Cryptocurrency auxiliary supervision method based on composite feature node category mining

By utilizing traffic data and external auxiliary systems in the cryptocurrency network, combined with machine learning models, accurately identifying cryptocurrency node categories, the accuracy of node type judgment in the existing technology is solved, and the regulatory efficiency and scope of cryptocurrency transactions are improved. .

CN119067661APending Publication Date: 2024-12-03BEIJING INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410983488.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing methods for judging cryptocurrency node types generally adopt heuristic rules, which cannot accurately reflect the inherent differences between full nodes and light nodes, resulting in judgment conflicts and inability to judge, making it difficult to effectively regulate cryptocurrency transactions in an anonymous environment.

Method used

Adopt traffic data analysis and external auxiliary judgment system based on blockchain network, combined with machine learning models, and through traffic analysis, message statistics, feature engineering and model training, accurate mining of cryptocurrency node categories is achieved, and effective supervision of the regulatory platform is improved. scope.

Benefits of technology

By building a high confidence machine learning model, accurately identifying light nodes and full nodes, enhancing the supervision of the regulatory platform, and improving the regulatory efficiency and accuracy of cryptocurrency transactions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067661A_ABST
    Figure CN119067661A_ABST
Patent Text Reader

Abstract

The invention relates to a cryptocurrency auxiliary supervision method based on composite feature node category mining, and belongs to the field of block chain supervision. The method comprises the following steps of: firstly, analyzing broadcast traffic for capturing cryptocurrency transaction from a block chain network, and obtaining a node list participating in an interaction behavior and a corresponding message type; and summarizing and classifying the traffic data within a period of time to obtain the statistical magnitude of message types. And the grabbed flow data is transmitted to an external auxiliary system, and category judgment results of part of nodes and a list of nodes which cannot be judged are returned. The quality of training samples is improved, a machine learning model is constructed based on a clustering thought, and the model is adjusted according to a coarse-grained label provided by an external system. And the residual nodes which cannot be judged are predicted by using the trained model to obtain a node category with high confidence, light nodes in the node category are taken into a key supervision target, and the range supervision strength is enhanced. According to the invention, encryption currency node category mining is realized, and the effective supervision range proportion of a supervision platform is increased.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a cryptocurrency auxiliary supervision method based on composite feature node category mining, belonging to the technical field of blockchain supervision. Background Art

[0002] The core foundation of cryptocurrency lies in blockchain technology, which is a distributed database system that uses a chain block architecture to record transaction history. It constitutes a distributed ledger system that is globally shared, does not require centralized authority management, and has transparency and decentralization.

[0003] Here is an overview of the main features and functions of cryptocurrency networks:

[0004] 1. Decentralized architecture: The cryptocurrency network abandons the traditional centralized management model and adopts a decentralized architecture design. This means that the operation and maintenance of the network no longer rely on a single central agency, but are coordinated by many participants (nodes) around the world through a peer-to-peer network.

[0005] 2. Distributed ledger: As the core technology of cryptocurrency, blockchain is essentially an open, transparent, and tamper-proof distributed ledger system. Each network node maintains a complete copy of the ledger, and the data consistency between nodes is ensured through the consensus process.

[0006] 3. Consensus mechanism: Consensus mechanism plays a vital role in the cryptocurrency network. It is the key means to ensure consensus in a decentralized system. These mechanisms ensure that only transactions that are confirmed as valid by the majority of nodes in the network will be integrated into the blockchain.

[0007] 4. Network communication and topology: The nodes within the cryptocurrency network are interconnected according to a specific communication protocol, forming a complex topological structure similar to a tree or mesh. Such a structure is not only conducive to the rapid dissemination of information and the rapid achievement of global consensus, but also enhances the network's resistance to attacks and failures.

[0008] 5. Security and anonymity: Thanks to the properties of distributed systems, cryptocurrencies have demonstrated a high degree of security at the network level, because in most cases, tampering with recorded historical data requires the control of a large number of nodes, which is extremely difficult in a large-scale decentralized network. At the same time, although transaction details are publicly available on the blockchain, with the help of advanced public key cryptography technology, cryptocurrencies can provide certain anonymity protection or pseudonymous transaction features.

[0009] 6. Incentive mechanism: In order to motivate nodes to actively participate in transaction verification, new block generation and network security, the cryptocurrency network has introduced a complete set of economic incentive mechanisms. When a full node successfully creates a new block, it will receive a newly issued cryptocurrency as a reward. This positive feedback mechanism has effectively promoted the self-stabilization and sustainable development of the network.

