Self-adaptive cross-block chain anonymous account supervision method and device based on space-time diagram learning

Abstract

The invention discloses a self-adaptive cross-block chain anonymous account supervision method and device based on space-time diagram learning, and the method comprises the steps: constructing an anonymous account space-time behavior diagram model for the time dynamics and space heterogeneity of block chain anonymous account behaviors; based on the anonymous account spatio-temporal behavior graph model, for a single anonymous account spatio-temporal behavior graph, extracting spatio-temporal behavior characteristics in the single anonymous account spatio-temporal behavior graph by using a designed time sequence heterogeneous graph learning mechanism; based on an anonymous account space-time behavior graph model, aiming at a plurality of anonymous account space-time behavior graphs, utilizing a designed time sequence heterogeneous graph twin network to measure space-time behavior characteristics among different anonymous accounts, and obtaining space-time behavior similarity among different anonymous accounts; for different block chain environments, the measurement of spatial-temporal behavior similarity among different anonymous accounts in a cross-block chain environment is adaptively adjusted and optimized by using a designed similar perception graph-based transfer learning mechanism, and adaptive cross-block chain anonymous account identity association is realized. The objective of the invention is to accurately model real-time and diversified transaction behaviors of the block chain anonymous account, improve the transaction feature expression ability of the real-time anonymous account, effectively mine the spatio-temporal behavior features of the anonymous account, and realize adaptive and efficient anonymous account identity association in a cross-block chain transaction ecology.

Description

Technical Field

[0001] The present invention belongs to the field of blockchain security, and specifically relates to an adaptive cross-blockchain anonymous account supervision method and device based on spatiotemporal graph learning. Background Art

[0002] With the rapid development of decentralized finance in blockchain systems, on-chain anonymity enhancement services represented by mixers have been widely deployed and applied. By effectively confusing the flow of transaction funds, these services are used by a large number of on-chain criminals. Criminals often use multiple sub-accounts to use on-chain anonymity enhancement services to conceal attack transactions and confuse attack proceeds. They complete contract vulnerability exploitation and illegal proceeds laundering without revealing their true identities, making blockchain systems a hotbed for illegal activities. Therefore, there is an urgent need to regulate blockchain systems. Due to the transparency of blockchain systems, the complete transaction history of each on-chain account is available to regulators, making it possible to identify and track malicious participants. However, traditional account identity classification and supervision schemes rely on known account behavior patterns to classify account identities, and it is difficult to identify unknown account behavior patterns hidden under on-chain anonymity enhancement services.

[0003] The account identity association supervision scheme associates multiple anonymous sub-accounts controlled by the same criminal based on the similarity of transaction patterns and behaviors, thereby identifying the behavior patterns of unknown accounts in the context of on-chain anonymity enhancement services. However, 1) the real blockchain transaction environment is heterogeneous and changes in real time. Existing account identity association supervision schemes usually use static and isomorphic account transaction feature modeling, which makes it difficult to accurately characterize the dynamic transaction behavior characteristics of accounts; 2) Existing account identity association supervision schemes are mostly based on heuristic rules and are designed for specific blockchain ecosystems, making it difficult to achieve portable account identity association across blockchains; 3) Existing account identity association supervision schemes only simply mine the topological behavior characteristics of on-chain account transactions, and lack in-depth mining of the spatiotemporal characteristics of account transaction behaviors. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides an adaptive cross-blockchain anonymous account supervision method and device based on spatiotemporal graph learning, which aims to accurately model the real-time and diversified transaction behaviors of blockchain anonymous accounts, improve the ability to express real-time anonymous account transaction characteristics, effectively mine the spatiotemporal behavior characteristics of anonymous accounts, and realize adaptive and efficient anonymous account identity association in the cross-blockchain transaction ecosystem, thereby realizing effective and safe supervision of criminal anonymous sub-accounts in large-scale on-chain transactions.

[0005] In order to solve the above technical problems, the present invention is implemented by the following technical solutions:

[0006] According to a first aspect of the present invention, there is provided an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning, comprising:

[0007] Aiming at the temporal dynamics and spatial heterogeneity of anonymous account behavior in blockchain, a spatiotemporal behavior graph model of anonymous accounts is constructed;

[0008] Based on the anonymous account spatiotemporal behavior graph model, for a single anonymous account spatiotemporal behavior graph, the designed temporal heterogeneous graph learning mechanism is used to extract the spatiotemporal behavior features in the spatiotemporal behavior graph of the single anonymous account;

[0009] Based on the anonymous account spatiotemporal behavior graph model, for multiple anonymous account spatiotemporal behavior graphs, the designed temporal heterogeneous graph twin network is used to measure the spatiotemporal behavior characteristics between different anonymous accounts to obtain the spatiotemporal behavior similarity between different anonymous accounts;

[0010] Aiming at different blockchain environments, the designed similarity perception graph-based transfer learning mechanism is used to adaptively tune the measurement of the spatiotemporal behavior similarity between different anonymous accounts in the cross-blockchain environment, so as to realize adaptive cross-blockchain anonymous account identity association.

[0011] In a possible implementation of the first aspect, the anonymous account spatiotemporal behavior graph model is constructed based on the temporal dynamics and spatial heterogeneity of the blockchain anonymous account behavior, specifically:

[0012] Based on the temporal dynamics of on-chain account transaction behaviors, as well as the heterogeneity of transaction types and transaction accounts, a native account spatiotemporal behavior graph model is constructed;

[0013] The native account spatiotemporal behavior graph model is denoised based on a heuristic graph pruning algorithm to obtain the anonymous account spatiotemporal behavior graph model.

