final classification model.
[0013] The working principle of this solution is as follows: Existing machine learning methods for deanonymizing cryptocurrency addresses have technical problems. This invention provides a blockchain address attribute classification method and system based on ensemble machine learning. It is based on an ensemble model of decision tree classifier and graph neural network subgraph classifier, thereby establishing an effective and accurate identification of the category of blockchain addresses based on UTXO, achieving the purpose of deanonymization.
[0014] This solution fully considers the shortcomings of existing methods in feature construction and model selection. It cleverly utilizes features from three aspects: blockchain addresses, entities, and entity relationships. Based on two machine learning techniques, XGBoost and GCN, it provides a more accurate and reliable method for address classification. The UTXO-based blockchain address clustering technology aggregates addresses of the same entity, resulting in more accurate entity information and providing richer features for subsequent classification. In addition to utilizing individual entity features, it also adds relationship features between entities by constructing entity transaction subgraphs, further enhancing the richness of entity features. This solution leverages two excellent machine learning models, XGBoost and GCN, and cleverly combines their prediction results through a weighted averaging strategy, significantly improving the robustness and reliability of the overall framework.
[0015] Setting up a UTXO-based blockchain full node provides complete transaction information for subsequent work. The main steps for setting up a UTXO-based blockchain full node are as follows: First, install the core cryptocurrency client software according to the local environment; then start the client software and synchronize the blockchain transaction details.
[0016] A further optimized solution is that the preprocessing includes the following methods:
[0017] The transaction information of the transaction address is extracted from the detailed transaction data to form an address information table. The transaction information includes: balance, number of outgoing transactions and number of incoming transactions.
[0018] The UTXO-based blockchain address clustering method aggregates addresses of the same entity together, integrates the aggregated address information to obtain basic entity information to form an entity information table, which includes: the number of entity-associated addresses, the number of entity-outgoing transfers, and the number of entity-ingoing transfers.
[0019] A further optimization is that the preprocessing also includes:
[0020] Collect tags for blockchain addresses from open-source sources;
[0021] The open-source channels include blockchain explorers, exchange announcements, social media, community forums, and authoritative third-party data analysis websites;
[0022] The labels include annotations and descriptions of blockchain address i, as well as label categories (such as mining, exchange, fraud, etc.) obtained through manual aggregation. This address information will provide an important basis for subsequent classification.
[0023] A further optimized solution involves constructing the entity transaction subgraph using the following methods:
[0024] Based on labeled address data, the target entities corresponding to the addresses are identified, and transactions involving all entities are processed to form a second-order directed entity transaction subgraph. This subgraph is stored as an adjacency matrix and used as input to the subsequent Graph Neural Network (GCN). Specifically:
[0025] Obtain the tags of blockchain addresses and retrieve the entities to which the blockchain addresses belong;
[0026] Centered on entity E, retrieve the entity's transaction data, and construct a first-order directed transaction subgraph with the counterparty entities {E1, E2, ..., En} in the transaction data as first-order neighbor nodes;
[0027] Centered on first-order neighbor nodes, retrieve transaction data for each entity in the first-order directed transaction subgraph, and use the counterparty entities in the transaction data as second-order neighbor nodes of entity E to form a second-order directed subgraph of entity E; wherein the retrieval time is limited when retrieving the transaction data of entities.
[0028] Due to the large volume of transactions, the memory pressure required to construct subgraphs using the full transaction data would also increase. Therefore, the time limit for retrieving transaction data is set to a certain period of time. The m time periods of entity E are retrieved to form m subgraphs of entity E. This not only reduces memory pressure but also increases the richness of the subgraphs. At the same time, this method oversamples a small number of labeled samples, which solves the problem of sample imbalance to some extent.