[0010] Due to the decentralized nature of blockchain, various cryptocurrency transactions are not constrained and regulated, which also poses challenges to monetary policy and economic stability. Therefore, active regulation of cryptocurrencies is an important measure to protect the security of multiple parties. Through blockchain analysis, we can understand its network topology and operation logic, and trace and deal with security threats or illegal activities when they occur.

[0011] Since cryptocurrency is anonymous at the transaction level, it is necessary to mine information at the network traffic level to assist supervision. There are mainly two types of network nodes with different behavioral characteristics in the blockchain network of cryptocurrency: full nodes and light nodes. The two have the following differences: full nodes store all historical data of the blockchain network, that is, every block from the genesis block to the present and all transaction records contained therein, while light nodes do not store complete blockchain data, only save the block header information of the blockchain or other necessary minimum data sets; full nodes execute all the rules and protocols of the blockchain, and can independently verify the validity of each transaction and block without relying on other nodes, while light nodes usually need to rely on the services provided by full nodes to verify transactions, and they cannot independently verify the complete legitimacy of transactions and blocks; due to the need to store the entire blockchain, full nodes require a large hard disk space, sufficient computing power and a stable network connection to continuously synchronize and verify new blocks, while light nodes require fewer resources and are suitable for running on mobile devices or low-end computers, which reduces the entry threshold. In summary, full nodes provide complete functionality and the highest security, but require more hardware resources; light nodes sacrifice some independence and security in exchange for lower operating costs and higher convenience, and are suitable for transaction verification and query for ordinary users or resource-constrained environments. For regulatory platforms, mining cryptocurrency node categories and screening out light nodes will help increase the effective regulatory scope of regulatory platforms.

[0012] Based on the above characteristics of full nodes and light nodes, their behaviors in cryptocurrency networks will also be significantly different. However, existing technologies generally use rule-making to determine and mine network node types, such as node online time records or communication specific field identification. These heuristic rules cannot fully reflect the fundamental differences between the two, so there will be multiple rule judgment conflicts or undecidable situations. Using machine learning combined with composite features can more accurately fit the intrinsic characteristics of full nodes and light nodes, efficiently assist in node type mining, and provide strong support for cryptocurrency regulation. Summary of the invention

[0013] The purpose of the present invention is to solve the technical problem of how to enhance effective supervision in the anonymization environment of cryptocurrency transactions, and creatively proposes a cryptocurrency assisted supervision method based on composite feature node category mining.

[0014] The present invention is based on a blockchain network and an external auxiliary judgment system, which respectively provide network traffic data and coarse-grained node category judgment.

[0015] Network traffic data: When a new transaction record is generated or a new block is encapsulated in the blockchain network, this information will be transmitted to each node in the network immediately. In this process, a series of interactive actions will be performed between nodes, including mutual authentication, synchronization of the latest block information, query of on-chain status, and response to various requests. These interactive behaviors naturally trigger the generation and flow of network traffic. Various interactive behaviors are represented by different traffic messages, and in turn different message types can also reflect different types of nodes.

[0016] External auxiliary judgment system: Based on the existing detection and analysis technology, a system for automatically judging the node category is constructed, including long-line detection based on spatiotemporal characteristics and rule judgment based on protocol characteristics. The different characteristic performances of data packets of different categories of nodes when performing specific behaviors are used to realize the classification of network node types.

[0017] Due to the limitations of related technologies, these two methods are difficult to give clear judgment results for some nodes.

[0018] like Figure 1 As shown, the method of the present invention includes the steps of traffic analysis, message statistics, external system assistance, feature engineering, model training, model prediction, etc.

[0019] First, we parse the broadcast traffic of cryptocurrency transactions captured from the blockchain network to obtain a list of nodes involved in the interaction and their corresponding message types. We then aggregate and classify the traffic data over a period of time to obtain statistics on message types.

[0020] At the same time, the captured traffic data is transmitted to the external auxiliary system, and the category judgment results of some nodes and the list of nodes that cannot be judged are returned.

[0021] Afterwards, feature screening, feature scaling and other techniques are used to improve the quality of training samples, a machine learning model is built based on the idea of ​​clustering, and the model is adjusted according to the coarse-grained labels provided by the external system.