[0014] In a possible implementation manner of the first aspect, the native account spatiotemporal behavior graph model is specifically:

[0015]

[0016] In the formula, It represents the self-blockchain transaction network centered on the anonymous account a on the chain to be supervised, that is, the native account spatiotemporal behavior graph model of the anonymous account a; Representing the Self-Blockchain Transaction Network The depth of breadth-first search traversal; V represents the set of on-chain address nodes traversed in breadth-first search; E represents the set of all on-chain transactions traversed in breadth-first search; Indicates the address type from the on-chain address node set V to the on-chain address type ε is the real external owner account, b is the blockchain robot account, and s is the blockchain smart contract account; Indicates the transaction type from the on-chain transaction set E to the on-chain transaction type ∈ is an external transaction, c is a contract transaction, ι is an internal transaction, and τ is a token transaction; Γ:E→Ω represents the time from the on-chain transaction set E to the on-chain transaction The mapping of ξ s Indicates the transaction start time. Indicates the duration of the transaction;

[0017] The heuristic graph pruning algorithm is used to denoise the native account spatiotemporal behavior graph model, specifically:

[0018] Identify an overloaded node set in the native account spatiotemporal behavior graph model, where the overloaded node set in the native account spatiotemporal behavior graph model is:

[0019]

[0020] Where Ψ(i) represents the on-chain address type of node i; D + (i) represents the in-degree of node i; D - (i) represents the out-degree of node i; Indicates the tolerance value of overloaded nodes;

[0021] The rotten node neighbor set under the branch subgraph connected to the overloaded node is classified, and the rotten node neighbor set under the branch subgraph connected to the overloaded node is:

[0022]

[0023] In the formula, Represents the spatiotemporal behavior graph model of the overloaded node m in the native account The neighbor nodes found by the first-order breadth-first search; Indicates the earliest on-chain transaction among neighbor nodes. neighbor nodes; is the denoising coefficient;

[0024] The branch subgraph connected to the neighbors of the corrupted nodes is pruned to complete the denoising of the native account spatiotemporal behavior graph model, and obtain the anonymous account spatiotemporal behavior graph model.

[0025] In a possible implementation manner of the first aspect, the designed temporal heterogeneous graph learning mechanism is specifically:

[0026] A heterogeneous temporal edge aggregation method is designed for edge signals in the spatiotemporal behavior graph model of anonymous accounts, and the initial node features of each graph node are calculated.

[0027] According to the initial node features obtained by the heterogeneous time-series edge aggregation method as the initial features of point aggregation, a heterogeneous time-series point aggregation method is designed for the point signals within the anonymous account spatiotemporal behavior graph model, and the node embedding of each graph node is calculated;

[0028] According to the node embedding of each graph node, a graph pooling method is designed for the subgraph signals in the anonymous account spatiotemporal behavior graph model, and the graph embedding of the anonymous account spatiotemporal behavior graph model is calculated.

[0029] In a possible implementation of the first aspect, the method for designing a heterogeneous temporal edge aggregation method for edge signals in an anonymous account spatiotemporal behavior graph model is specifically:

[0030] For a given node u in the anonymous account spatiotemporal behavior graph model, Indicates a given edge type The set of adjacent edges within u, each edge Attention score The calculation is as follows:

[0031]

[0032] In the formula, is the transposed matrix of attention parameters; f i Is the edge type The vectorized transaction information of edge i in the figure. For edge i, the nonlinear activation function Relu is used to calculate the attention score; for the attention score of edge i Further normalized by the Softmax function, the importance weight of edge i based on the attention mechanism The calculation is as follows:

[0033]

[0034] Based on the adjacent edge set The attention-based importance weights are applied to aggregate the information of adjacent edges. The node u is in the edge type Edge aggregation within The calculation is as follows:

[0035]

[0036]

[0037] In the formula, represents the connection operation; Relu represents the Relu nonlinear activation function; H is the number of attention heads used in the multi-head attention method; for each edge aggregation The timestamp is defined as The latest interaction in is calculated as follows:

[0038]

[0039] Aggregate by edge Timestamp The edge aggregation is sorted in ascending order, and the initial node feature of each node u is calculated as follows:

[0040]

[0041]

[0042] q u =LSTM([Sequence u ])

[0043] In the formula, LSTM is a long short-term memory network; Sort + Indicates that the time series is sorted in ascending order.

[0044] In a possible implementation of the first aspect, the method for designing a heterogeneous time-series point aggregation method for internal point signals of an anonymous account spatiotemporal behavior graph model is specifically:

[0045] For a given node x in the anonymous account spatiotemporal behavior graph model, Node Type The set of adjacent nodes in the node Time attribute From the adjacent side Latest timestamp The calculation is as follows:

[0046]

[0047] According to time attribute Generate the temporal importance weight of node u The calculation is as follows:

[0048]

[0049] According to the timing importance weight Generate Node Type Intra-class point aggregation of middle node x The calculation is as follows:

[0050]

[0051] In the formula, MLP (k) Indicates a multilayer perceptron with k layers; Is the node x in the node type The k-th layer embedding result in ; is the input of the kth layer, and Neighbors The temporal importance weight of , indicating that the influence of neighbors on node x changes over time;

[0052] According to the intra-class point aggregation, node x is embedded in the node of type β Attention score The calculation is as follows:

[0053]

[0054] Where LeakyRelu is the zero negative correction activation function; is the transposed attention parameter;

[0055] According to the node embedding of the node x in type β The attention score of node x, the aggregation of inter-class points, is the node embedding, which is calculated as follows:

[0056]

[0057] In the formula, ρ x is the between-class point aggregation of node x.

[0058] In a possible implementation of the first aspect, the graph pooling method for subgraph signals in the spatiotemporal behavior graph model for anonymous accounts is specifically:

[0059]

[0060] In the formula, ρ i is the heterogeneous temporal node embedding of node i; a i is the attention weight of node i, a i The specific calculation is as follows:

[0061]

[0062] In the formula, is the transpose of the learnable attention vector; W is the learnable weight matrix; V is the node set of the anonymous account spatiotemporal graph model.