[0029] A further optimized approach involves constructing entity transaction characteristics using the following methods:
[0030] Obtain the tag of the blockchain address, lock the target entity corresponding to the blockchain address, and select entity feature information from the entity information table of the target entity to construct the entity transaction feature vector;
[0031] The entity feature information includes quantity features, amount features, and time features; the quantity features include: number of sub-addresses, total number of transactions, number of inbound transactions, number of outbound transactions, percentage of inbound transactions, and percentage of outbound transactions; the amount features include: total transaction amount, total inbound amount, total outbound amount, net inbound amount, maximum inbound amount, maximum outbound amount, average inbound amount, and average outbound amount; the time features include: earliest transaction time, most recent transaction time, and average transaction frequency.
[0032] A further optimized solution involves constructing address transaction characteristics using the following methods:
[0033] Obtain the tag of the blockchain address, lock the target entity corresponding to the blockchain address, and then map multiple blockchain addresses according to the target entity: the blockchain addresses of the same target entity have the same entity characteristics and different address characteristics. The address characteristics are standardized to form an address transaction feature vector.
[0034] The address characteristics include quantity characteristics, amount characteristics, and time characteristics; the quantity characteristics include: total transaction volume, number of inbound transactions, number of outbound transactions, percentage of inbound transactions, percentage of outbound transactions, and number of transactions with a unique address; the amount characteristics include: total transaction amount, total amount of inbound transactions, total amount of outbound transactions, average inbound amount, average outbound amount, and total transaction fee; the time characteristics include: earliest transaction time, most recent transaction time, and average transaction frequency.
[0035] Further optimization involves training the XGBoost model using the following methods:
[0036] The address transaction characteristics corresponding to each blockchain address are concatenated with the entity transaction characteristics of the entity to which it belongs to form an address characteristic set;
[0037] The address feature set is divided into a training set and a test set. The hyperparameters of the XGBoost model are initialized, and the training set is used as input to train the XGBoost model. At the same time, the performance index of the XGBoost model is detected, thereby adjusting the model hyperparameters.
[0038] The performance of the XGBoost model was evaluated based on precision and recall metrics, and on a test set.
[0039] Further optimization involves training the GCN model using the following methods:
[0040] Construct a set of entity transaction subgraphs, where each entity transaction subgraph G = {A, E}, where A is the adjacency matrix, representing the connection relationship between entities, and E is the entity feature, representing the feature of each entity in the entity transaction subgraph.
[0041] The set of entity transaction subgraphs is divided into a training set and a test set;
[0042] Initialize the hyperparameters of the GCN model, use the training set as input to the GCN model, and simultaneously detect the performance metrics of the GCN model to adjust the model hyperparameters.
[0043] The performance of the XGBoost model was evaluated based on precision and recall metrics, and on a test set.
[0044] Further optimizations include the following model ensemble methods:
[0045] The results of the XGBoost and GCN models are combined using a weighted average strategy to obtain the final prediction result. The main steps are as follows:
[0046] Obtain the final performance metrics for the XGBoost and GCN models;
[0047] By combining the prediction results and final performance metrics of the XGBoost and GCN models, weights are assigned to the XGBoost and GCN models to obtain the final integrated classification model.
[0048] A small number of newly added samples are selected to test the final classification model and the weights are adjusted. These newly added samples are not in the training set or the prediction set.
[0049] This solution also provides a blockchain address attribute classification system based on ensemble machine learning, used to implement the aforementioned blockchain address attribute classification method based on ensemble machine learning, including:
[0050] The data acquisition module is used to build a full blockchain node based on UTXO, acquire detailed transaction data of the blockchain and perform preprocessing.
[0051] The preprocessing module is used to construct features based on the preprocessed detailed transaction data: address analysis is performed on the preprocessed detailed transaction data to obtain address transaction features, entity clustering analysis is performed to obtain entity transaction features, and an entity transaction subgraph is constructed.
[0052] The model training module is used to input address transaction features, entity transaction features, and entity transaction subgraphs into the XGBoost model and GCN model for training. The trained XGBoost model and GCN model are then integrated to obtain the final classification model.