[0022] Finally, the trained model is used to predict the remaining nodes that cannot be judged, and high-confidence node categories are obtained. The light nodes among them are included in the key supervision targets to strengthen the scope of supervision.

[0023] Beneficial Effects

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] The method of the present invention is based on the cryptocurrency flow data in the blockchain network, combined with the preliminary judgment of the external auxiliary system, and simply and efficiently realizes the cryptocurrency node category mining by constructing a machine learning model, thereby increasing the effective supervision scope ratio of the supervision platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the node category mining process of the present invention.

[0027] Figure 2 Flow chart of the method of the present invention.

[0028] Figure 3 This is a structural diagram of the relevant traffic analysis of the present invention. DETAILED DESCRIPTION

[0029] The method of the present invention is further described in detail below with reference to the accompanying drawings.

[0030] like Figure 2 As shown, a cryptocurrency auxiliary supervision method based on composite feature node category mining includes the following steps:

[0031] Step 1: Traffic analysis.

[0032] Specifically, the gateway is used to capture the traffic data in the cryptocurrency network for processing and analysis. Each data traffic packet contains a large amount of information, and it is necessary to filter out the required important information. Therefore, for the node category mining problem, the node IP address and the message type of the traffic sent as the source IP are obtained according to the traffic format.

[0033] Taking encrypted digital currency as an example, its traffic in the blockchain network contains 27 types of messages. The following are some of the core message types:

[0034] 1) version: When a node creates an outgoing connection, it will immediately announce the protocol version it is using. The remote node will respond with its version. No further communication is possible until the two nodes have exchanged versions.

[0035] 2) verack: The verack message is a reply to version. This message has only one header.

[0036] 3) addr: address. Provides information about known nodes on the network. A node can send an address list to other nodes through this message.

[0037] 4) inv: Nodes can use this message to announce known transaction information. This message can be sent actively or used to respond to getblocks messages.

[0038] 5) getdata: Get data. Used to respond to inv and retrieve the content of a specific transaction. Usually sent after receiving the inv packet and filtering known transaction information.

[0039] In order to distinguish different message types, the captured traffic needs to be parsed. Figure 3 It shows the structure of the traffic packet of cryptocurrency, including the message header and data payload. The 12-byte instruction identifier indicates the message type of the current traffic; the magic number is used to find the next message when the information flow status is unknown; the data length can indicate the length of the data payload; the checksum is used to ensure the integrity of the message; and the data payload contains the specific content of the message.

[0040] Therefore, the IP addresses of the nodes involved in transaction verification and the types of messages they send can be obtained from the captured traffic packets.

[0041] Step 2: Message statistics.

[0042] Specifically, after a period of traffic capture and traffic analysis, all node IP addresses are extracted from the node IP addresses and their statistical data to form a set C IP .

[0043] Summarize the traffic associated with the same node and count the number of times each message type appears during this period of time N i , the message type statistics of all nodes constitute the set N IP .

[0044] Since each message has different functions, the statistics of message types can reflect the behavioral characteristics of different nodes, thus serving as a composite feature for judging their node types.

[0045] Step 3: External system assistance.

[0046] Specifically, before using composite features to mine node categories, an existing external system is first used to perform category judgment to obtain category labels for some nodes.

[0047] Among them, the external system uses the spatiotemporal behavior characteristics and protocol characteristics of the nodes for coarse-grained judgment. Due to the limitations of the external system, it cannot completely and accurately judge the node category, and it will be impossible to judge for some nodes with unclear behavior characteristics. Therefore, after the external system assists in decision-making, the set C IP All IP addresses are given a preliminary judgment result and divided into three subsets: C SPV , C FULL and C OTHER , representing the light node set, full node set and undeterminable set respectively.

[0048] Step 4: Feature Engineering.

[0049] Specifically, feature construction and preprocessing have an important impact on subsequent model training and are an indispensable step in the field of machine learning. This step includes feature selection and feature scaling.

[0050] First, the importance of features is evaluated by calculating statistical indicators between features (such as correlation, variance, information gain, etc.). Then the most important features are selected to improve model performance, reduce computational costs, and reduce overfitting. After that, normalization is used to scale the feature data to a specific range, namely Min-Max Scaling. The normalized feature X norm It is expressed as: Where X is the original feature, X max is the maximum value of the feature, X min is the minimum value of the feature.