[0063] In a possible implementation manner of the first aspect, the designed temporal heterogeneous graph twin network is specifically:

[0064] Construct a graph-level twin network for multiple anonymous account spatiotemporal behavior graphs. For a given anonymous account spatiotemporal behavior graph j, the graph embedding f j The calculation is as follows:

[0065]

[0066] Where Pool is the graph pooling method; V j is the node set of the anonymous account spatiotemporal behavior graph j; It is the node embedding of node v in the last layer after L layers of heterogeneous time series edge aggregation and heterogeneous time series point aggregation:

[0067]

[0068] Where, HTAGG n It is a heterogeneous time-series point aggregation method; is the set of neighbor nodes of node v; is the aggregate edge embedding of node v at level l+1:

[0069]

[0070] Where, HTAGG e It is a heterogeneous time series edge aggregation method; is the edge embedding from node u to node v at layer l;

[0071] Initialize with a random vector;

[0072] The designed time-series heterogeneous graph twin network is used to measure the spatiotemporal behavior characteristics between different anonymous accounts, specifically:

[0073] According to the time-space behavior graph of the anonymous account to be compared and The graph embedding f q and f p , use Euclidean distance to measure the similarity of spatiotemporal behaviors between different anonymous accounts:

[0074]

[0075] Where d is the graph embedding f q and f p The vector dimension of .

[0076] In a possible implementation of the first aspect, the method of adaptively tuning the measure of spatiotemporal behavior similarity between different anonymous accounts in a cross-blockchain environment by using the designed similarity perception graph transfer learning mechanism is specifically as follows:

[0077] For a single blockchain scenario, the anonymous account spatiotemporal behavior graph model is pre-trained to obtain a pre-trained basic model. is a dataset on the source blockchain, and is the spatiotemporal behavior graph of the anonymous account for the i-th query-comparison sample pair, Represents the self nodes in the two graphs, i.e., whether the comparison account and query account are linked to the same off-chain identity, N S Represents the number of query-comparison pairs in the dataset constructed from the source blockchain; the purpose of pre-training is to minimize the number of samples in the source dataset for the temporal heterogeneous graph twin network. The loss function on

[0078]

[0079] In the formula, and They are and Graph embedding of is the Euclidean distance between the two graph embeddings, and the parameters θ of the temporal heterogeneous graph twin network are updated by gradient descent:

[0080]

[0081] Where η is the learning rate;

[0082] According to the pre-trained basic model, cross-blockchain model tuning is performed based on the similarity perception graph transfer learning mechanism. is a dataset constructed from the target blockchain, and the goal is to minimize the similarity-aware loss function on the dataset

[0083]

[0084] Where N T is the number of query-comparison pairs in the target blockchain dataset.

[0085] According to a second aspect of the present invention, an adaptive cross-blockchain anonymous account supervision device based on spatiotemporal graph learning is provided, comprising:

[0086] A construction module is used to construct a spatiotemporal behavior graph model of anonymous accounts based on the temporal dynamics and spatial heterogeneity of blockchain anonymous account behaviors;

[0087] An extraction module is used to extract the spatiotemporal behavior features in a single anonymous account spatiotemporal behavior graph based on the anonymous account spatiotemporal behavior graph model and using a designed temporal heterogeneous graph learning mechanism for a single anonymous account spatiotemporal behavior graph;

[0088] A measurement module is used to measure the spatiotemporal behavior characteristics between different anonymous accounts based on the anonymous account spatiotemporal behavior graph model and multiple anonymous account spatiotemporal behavior graphs using a designed temporal heterogeneous graph twin network to obtain the spatiotemporal behavior similarity between different anonymous accounts;

[0089] The tuning module is used for different blockchain environments. It uses the designed similarity perception graph transfer learning mechanism to adaptively tune the measurement of the spatiotemporal behavior similarity between different anonymous accounts in the cross-blockchain environment, and realizes adaptive cross-blockchain anonymous account identity association.

[0090] According to a third aspect of the present invention, a device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for adaptive cross-blockchain anonymous account supervision based on spatiotemporal graph learning is implemented.

[0091] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for adaptive cross-blockchain anonymous account supervision based on spatiotemporal graph learning is implemented.

[0092] Compared with the prior art, the present invention has at least the following beneficial effects:

[0093] (1) This paper proposes an anonymous account spatiotemporal behavior graph model, which not only effectively captures the temporal dynamics and spatial heterogeneity of on-chain account transaction behaviors, but also designs a heuristic rule-based graph pruning algorithm to denoise irrelevant transaction subgraphs, accurately modeling the real-time and diversified transaction behaviors of blockchain anonymous accounts, while reducing the space overhead of graph learning training.

[0094] (2) The present invention proposes a temporal heterogeneous graph learning mechanism in a blockchain environment. For a single anonymous account's spatiotemporal behavior graph, the present invention fully mines its spatiotemporal behavior characteristics from the perspective of temporal heterogeneous edge signals, point signals, and subgraph signals, thereby improving the ability to express the characteristic behavior of blockchain anonymous account transactions.

[0095] (3) The present invention proposes a twin network architecture for time-series heterogeneous graphs, constructs a graph-level twin network for the spatiotemporal behavior graphs of multiple anonymous accounts, and realizes the parallel measurement of the spatiotemporal behavior similarities between anonymous accounts on different chains in the same Euclidean space, thereby reducing the time overhead of learning and training time-series heterogeneous graphs and improving the efficiency of large-scale graph-level blockchain anonymous account behavior analysis.