[0053] The output module is used to predict the classification of blockchain address attributes based on the final classification model.
[0054] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0055] This invention provides a blockchain address attribute classification method and system based on ensemble machine learning. Attached Figure Description
[0056] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:
[0057] Figure 1 This is a flowchart illustrating a blockchain address attribute classification method based on ensemble machine learning.
[0058] Figure 2 This is a schematic diagram illustrating the principle of blockchain address attribute classification based on ensemble machine learning. Detailed Implementation
[0059] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.
[0060] Example 1
[0061] This embodiment provides a blockchain address attribute classification method based on ensemble machine learning, such as... Figure 1 and Figure 2 The following are included:
[0062] Step 1: Set up a full blockchain node based on UTXO to obtain detailed transaction data from the blockchain and perform preprocessing; preprocessing includes the following methods:
[0063] The transaction information of the transaction address is extracted from the detailed transaction data to form an address information table. The transaction information includes: balance, number of outgoing transactions and number of incoming transactions.
[0064] The UTXO-based blockchain address clustering method aggregates addresses of the same entity together, integrates the aggregated address information to obtain basic entity information to form an entity information table, which includes: the number of entity-associated addresses, the number of entity-outgoing transfers, and the number of entity-ingoing transfers.
[0065] Preprocessing also includes: collecting tags for blockchain addresses from open-source sources; open-source sources include blockchain explorers, exchange announcements, social media, community forums, and authoritative third-party data analysis websites; tags include annotations and descriptions of blockchain address i, as well as tag categories obtained through manual integration.
[0066] Step 2: Feature construction based on preprocessed detailed transaction data: Address analysis is performed on the preprocessed detailed transaction data to obtain address transaction features, entity clustering analysis is performed to obtain entity transaction features, and an entity transaction subgraph is constructed.
[0067] The methods for constructing entity transaction subgraphs include:
[0068] Obtain the tags of blockchain addresses and retrieve the entities to which the blockchain addresses belong;
[0069] Centered on entity E, retrieve the entity's transaction data, and construct a first-order directed transaction subgraph with the counterparty entities {E1, E2, ..., En} in the transaction data as first-order neighbor nodes;
[0070] Centered on first-order neighbor nodes, retrieve transaction data for each entity in the first-order directed transaction subgraph, and use the counterparty entities in the transaction data as second-order neighbor nodes of entity E to form a second-order directed subgraph of entity E; wherein the retrieval time is limited when retrieving the transaction data of entities.
[0071] Methods for constructing entity transaction characteristics include:
[0072] Obtain the tag of the blockchain address, lock the target entity corresponding to the blockchain address, and select entity feature information from the entity information table of the target entity to construct the entity transaction feature vector;
[0073] The entity feature information includes quantity features, amount features, and time features; the quantity features include: number of sub-addresses, total number of transactions, number of inbound transactions, number of outbound transactions, percentage of inbound transactions, and percentage of outbound transactions; the amount features include: total transaction amount, total inbound amount, total outbound amount, net inbound amount, maximum inbound amount, maximum outbound amount, average inbound amount, and average outbound amount; the time features include: earliest transaction time, most recent transaction time, and average transaction frequency.
[0074] Methods for constructing address transaction features include:
[0075] Obtain the tag of the blockchain address, lock the target entity corresponding to the blockchain address, and then map multiple blockchain addresses according to the target entity: the blockchain addresses of the same target entity have the same entity characteristics and different address characteristics. The address characteristics are standardized to form an address transaction feature vector.
[0076] The address characteristics include quantity characteristics, amount characteristics, and time characteristics; the quantity characteristics include: total transaction volume, number of inbound transactions, number of outbound transactions, percentage of inbound transactions, percentage of outbound transactions, and number of transactions with a unique address, etc.; the amount characteristics include: total transaction amount, total amount of inbound transactions, total amount of outbound transactions, average amount of inbound transactions, average amount of outbound transactions, and total transaction fees, etc.; the time characteristics include: earliest transaction time, most recent transaction time, and average transaction frequency, etc.