[0051] After feature engineering, the statistical data corresponding to each IP address is converted into its feature vector, corresponding to the three subsets divided in step 3, and its node feature subset is X SPV , X FULL and X OTHER .

[0052] Step 5: Model training.

[0053] Specifically, by encoding and feature engineering the composite features of the nodes, we obtain feature vectors that can implicitly represent the properties of the nodes. Combined with machine learning methods for training, we obtain a high-performance model with category prediction capabilities to achieve automated and large-scale prediction.

[0054] In order to finally classify the nodes into light nodes and full nodes, the idea of ​​clustering is adopted to perform category mining on the nodes, and the number of categories is set to 2.

[0055] The light nodes and full nodes that have been preliminarily judged are used as training sets, and their corresponding features are fused into the training set feature X train =X SPV ∪X FULL , the training set label is C train =C SPV ∪C FULL . Select clustering model M, based on feature set X train Train the model.

[0056] The training process includes:

[0057] First, from the feature set X train Randomly select the feature vector x of 2 IPs 1 ,x 2 As the initial cluster center, denoted as Represents the two cluster centers updated at the 0th time and forms two corresponding category sets

[0058] Then, calculate X train The feature vector x of other nodes i in i To the cluster center and distance (such as Euclidean distance), we get d i1 ,d i2 .

[0059] Compare distance d i1 ,d i2 The size of d i1 <d i2 , then let C 1 =C 1 ∪{x i}, otherwise let C 2 =C 2 ∪{x i}. Repeat this process until all IPs are added to both sets.

[0060] Afterwards, calculate the set C 1 The point with the smallest average distance to all IP feature vectors And use it as the new cluster center of category 1, C 2 Similarly, the new cluster center of category 2 is obtained.

[0061] Repeat the above process until the cluster center Convergence: The convergence judgment rule is that the distance between the new cluster center and the previous round of cluster center is less than the predetermined value ε.

[0062] Finally, according to C train Evaluate the clustering accuracy of model M and adjust the model based on the results.

[0063] Try to train all clustering models, select the best performing model as the algorithm's final model M, and identify two cluster centers μ SPV ,μ FULL , represent the cluster centers of light nodes and full nodes respectively.

[0064] Step 6: Model prediction.

[0065] After training and tuning, we get a machine learning model M with discrimination ability. We use the model to identify the IP address set C that cannot be identified in the initial judgment of the external system. OTHER Perform clustering and update the set C of light nodes and full nodes SPV ,C FULL .

[0066] Specifically, model prediction includes the following steps:

[0067] Step 1: Calculate the feature set X OTHER The feature vector x of each node i in i To two cluster centers μ SPV , μ FULL distance (such as Euclidean distance), we get d i1 d i2 .

[0068] Step 2: Compare distance d i1 ,d i2 The size of d i1 <d i2 , then let C FULL =C FULL ∪{x i}, otherwise let C SPV =C SPV ∪{x i}. Use the model to calculate the confidence level Conf of the current judgment result i .

[0069] Repeat the above process until all IP addresses are added to both sets.

[0070] The model prediction results are rewritten into the external auxiliary judgment system, and the IP corresponding to the light node is marked and included in the key monitoring and regulatory target list to provide direction for cryptocurrency regulation and narrow its scope.

Claims

1. A cryptocurrency assisted supervision method based on composite feature node category mining, based on a blockchain network and an external auxiliary judgment system, characterized by: First, the broadcast traffic of cryptocurrency transactions captured from the blockchain network is parsed to obtain a list of nodes participating in the interactive behavior and their corresponding message types; the traffic data within a period of time is summarized and classified to obtain the statistics of message types; at the same time, the captured traffic data is passed to an external auxiliary judgment system, and the category judgment results of some nodes and a list of nodes that cannot be judged are returned; After that, feature selection and feature scaling are used to improve the quality of training samples, a machine learning model is built based on the idea of ​​clustering, and the model is adjusted according to the coarse-grained labels provided by the external auxiliary judgment system; Finally, the trained model is used to predict the remaining nodes that cannot be judged, and the node categories with high confidence are obtained. The node categories are finally divided into light nodes and full nodes, and the light nodes are included in the key supervision targets.