[0096] (4) The present invention proposes a similarity-aware graph transfer learning mechanism. By pre-training the model in a single blockchain network and adjusting the model similarity metric in real time in a cross-blockchain environment, the problem of shortage of labeled data in existing models for graph learning in cross-blockchain scenarios is alleviated. Adaptive and efficient anonymous account identity association is achieved in the cross-blockchain ecosystem, thereby completing effective supervision of criminal anonymous sub-accounts.

[0097] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] In order to more clearly illustrate the technical solutions in the specific implementation modes of the present invention, the drawings required for use in the description of the specific implementation modes will be briefly introduced below. Obviously, the drawings described below are some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0099] Figure 1 This is a flow chart of an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning of the present invention;

[0100] Figure 2 This is a flow chart of an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning according to an embodiment of the present invention;

[0101] Figure 3 This is a model diagram of the spatiotemporal behavior graph of a native account according to an embodiment of the present invention;

[0102] Figure 4 A design diagram of a temporal heterogeneous edge aggregation method in a temporal heterogeneous graph learning mechanism according to an embodiment of the present invention;

[0103] Figure 5 It is a design diagram of a method for aggregating time series heterogeneous points in a time series heterogeneous graph learning mechanism according to an embodiment of the present invention;

[0104] Figure 6 This is a twin network model diagram of a time-series heterogeneous graph according to an embodiment of the present invention. DETAILED DESCRIPTION

[0105] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0106] As a specific embodiment of the present invention, Figure 1 and Figure 2 As shown, the present invention provides an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning, which specifically includes the following steps:

[0107] Step 1: Aiming at the temporal dynamics and spatial heterogeneity of the anonymous account behavior in the blockchain, a spatiotemporal behavior graph model of anonymous accounts is constructed to accurately model the spatiotemporal behavior of accounts on the chain.

[0108] In one achievable manner, the temporal dynamics and spatial heterogeneity of the anonymous account behavior of the blockchain are used to construct an anonymous account spatiotemporal behavior graph model, which specifically includes:

[0109] Step 1.1: Construct a native account spatiotemporal behavior graph model based on the temporal dynamics of on-chain account transaction behaviors and the heterogeneity of transaction types and transaction accounts.

[0110] In this implementation, the native account spatiotemporal behavior graph model is specifically:

[0111]

[0112] In the formula, It represents the self-blockchain transaction network centered on the anonymous account a on the chain to be supervised, that is, the native account spatiotemporal behavior graph model of the anonymous account a; Representing the Self-Blockchain Transaction Network The depth of breadth-first search traversal; V represents the set of on-chain address nodes traversed in breadth-first search; E represents the set of all on-chain transactions traversed in breadth-first search; Indicates the address type from the on-chain address node set V to the on-chain address type ε is the real external owner account, b is the blockchain robot account, and s is the blockchain smart contract account; Indicates the transaction type from the on-chain transaction set E to the on-chain transaction type ∈ is an external transaction, c is a contract transaction, ι is an internal transaction, and τ is a token transaction; Γ:E→Ω represents the time from the on-chain transaction set E to the on-chain transaction The mapping of ξ s Indicates the transaction start time. Indicates the duration of the transaction.

[0113] More specifically, the native account spatiotemporal behavior graph model is as follows Figure 3 As shown in the figure, its temporal dynamics is reflected in the native account spatiotemporal behavior graph. In the chain to be supervised, anonymous accounts continuously generate new transactions, forming new transaction nodes and transaction edges, which leads to the continuous evolution of the transaction behavior of anonymous accounts on the chain to be supervised. Its spatial heterogeneity is reflected in heterogeneous points (multiple types of chain addresses) and heterogeneous edges (multiple types of chain transactions), mapping Used to model the heterogeneity of on-chain address types in the native account spatiotemporal behavior graph. This mapping is from the address set Mapping to address type The real external owner account ε, blockchain robot account b and blockchain smart contract account s are defined as:

[0114] Definition 1: A real external owner account is an on-chain account controlled by a real person off-chain through a corresponding private key. The real person off-chain freely decides when, how and with whom to trade.

[0115] Definition 2: A blockchain robot account is an on-chain account that uses a preset script to control its corresponding private key to complete automated transaction control. It often plays the role of an automated trading robot in the blockchain ecosystem.

[0116] Definition 3: A blockchain smart contract account is a self-executing public script program that does not hold a private key and can automatically execute its preset script functions in the blockchain virtual machine. It is controlled by the smart contract code and its associated data storage.

[0117] Mapping Used to model the heterogeneity of on-chain transaction types in the native account spatiotemporal behavior graph, this mapping maps the transaction set ε to its type Among them, external transactions ∈, contract transactions c, internal transactions ι and token transactions τ are defined as:

[0118] Definition 4: External transactions are initiated and received by non-blockchain smart contract accounts and are used to transfer cryptocurrencies between non-blockchain smart contract accounts;

[0119] Definition 5: Contract transaction, initiated by a non-blockchain smart contract account and received by a blockchain smart contract account, is used to request blockchain smart contract services;

[0120] Definition 6: Internal transactions are initiated by blockchain smart contract accounts and received by non-blockchain smart contract accounts, and are used to withdraw cryptocurrencies hosted in blockchain smart contract services;

[0121] Definition 7: Token transaction: the transaction object is blockchain tokens, which are used to transfer token assets between blockchain accounts.

[0122] Step 1.2: De-noise the native account spatiotemporal behavior graph model based on a heuristic graph pruning algorithm to obtain the anonymous account spatiotemporal behavior graph model.

[0123] In this implementation, the native account spatiotemporal behavior graph model is denoised based on the heuristic graph pruning algorithm, as follows:

[0124] First, identify the set of overloaded nodes in the native account spatiotemporal behavior graph model:

[0125]

[0126] Where Ψ(i) represents the on-chain address type of node i; D + (i) represents the in-degree of node i; D -(i) represents the out-degree of node i; Indicates the tolerance value of overloaded nodes.