[0077] Address transaction features, as input to the XGBoost model, play a crucial role. Based on labeled address data, the target entities corresponding to the addresses are identified, and then more addresses are mapped from these entities. Addresses of the same entity share the same entity features, but their address features are distinct. Furthermore, there are significant differences in the number of entities contained in different categories and the number of addresses contained within different entities. To construct a balanced training dataset, a specific random sampling method is required. The sampling principles follow these three points:
[0078] 1. Retain tagged entities.
[0079] 2. The number of addresses in different categories should be kept as consistent as possible.
[0080] 3. Sampling is performed on small entities (containing fewer addresses) and on large entities (containing more addresses).
[0081] After completing the address sampling, the obtained address set D{d1,d2,...,d...} n This will be used as the training object for the XGBoost model. Features will be constructed for each address by combining the address information table and transaction data.
[0082] Step 3: Input the address transaction features, entity transaction features, and entity transaction subgraph into the XGBoost model and GCN model for training, and integrate the trained XGBoost model and GCN model to obtain the final classification model;
[0083] The training methods for the XGBoost model include:
[0084] Let the address set D{d1,d2,...,d n The XGBoost model is trained using the input of the input as follows:
[0085] The address transaction characteristics corresponding to each blockchain address are concatenated with the entity transaction characteristics of the entity to which it belongs to form an address characteristic set;
[0086] The address feature set is divided into a training set and a test set. The hyperparameters of the XGBoost model (such as the depth of the decision tree, the learning rate, etc.) are initialized, and the training set is used as input to train the XGBoost model. At the same time, the performance indicators of the XGBoost model are detected to adjust the model hyperparameters.
[0087] The performance of the XGBoost model was evaluated based on precision and recall metrics, and on a test set.
[0088] Training methods for GCN models include:
[0089] The GCN model is trained by using the set of entity subgraphs as input:
[0090] Construct a set of entity transaction subgraphs, where each entity transaction subgraph G = {A, E}, where A is the adjacency matrix, representing the connection relationship between entities, and E is the entity feature, representing the feature of each entity in the entity transaction subgraph.
[0091] The set of entity transaction subgraphs is divided into a training set and a test set;
[0092] Initialize the hyperparameters of the GCN model (such as the number of convolutional layers and the learning rate), and use the training set as the input to the GCN model. At the same time, detect the performance indicators of the GCN model to adjust the model hyperparameters.
[0093] The performance of the XGBoost model was evaluated based on precision and recall metrics, and on a test set.
[0094] Model ensemble methods include:
[0095] The results of the XGBoost and GCN models are combined using a weighted average strategy to obtain the final prediction result. The main steps are as follows:
[0096] Obtain the final performance metrics for the XGBoost and GCN models;
[0097] By combining the prediction results and final performance metrics of the XGBoost and GCN models, weights are assigned to the XGBoost and GCN models to obtain the final integrated classification model.
[0098] A small number of newly added samples are selected to test the final classification model and the weights are adjusted. These newly added samples are not in the training set or the prediction set.
[0099] Step 4: Predict the classification of blockchain address attributes based on the final classification model.
[0100] Example 2
[0101] This embodiment provides a blockchain address attribute classification system based on ensemble machine learning, used to implement the blockchain address attribute classification method based on ensemble machine learning described in the previous embodiment, including:
[0102] The data acquisition module is used to build a full blockchain node based on UTXO, acquire detailed transaction data of the blockchain and perform preprocessing.
[0103] The preprocessing module is used to construct features based on the preprocessed detailed transaction data: address analysis is performed on the preprocessed detailed transaction data to obtain address transaction features, entity clustering analysis is performed to obtain entity transaction features, and an entity transaction subgraph is constructed.