2. A cryptocurrency assisted supervision method based on composite feature node category mining as claimed in claim 1, characterized in that: When parsing broadcast traffic, for the node category mining problem, the node IP address and the message type of the traffic sent as the source IP are obtained according to the traffic format; The message types include: version: version; when a node creates an outgoing connection, it will immediately announce the protocol version it is using; the remote node will respond with its version; no further communication can take place until the two nodes have exchanged versions; verack: The verack message is a reply to version. This message has only one header. addr: address; provides information about known nodes on the network; the node sends an address list to other nodes through this message; inv: The node announces known transaction information through this message; getdata: Get data; used to respond to inv and retrieve the content of a specific transaction; after receiving the inv packet, it filters the known transaction information and sends it.

3. A cryptocurrency assisted supervision method based on composite feature node category mining as claimed in claim 1, characterized in that: After a period of traffic capture and traffic analysis, all node IP addresses are extracted from the node IP addresses and their statistical data to form a set C IP ; Summarize the traffic associated with the same node and count the number of times each message type appears during this period of time N i , the message type statistics of all nodes constitute the set N IP ; The external auxiliary judgment system uses the spatiotemporal behavior characteristics and protocol characteristics of the nodes to make coarse-grained judgments on the set C IP All IP addresses are given a preliminary judgment result and divided into three subsets: C SPV , C FULL and C OTHER , respectively represent the light node set, the full node set and the undeterminable set; Use feature selection and feature scaling to improve the quality of training samples. First, evaluate the importance of features by calculating statistical indicators between features. Then the most important features are selected to improve model performance, reduce computational cost, and reduce overfitting; after that, normalization is used to scale the feature data to a specific range, i.e., minimum-maximum scaling; the normalized feature X norm It is expressed as: Where X is the original feature, X max is the maximum value of the feature, X min is the minimum value of the feature; After the above processing, the statistical data corresponding to each node IP address is converted into its feature vector, corresponding to the three subsets divided above, and its node feature subset is X SPV , X FULL and X OTHER ; In order to finally classify the node categories into light nodes and full nodes, the idea of ​​clustering is adopted to mine the node categories, and the number of categories is set to 2; the light nodes and full nodes that have been preliminarily judged are used as training sets, and their corresponding features are fused into the training set feature X train =X SPV ∪X FULL , the training set label is C train =C SPV ∪C FULL ; Select clustering model M, according to the feature set X train Train the model; After training and tuning, a machine learning model M with discrimination ability is obtained. The model is used to judge the IP address set X that cannot be judged in the initial judgment of the external auxiliary judgment system. OTHER Perform clustering and update the set C of light nodes and full nodes SPV , C FULL .

4. A cryptocurrency assisted supervision method based on composite feature node category mining as described in claim 3, characterized in that: The model training process is: First, from the feature set X train Randomly select the feature vectors x1 and x2 of two IPs as the initial clustering centers, denoted as Represents the two cluster centers updated at the 0th time and forms two corresponding category sets Then, calculate X train The feature vector x of other nodes i in i To the cluster center and distance (such as Euclidean distance), we get d i1 ,d i2 ; Compare distance d i1 ,d i2 The size of d i1 <d i2 , then let C1=C1∪{x i }, otherwise let C2=C2∪{x i }; Repeat this process until all IPs are added to both sets; After that, calculate the point in set C1 with the smallest average distance to all IP feature vectors And use it as the new cluster center of category 1, and the same goes for C2 to get the new cluster center of category 2; Repeat the above process until the cluster center Convergence: The convergence rule is that the distance between the new cluster center and the previous cluster center is less than the preset value ε; Finally, according to C train Evaluate the clustering accuracy of model M and adjust the model based on the results; Try to train all clustering models, select the best performing model as the algorithm's final model M, and identify two cluster centers μ SPV ,μ FULL , represent the cluster centers of light nodes and full nodes respectively.

5. A cryptocurrency assisted supervision method based on composite feature node category mining as claimed in claim 4, characterized in that: Model predictions The following steps are involved: first step: Calculate the feature set X OTHER The feature vector x of each node i in i To two cluster centers μ SPV , μ FULL The distance d i1 d i2 ; Step 2: Compare distance d i1 ,d i2 The size of d i1 <d i2 , then let C FULL =C FULL ∪{x i }, otherwise let C SPV =C SPV ∪{x i }; Use the model to calculate the confidence level Conf of the current judgment result i ; Repeat the above process until all IPs are added to the two sets; The model prediction results are rewritten into the external auxiliary judgment system, and the IP corresponding to the light node is marked and included in the key monitoring and regulatory target list.