[0127] Secondly, classify the set of stale node neighbors under the branch subgraph connected to the overloaded node:

[0128]

[0129] In the formula, Represents the spatiotemporal behavior graph model of the overloaded node m in the native account The neighbor nodes found by the first-order breadth-first search; Indicates the earliest on-chain transaction among neighbor nodes. neighbor nodes; is the denoising coefficient.

[0130] Finally, the branch subgraphs connected to the neighbors of the corrupted nodes are pruned to complete the denoising of the native account spatiotemporal behavior graph model and obtain the anonymous account spatiotemporal behavior graph model.

[0131] Step 2: Based on the anonymous account spatiotemporal behavior graph model, for a single anonymous account spatiotemporal behavior graph, the designed temporal heterogeneous graph learning mechanism is used to extract the spatiotemporal behavior features in the spatiotemporal behavior graph of the single anonymous account.

[0132] As a preferred implementation, the temporal heterogeneous graph learning mechanism of the design is specifically as follows:

[0133] Step 2.1: Design a heterogeneous temporal edge aggregation method for the edge signals in the anonymous account spatiotemporal behavior graph model, and calculate the initial node features of each graph node.

[0134] Specifically, if Figure 4 As shown, in this embodiment, a heterogeneous time-series edge aggregation method is designed for edge signals in an anonymous account spatiotemporal behavior graph model, specifically including:

[0135] Step 2.1.1: Based on the anonymous account spatiotemporal behavior graph model, perform attention-based aggregation on the adjacent edges of each edge type, introduce the transaction information of the adjacent edges into the edge feature aggregation process, and obtain the intra-class edge aggregation, specifically:

[0136] For a given node u in the anonymous account spatiotemporal behavior graph model, Indicates a given edge type The set of adjacent edges within u, each edge Attention score The calculation is as follows:

[0137]

[0138] In the formula, is the transposed matrix of attention parameters; f i Is the edge type The vectorized transaction information of edge i in the figure. For edge i, the nonlinear activation function Relu is used to calculate the attention score; for the attention score of edge i Further normalized by the Softmax function, the importance weight of edge i based on the attention mechanism The calculation is as follows:

[0139]

[0140] Based on the adjacent edge set The attention-based importance weights are applied to aggregate the information of adjacent edges. The node u is in the edge type Edge aggregation within The calculation is as follows:

[0141]

[0142]

[0143] In the formula, Represents the connection operation; Relu represents the Relu nonlinear activation function; H is the number of attention heads used in the multi-head attention method.

[0144] Step 2.1.2: Based on the intra-class edge aggregation, a long short-term memory network is used to fuse heterogeneous intra-class edge aggregations in chronological order to generate initial node features in the anonymous account spatiotemporal behavior graph model, specifically:

[0145] Aggregate the edges within each edge type class As time series input, this implementation method is for each Assigning timestamps This timestamp is the time at which node u is in the neighborhood With edge type The latest interaction time is calculated as follows:

[0146]

[0147] Aggregate by edge Timestamp Sort the edge aggregation in ascending order and calculate the initial node features of each node u as follows:

[0148]

[0149]

[0150] q u=LSTM([Sequence u ])

[0151] In the formula, LSTM is a long short-term memory network; Sort + Indicates that the time series is sorted in ascending order.

[0152] Step 2.2: Using the initial node features obtained by the heterogeneous time series edge aggregation method as the initial features for point aggregation, a heterogeneous time series point aggregation method is designed for the point signals within the anonymous account spatiotemporal behavior graph model, and the node embedding of each graph node is calculated.

[0153] Specifically, if Figure 5 As shown, in this embodiment, a heterogeneous time series point aggregation method is designed for the internal point signal of the anonymous account spatiotemporal behavior graph model, specifically including:

[0154] Step 2.2.1, using the node features generated by heterogeneous temporal edge aggregation as the initial features of point aggregation, design a temporal importance mechanism to generate weights for neighbor nodes, which can effectively evaluate the impact of neighbor nodes on the target node;

[0155] In this implementation, the timing importance mechanism is as follows:

[0156] For a given node x in the anonymous account spatiotemporal behavior graph model, Node Type The set of adjacent nodes in the node Time attribute From the adjacent side Latest timestamp The calculation is as follows:

[0157]

[0158] According to time attribute Generate the temporal importance weight of node u The calculation is as follows:

[0159]

[0160] Step 2.2.2: Based on the temporal importance weight, a graph isomorphic network is used to perform point aggregation operations within a single node type, and the intra-class point aggregation is calculated, as follows:

[0161] According to the timing importance weight Generate Node Type Intra-class point aggregation of middle node x The calculation is as follows:

[0162]

[0163] In the formula, MLP (k) Indicates a multilayer perceptron with k layers; Is the node x in the node type The k-th layer embedding result in ; is the input of the kth layer, and Neighbors The temporal importance weight of , indicating that the influence of neighbors on node x changes over time.

[0164] Step 2.2.3: Based on the intra-class point aggregation, the graph attention mechanism is used to fuse heterogeneous types of intra-class point aggregation to generate inter-class point aggregation, and complete the information aggregation of the point signal in the anonymous account spatiotemporal behavior graph model, i.e., node embedding, as follows:

[0165] According to the intra-class point aggregation, node x is embedded in the node of type β Attention score The calculation is as follows:

[0166]

[0167] Where LeakyRelu is the zero negative correction activation function; is the transposed attention parameter;

[0168] According to the node embedding of the node x in type β The attention score of node x, the aggregation of inter-class points, is the node embedding, which is calculated as follows:

[0169]

[0170] In the formula, ρ x is the between-class point aggregation of node x.