[0104] The model training module is used to input address transaction features, entity transaction features, and entity transaction subgraphs into the XGBoost model and GCN model for training. The trained XGBoost model and GCN model are then integrated to obtain the final classification model.
[0105] The output module is used to predict the classification of blockchain address attributes based on the final classification model.
[0106] The above embodiments fully consider the shortcomings of existing methods in feature construction and model selection, cleverly utilizing features from three aspects: address, entity, and entity relationship. They employ two machine learning techniques, XGBoost and GCN models, to provide a more accurate and reliable method for address classification. The UTXO-based blockchain address clustering technology aggregates addresses of the same entity, thereby obtaining more accurate entity information and providing richer features for subsequent classification. In addition to utilizing individual entity features, the construction of entity transaction subgraphs adds relational features between entities, further enhancing the richness of entity features. The above embodiments utilize two excellent machine learning models, XGBoost and GCN, and cleverly combine the prediction results of the two models through a weighted averaging strategy, thus greatly improving the robustness and reliability of the overall framework. During the construction of entity transaction subgraphs, by extracting transaction subgraphs from different time periods of the current entity, the problem of imbalanced entity label samples is addressed to some extent.
[0107] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. 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 blockchain address attribute classification method based on integrated machine learning, characterized in that, The application relates to a blockchain address attribute classification method based on a UTXO (unspent transaction output) blockchain. Step one: a UTXO-based blockchain full node is built, detailed transaction data of the blockchain is acquired, and preprocessing is performed; Step two: feature construction is performed based on the preprocessed detailed transaction data: address transaction features are obtained by performing address analysis on the preprocessed detailed transaction data, entity transaction features are obtained by performing entity clustering analysis, and an entity transaction subgraph is constructed; Step three: the address transaction features, the entity transaction features and the entity transaction subgraph are input into an XGBoost model and a GCN model for training, and the trained XGBoost model and the GCN model are integrated to obtain a final classification model; The training method of the XGBoost model comprises the following steps: The address transaction features corresponding to each blockchain address are spliced with the entity transaction features of the entity to which the blockchain address belongs, so as to form an address feature set; The address feature set is divided into a training set and a test set, the hyperparameters of the XGBoost model are initialized, the training set is input into the XGBoost model for training, and the performance indicators of the XGBoost model are detected, so that the hyperparameters of the model are adjusted; The accuracy and the recall rate are taken as indicators, and the performance of the XGBoost model is evaluated based on the test set; The training method of the GCN model comprises the following steps: An entity transaction subgraph set is constructed, each entity transaction subgraph G in the entity transaction subgraph set is composed of an adjacency matrix A and a connection relationship between entities E, and the features of each entity in the entity transaction subgraph are entity features E; The entity transaction subgraph set is divided into a training set and a test set; The hyperparameters of the GCN model are initialized, the training set is input into the GCN model, and the performance indicators of the GCN model are detected, so that the hyperparameters of the model are adjusted; The accuracy and the recall rate are taken as indicators, and the performance of the GCN model is evaluated based on the test set; The method of model integration comprises the following steps: The results of the XGBoost model and the GCN model are comprehensively processed based on a weighted average value strategy to obtain a final prediction result; the steps are as follows: The final performance indicators of the XGBoost model and the GCN model are acquired; The XGBoost model and the GCN model are assigned weights based on the prediction results and the final performance indicators of the XGBoost model and the GCN model to obtain an integrated final classification model; A small amount of new samples are selected to test the final classification model and adjust the weights, and the new samples are not in the training set and the prediction set; Step four: the final classification model is used to predict the attributes of blockchain addresses.
2. The integrated machine learning based blockchain address property classification method of claim 1, wherein, The preprocessing comprises the following steps: Transaction information of a transaction address is extracted from detailed transaction data to form an address information table, and the transaction information comprises a balance, a number of outgoing transaction records and a number of incoming transaction records; UTXO-based blockchain address clustering is performed to aggregate addresses of the same entity, and the aggregated address information is integrated to obtain entity basic information to form an entity information table, and the basic information comprises a number of entity associated addresses, a number of entity outgoing transaction records and a number of entity incoming transaction records. 3.The integrated machine learning based blockchain address property classification method of claim 2, wherein, The preprocessing further comprises the following steps: Labels of blockchain addresses are collected from open source channels; The open source channel includes a blockchain browser, an exchange announcement, social media, a community forum, and an authoritative third-party data analysis website; The label includes an annotation, a description, and a label category obtained by artificial integration of the blockchain address i. 4.The integrated machine learning based blockchain address attribute classification method of claim 1, wherein, The construction method of the entity transaction subgraph includes: Obtaining the label of the blockchain address and retrieving the entity to which the blockchain address belongs; Taking the entity E as the center, retrieving the transaction data of the entity, and constructing a first-order directed transaction subgraph with the counterparty entities {E1, E2,..., En} in the transaction data as the first-order neighbor nodes of the entity E; Taking the first-order neighbor nodes as the center, retrieving the transaction data of each entity in the first-order directed transaction subgraph, and taking the counterparty entities in the transaction data as the second-order neighbor nodes of the entity E to form a second-order directed subgraph of the entity E; wherein the retrieval time of the transaction data of the entity is limited. 5.The integrated machine learning based blockchain address attribute classification method of claim 1, wherein, The construction method of the entity transaction feature includes: Obtaining the label of the blockchain address, locking the target entity corresponding to the blockchain address, and selecting entity feature information in the entity information table of the target entity to construct an entity transaction feature vector; The entity feature information includes quantity characteristics, amount characteristics, and time characteristics; the quantity characteristics include: sub-address quantity, total transaction quantity, incoming transaction quantity, outgoing transaction quantity, incoming transaction proportion, and outgoing transaction proportion; the amount characteristics include: total transaction amount, incoming total amount, outgoing total amount, net incoming amount, maximum incoming amount, maximum outgoing amount, average incoming amount, and average outgoing amount; the time characteristics include: earliest transaction time, latest transaction time, and average transaction frequency. 6.The integrated machine learning based blockchain address attribute classification method of claim 1, wherein, The construction method of the address transaction feature includes: Obtaining the label of the blockchain address, locking the target entity corresponding to the blockchain address, and mapping a plurality of blockchain addresses according to the target entity: the blockchain addresses of the same target entity have the same entity features and different address features, and the address features are standardized to form an address transaction feature vector; The address feature includes quantity characteristics, amount characteristics, and time characteristics; the quantity characteristics include: total transaction amount, incoming transaction quantity, outgoing transaction quantity, incoming transaction proportion, outgoing transaction proportion, and unique address transaction quantity; the amount characteristics include: total transaction amount, incoming transaction total amount, outgoing transaction total amount, average incoming amount, average outgoing amount, and total handling fee; the time characteristics include: earliest transaction time, latest transaction time, and average transaction frequency.
7. A blockchain address attribute classification system based on integrated machine learning, characterized by, The method for realizing the blockchain address attribute classification based on integrated machine learning according to any one of claims 1-6 includes: The acquisition module is used to build a UTXO-based blockchain full node, obtain detailed transaction data of the blockchain, and perform preprocessing; The preprocessing module is used to perform feature construction based on the preprocessed detailed transaction data: address analysis is performed on the preprocessed detailed transaction data to obtain address transaction features, entity clustering analysis is performed to obtain entity transaction features, and an entity transaction subgraph is constructed; The model training module is configured to input the address transaction features, the entity transaction features and the entity transaction subgraph into an XGBoost model and a GCN model for training, and to integrate the trained XGBoost model and the GCN model to obtain a final classification model. The output module is configured to predict the blockchain address attribute classification based on the final classification model.