[0171] Step 2.3: Based on the node embedding of each graph node, a graph pooling method is designed for the subgraph signal in the anonymous account spatiotemporal behavior graph model, and the graph embedding of the anonymous account spatiotemporal behavior graph model is calculated.

[0172] As a preferred implementation, this implementation designs an attention-based pooling mechanism, which assigns attention weights to the embedding of each node and combines them to form the final graph embedding, so that nodes that are more relevant to the spatiotemporal pattern of the anonymous account α to be supervised obtain higher weights. The graph embedding of the anonymous account spatiotemporal behavior graph model is defined as follows:

[0173]

[0174] In the formula, ρ i is the heterogeneous temporal node embedding of node i; a iis the attention weight of node i, a i The specific calculation is as follows:

[0175]

[0176] In the formula, q T is the transpose of the learnable attention vector; W is the learnable weight matrix; V is the node set of the anonymous account spatiotemporal graph model.

[0177] Step 3: Based on the anonymous account spatiotemporal behavior graph model, for multiple anonymous account spatiotemporal behavior graphs, the designed temporal heterogeneous graph twin network is used to measure the spatiotemporal behavior characteristics between different anonymous accounts to obtain the spatiotemporal behavior similarity between different anonymous accounts.

[0178] As a preferred embodiment, Figure 6 As shown, the time-series heterogeneous graph twin network specifically includes:

[0179] Step 3.1: Build a graph-level twin network for multiple anonymous account spatiotemporal behavior graphs to implement large-scale graph-level anonymous account behavior analysis in parallel, as follows:

[0180] For a given anonymous account spatiotemporal behavior graph j, graph embedding f j The calculation is as follows:

[0181]

[0182] Where Pool is the graph pooling method; V j is the node set of the anonymous account spatiotemporal behavior graph j; It is the node embedding of node v in the last layer after L layers of heterogeneous time series edge aggregation and heterogeneous time series point aggregation:

[0183]

[0184] In the formula, HTAGG n It is a heterogeneous time-series point aggregation method; is the set of neighbor nodes of node v; is the aggregate edge embedding of node v at level l+1:

[0185]

[0186] In the formula, HTAGG e It is a heterogeneous time series edge aggregation method; is the edge embedding from node u to node v at layer l;

[0187] Initialized by a random vector.

[0188] In this embodiment, each time-series heterogeneous graph twin network includes graph twin sub-network A and graph twin sub-network B. The network weights are shared between the twin sub-networks, and can simultaneously support the parallel generation of graph embeddings of two anonymous account spatiotemporal behavior graphs for similarity comparison.

[0189] Step 3.2: Use the designed time-series heterogeneous graph twin network to measure the spatiotemporal behavior characteristics between different anonymous accounts, as follows:

[0190] According to the time-space behavior graph of the anonymous account to be compared and The graph embedding f q and f p , use Euclidean distance to measure the similarity of spatiotemporal behaviors between different anonymous accounts:

[0191]

[0192] Where d is the graph embedding f q and f p The vector dimension of .

[0193] Step 4: For different blockchain environments, the designed similarity perception graph-based transfer learning mechanism is used to adaptively tune the measurement of the spatiotemporal behavior similarity between different anonymous accounts in the cross-blockchain environment, realize adaptive cross-blockchain anonymous account identity association, and then realize anonymous account supervision.

[0194] As a preferred implementation, the similarity perception graph-based transfer learning mechanism specifically includes:

[0195] Step 4.1: Basic model pre-training, as follows:

[0196] For a single blockchain scenario, the anonymous account spatiotemporal behavior graph model is pre-trained to obtain a pre-trained basic model. is a dataset on the source blockchain, and is the spatiotemporal behavior graph of the anonymous account for the i-th query-comparison sample pair, Represents the self nodes in the two graphs, i.e., whether the comparison account and query account are linked to the same off-chain identity, N S Represents the number of query-comparison pairs in the dataset constructed from the source blockchain; the purpose of pre-training is to minimize the number of samples in the source dataset for the temporal heterogeneous graph twin network. The loss function on

[0197]

[0198] In the formula, and They are and Graph embedding of is the Euclidean distance between the two graph embeddings, and the parameters θ of the temporal heterogeneous graph twin network are updated by gradient descent:

[0199]

[0200] Where η is the learning rate.

[0201] Step 4.2: Perform cross-blockchain model optimization based on similarity perception graph transfer learning mechanism, as follows;

[0202] According to the pre-trained basic model, cross-blockchain model tuning is performed based on the similarity perception graph transfer learning mechanism. is a dataset constructed from the target blockchain, and the goal is to minimize the similarity-aware loss function on the dataset

[0203]

[0204] Where N T is the number of query-comparison pairs in the target blockchain dataset.

[0205] In another embodiment of the present invention, an adaptive cross-blockchain anonymous account supervision device based on spatiotemporal graph learning is provided, which is used to implement an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning in the above embodiment, specifically including:

[0206] A construction module is used to construct a spatiotemporal behavior graph model of anonymous accounts based on the temporal dynamics and spatial heterogeneity of blockchain anonymous account behaviors;

[0207] An extraction module is used to extract the spatiotemporal behavior features in a single anonymous account spatiotemporal behavior graph based on the anonymous account spatiotemporal behavior graph model and using a designed temporal heterogeneous graph learning mechanism for a single anonymous account spatiotemporal behavior graph;

[0208] A measurement module is used to measure the spatiotemporal behavior characteristics between different anonymous accounts based on the anonymous account spatiotemporal behavior graph model and multiple anonymous account spatiotemporal behavior graphs using a designed temporal heterogeneous graph twin network to obtain the spatiotemporal behavior similarity between different anonymous accounts;

[0209] The tuning module is used for different blockchain environments. It uses the designed similarity perception graph transfer learning mechanism to adaptively tune the measurement of the spatiotemporal behavior similarity between different anonymous accounts in the cross-blockchain environment, and realizes adaptive cross-blockchain anonymous account identity association.

[0210] All relevant contents of each step involved in the embodiment of the aforementioned adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning can be cited in the functional description of the functional module corresponding to the adaptive cross-blockchain anonymous account supervision device based on spatiotemporal graph learning in the embodiment of the present invention, which will not be repeated here. The division of modules in the embodiment of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention can be integrated in a processor, or it can exist physically separately, or two or more modules can be integrated in one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0211] In another embodiment of the present invention, a computer device is provided, the computer device including a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning.

[0212] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in a computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the above-mentioned embodiment regarding an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning.

[0213] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0214] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0215] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0216] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0217] In the present invention, the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0218] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning, characterized in that: include: Aiming at the temporal dynamics and spatial heterogeneity of anonymous account behavior in blockchain, a spatiotemporal behavior graph model of anonymous accounts is constructed; Based on the anonymous account spatiotemporal behavior graph model, for a single anonymous account spatiotemporal behavior graph, the designed temporal heterogeneous graph learning mechanism is used to extract the spatiotemporal behavior features in the spatiotemporal behavior graph of the single anonymous account; Based on the anonymous account spatiotemporal behavior graph model, for multiple anonymous account spatiotemporal behavior graphs, the designed temporal heterogeneous graph twin network is used to measure the spatiotemporal behavior characteristics between different anonymous accounts to obtain the spatiotemporal behavior similarity between different anonymous accounts; Aiming at different blockchain environments, the designed similarity perception graph-based transfer learning mechanism is used to adaptively tune the measurement of the spatiotemporal behavior similarity between different anonymous accounts in the cross-blockchain environment, so as to realize adaptive cross-blockchain anonymous account identity association.

2. According to claim 1, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The temporal dynamics and spatial heterogeneity of the anonymous account behavior in the blockchain are used to construct a spatiotemporal behavior graph model of anonymous accounts, specifically: Based on the temporal dynamics of on-chain account transaction behaviors, as well as the heterogeneity of transaction types and transaction accounts, a native account spatiotemporal behavior graph model is constructed; The native account spatiotemporal behavior graph model is denoised based on a heuristic graph pruning algorithm to obtain the anonymous account spatiotemporal behavior graph model.

3. According to claim 2, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The native account spatiotemporal behavior graph model is specifically: In the formula, It represents the self-blockchain transaction network centered on the anonymous account a on the chain to be supervised, that is, the native account spatiotemporal behavior graph model of the anonymous account a; Representing the Self-Blockchain Transaction Network The depth of breadth-first search traversal; V represents the set of on-chain address nodes traversed in breadth-first search; E represents the set of all on-chain transactions traversed in breadth-first search; Indicates the address type from the on-chain address node set V to the on-chain address type ε is the real external owner account, b is the blockchain robot account, and s is the blockchain smart contract account; Indicates the transaction type from the on-chain transaction set E to the on-chain transaction type ∈ is an external transaction, c is a contract transaction, ι is an internal transaction, and τ is a token transaction; Γ:E→Ω represents the time from the on-chain transaction set E to the on-chain transaction The mapping of ξ s Indicates the transaction start time. Indicates the duration of the transaction; The heuristic graph pruning algorithm is used to denoise the native account spatiotemporal behavior graph model, specifically: Identify an overloaded node set in the native account spatiotemporal behavior graph model, where the overloaded node set in the native account spatiotemporal behavior graph model is: Where Ψ(i) represents the on-chain address type of node i; D + (i) represents the in-degree of node i; D _ (i) represents the out-degree of node i; Indicates the tolerance value of overloaded nodes; The rotten node neighbor set under the branch subgraph connected to the overloaded node is classified, and the rotten node neighbor set under the branch subgraph connected to the overloaded node is: In the formula, Represents the spatiotemporal behavior graph model of the overloaded node m in the native account The neighbor nodes found by the first-order breadth-first search; Indicates the earliest on-chain transaction among neighbor nodes. neighbor nodes; is the denoising coefficient; The branch subgraph connected to the neighbors of the corrupted nodes is pruned to complete the denoising of the native account spatiotemporal behavior graph model, and obtain the anonymous account spatiotemporal behavior graph model.

4. According to claim 3, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The temporal heterogeneous graph learning mechanism designed is specifically as follows: A heterogeneous temporal edge aggregation method is designed for edge signals in the spatiotemporal behavior graph model of anonymous accounts, and the initial node features of each graph node are calculated. According to the initial node features obtained by the heterogeneous time-series edge aggregation method as the initial features of point aggregation, a heterogeneous time-series point aggregation method is designed for the point signals within the anonymous account spatiotemporal behavior graph model, and the node embedding of each graph node is calculated; According to the node embedding of each graph node, a graph pooling method is designed for the subgraph signals in the anonymous account spatiotemporal behavior graph model, and the graph embedding of the anonymous account spatiotemporal behavior graph model is calculated.

5. According to claim 4, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The heterogeneous time series edge aggregation method for edge signal design in the anonymous account spatiotemporal behavior graph model is specifically as follows: For a given node u in the anonymous account spatiotemporal behavior graph model, Indicates a given edge type The set of adjacent edges within u, each edge Attention score The calculation is as follows: In the formula, is the transposed matrix of attention parameters; f i Is the edge type The vectorized transaction information of edge i. For edge i, the nonlinear activation function Relu is used to calculate the attention score; Attention score for edge i Further normalized by the Softmax function, the importance weight of edge i based on the attention mechanism The calculation is as follows: Based on the adjacent edge set The attention-based importance weights are applied to aggregate the information of adjacent edges. The node u is in the edge type Edge aggregation within The calculation is as follows: In the formula, Indicates a connection operation; Relu represents the Relu nonlinear activation function; H is the number of attention heads used in the multi-head attention method; for each edge aggregation The timestamp is defined as The latest interaction in is calculated as follows: Aggregate by edge Timestamp The edge aggregation is sorted in ascending order, and the initial node feature of each node u is calculated as follows: q u =LSTM([Sequence u ]) In the formula, LSTM is a long short-term memory network; Sort + Indicates that the time series is sorted in ascending order.

6. According to claim 5, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The heterogeneous time-series point aggregation method for the internal point signal design of the anonymous account spatiotemporal behavior graph model is specifically: For a given node x in the anonymous account spatiotemporal behavior graph model, Node Type The set of adjacent nodes in the node Time attribute From the adjacent side Latest timestamp The calculation is as follows: According to time attribute Generate the temporal importance weight of node u The calculation is as follows: According to the timing importance weight Generate Node Type Intra-class point aggregation of middle node x The calculation is as follows: In the formula, MLP (k) Indicates a multilayer perceptron with k layers; Is the node x in the node type The k-th layer embedding result in ; is the input of the kth layer, and θ l Neighbors The temporal importance weight of , indicating that the influence of neighbors on node x changes over time; According to the intra-class point aggregation, node x is embedded in the node of type β Attention score The calculation is as follows: Where LeakyRelu is the zero negative correction activation function; is the transposed attention parameter; According to the node embedding of the node x in type β The attention score of node x, the aggregation of inter-class points, is the node embedding, which is calculated as follows: In the formula, ρ x is the between-class point aggregation of node x.

7. According to claim 6, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The graph pooling method for subgraph signals in the spatiotemporal behavior graph model for anonymous accounts is specifically as follows: In the formula, ρ i is the heterogeneous temporal node embedding of node i; a i is the attention weight of node i, a i The specific calculation is as follows: In the formula, is the transpose of the learnable attention vector; W is the learnable weight matrix; V is the node set of the anonymous account spatiotemporal graph model.

8. According to claim 7, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The designed time-series heterogeneous graph twin network is specifically: Construct a graph-level twin network for multiple anonymous account spatiotemporal behavior graphs. For a given anonymous account spatiotemporal behavior graph j, the graph embedding f j The calculation is as follows: Where Pool is the graph pooling method; V j is the node set of the anonymous account spatiotemporal behavior graph j; It is the node embedding of node v in the last layer after L layers of heterogeneous time series edge aggregation and heterogeneous time series point aggregation: Where, HTAGG n It is a heterogeneous time-series point aggregation method; is the set of neighbor nodes of node v; is the aggregate edge embedding of node v at level l+1: Where, HTAGG e It is a heterogeneous time series edge aggregation method; is the edge embedding from node u to node v at layer l; Initialize with a random vector; The designed time-series heterogeneous graph twin network is used to measure the spatiotemporal behavior characteristics between different anonymous accounts, specifically: According to the time-space behavior graph of the anonymous account to be compared and The graph embedding f q and f p , use Euclidean distance to measure the similarity of spatiotemporal behaviors between different anonymous accounts: Where d is the graph embedding f q and f p The vector dimension of .

9. According to claim 8, an adaptive cross-blockchain anonymous account supervision method based on spatiotemporal graph learning is characterized in that: The method of adaptively tuning the measurement of spatiotemporal behavior similarity between different anonymous accounts in a cross-blockchain environment by using the designed similarity perception graph transfer learning mechanism is as follows: For a single blockchain scenario, the anonymous account spatiotemporal behavior graph model is pre-trained to obtain a pre-trained basic model. is a dataset on the source blockchain, and is the spatiotemporal behavior graph of the anonymous account for the i-th query-comparison sample pair, Represents the self nodes in the two graphs, i.e., whether the comparison account and query account are linked to the same off-chain identity, N S Represents the number of query-comparison pairs in the dataset constructed from the source blockchain; the purpose of pre-training is to minimize the number of samples in the source dataset for the temporal heterogeneous graph twin network. The loss function on In the formula, and They are and Graph embedding of is the Euclidean distance between the two graph embeddings, and the parameters θ of the temporal heterogeneous graph twin network are updated by gradient descent: Where η is the learning rate; According to the pre-trained basic model, cross-blockchain model tuning is performed based on the similarity perception graph transfer learning mechanism. is a dataset constructed from the target blockchain, and the goal is to minimize the similarity-aware loss function on the dataset Where N T is the number of query-comparison pairs in the target blockchain dataset.

10. An adaptive cross-blockchain anonymous account supervision device based on spatiotemporal graph learning, characterized in that: include: A construction module is used to construct a spatiotemporal behavior graph model of anonymous accounts based on the temporal dynamics and spatial heterogeneity of blockchain anonymous account behaviors; An extraction module is used to extract the spatiotemporal behavior features in a single anonymous account spatiotemporal behavior graph based on the anonymous account spatiotemporal behavior graph model and using a designed temporal heterogeneous graph learning mechanism for a single anonymous account spatiotemporal behavior graph; A measurement module is used to measure the spatiotemporal behavior characteristics between different anonymous accounts based on the anonymous account spatiotemporal behavior graph model and multiple anonymous account spatiotemporal behavior graphs using a designed temporal heterogeneous graph twin network to obtain the spatiotemporal behavior similarity between different anonymous accounts; The tuning module is used for different blockchain environments. It uses the designed similarity perception graph transfer learning mechanism to adaptively tune the measurement of the spatiotemporal behavior similarity between different anonymous accounts in the cross-blockchain environment, and realizes adaptive cross-blockchain anonymous account identity association.

Images