User identity identification method, system, device and storage medium

CN118551307BActive Publication Date: 2024-11-12GUANGZHOU UNIVERSITY
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410633064.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-21
Publication Date
2024-11-12
Estimated Expiration
2044-05-21

AI Technical Summary

Technical Problem

[0003]虽然手动特征工程在账户识别中发挥着重要的作用,但是手工特征高度依赖于特征工程师的主观判断与预先设定的假设,难以充分挖掘和表达区块链数据深处的复杂模式,造成特征选取的局限性和表达效能不足,从而导致分析模型对用户身份识别结果的准确性比较低

Benefits of technology

[0038] The beneficial effects of this application are: different from the existing technology, this application extracts the first-order subgraph of the target user node and first uses the local self-attention mechanism to calculate the attention weight of a certain local neighborhood such as the first-order subgraph. , determine the subgraph vector representation corresponding to each first-order subgraph. This method can more accurately capture the local behavioral characteristics of each user in the transaction network, that is, the local self-attention mechanism can be used to retain the local subgraph. Deep feature extraction based on structural information can more comprehensively reflect the user's actual behavior pattern and improve the expression ability of features and the generalization ability of the model compared to relying on predefined features.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118551307B_ABST
    Figure CN118551307B_ABST
Patent Text Reader

Abstract

This application discloses a user identification method, system, device, and storage medium. The method includes: acquiring an original graph to be processed, wherein the original graph includes multiple target user nodes; extracting a first-order subgraph corresponding to each target user node from the original graph; determining the subgraph vector representation corresponding to each first-order subgraph based on a local self-attention mechanism; updating the vector representation of each subgraph using a global self-attention mechanism to obtain the target node representation corresponding to each first-order subgraph; predicting the target node representation using a pre-trained multilayer perceptron to obtain the prediction result output by the multilayer perceptron; and determining the target user's identity category information based on the prediction result. Through the above method, this application can improve the accuracy of user identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, system and storage medium for user identity recognition based on local information and global information. Background Art

[0002] In recent years, in the field of blockchain analysis and account deanonymization, researchers have adopted a variety of methods and technologies to achieve the inference and identification of account identities. For example, manual feature engineering that relies on expert prior knowledge can infer account identities by designing specific statistical features (including transaction rates, timestamp-related features, etc.) or behavioral pattern features.

[0003] Although manual feature engineering plays an important role in account identification, manual features are highly dependent on the subjective judgment and pre-set assumptions of feature engineers, making it difficult to fully explore and express the complex patterns deep within blockchain data, resulting in limitations in feature selection and insufficient expression efficiency, which in turn leads to relatively low accuracy of user identity recognition results by the analysis model. Summary of the invention

[0004] The present application provides a user identity recognition method, device, system and storage medium based on local information and global information, which can improve the accuracy of user identity recognition results.

[0005] In a first aspect, the present application provides a method for identifying a user based on local information and global information, the method comprising:

[0006] Acquire an original graph to be processed, wherein the original graph includes a plurality of target user nodes;

[0007] Extract the first-order subgraph corresponding to each target user node from the original graph;

[0008] Determine the sub-graph vector representation corresponding to each first-order sub-graph based on the local self-attention mechanism;

[0009] The global self-attention mechanism is used to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph;

[0010] Use the trained multi-layer perceptron to predict the target node representation and obtain the prediction result output by the multi-layer perceptron;

[0011] The identity category information of the target user is determined based on the prediction results.

[0012] Its further technical solution is that the first-order subgraph includes a plurality of first-order user nodes corresponding to the target user;

[0013] The subgraph vector representation corresponding to each first-order subgraph is determined based on the local self-attention mechanism, including:

[0014] The local self-attention mechanism is used to preliminarily update the initial vector representation of each first-order user node to obtain the intermediate vector representation corresponding to each first-order user node;

[0015] Each intermediate vector representation is pooled to obtain the sub-graph vector representation corresponding to the first-order sub-graph.

[0016] A further technical solution is to use a local self-attention mechanism to preliminarily update the initial vector representation of each first-order user node to obtain an intermediate vector representation corresponding to each first-order user node, including:

[0017] Obtain the user nodes to be updated from the first-order subgraph in sequence;

[0018] Determine in sequence the related first-order user nodes corresponding to the user nodes to be updated;

[0019] The weight ratio between the user node to be updated and the related first-order user nodes is determined in sequence using the local self-attention mechanism;

[0020] Based on the weight ratios, the intermediate vector representations corresponding to the user nodes to be updated are obtained in turn.

[0021] A further technical solution is to perform pooling on each intermediate vector representation to obtain a sub-graph vector representation corresponding to a first-order sub-graph, including:

[0022] Each intermediate vector representation is average-pooled to obtain the sub-graph vector representation corresponding to the first-order sub-graph.

[0023] A further technical solution thereof is that the intermediate vector representation includes a first dimension value and a second dimension value, and the sub-graph vector representation also includes a first dimension value and a second dimension value;

[0024] Each intermediate vector representation is averaged and pooled to obtain the sub-graph vector representation corresponding to the first-order sub-graph, including:

[0025] The first dimension values ​​in all the intermediate vector representations are averaged to obtain the first dimension value of the sub-graph vector representation;

[0026] The second dimension values ​​in all the intermediate vector representations are averaged to obtain the second dimension value of the sub-graph vector representation;

[0027] According to the first dimension value and the second dimension value of the sub-graph vector representation, the sub-graph vector representation corresponding to the first-order sub-graph is obtained.

[0028] Its further technical solution is to use the global self-attention mechanism to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph, including:

[0029] The global self-attention mechanism is used to sequentially calculate the relationship weights between each sub-graph vector representation and all other sub-graph vector representations;

[0030] Based on the relationship weights, the target node representations corresponding to each first-order subgraph are obtained in turn.

[0031] A further technical solution is that the prediction result includes a probability value of the identity category, and the probability value is between 0 and 1, and the identity category information of the target user includes a normal user and an abnormal user;

[0032] Determine the target user's identity category information based on the prediction results, including:

[0033] If the probability value is greater than or equal to the preset value, the target user is an abnormal user;

[0034] If the probability value is less than the preset value, the target user is a normal user.

[0035] In a second aspect, the present application provides a user identity recognition system based on local information and global information, and the user identity recognition system based on local information and global information includes a unit for executing any of the above methods.

[0036] In a third aspect, the present application provides a computer device, the computer device comprising a memory and a processor, the memory being used to store program data, and the processor being used to execute the program data to implement the steps of any of the above methods.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium, which is used to store a computer program, and when the computer program is executed by a processor, it is used to implement the steps of any of the above methods.

[0038] The beneficial effect of the present application is as follows: Different from the prior art, the present application extracts the first-order subgraph of the target user node and first uses the local self-attention mechanism to calculate the attention weight of a local neighborhood such as the first-order subgraph, and determines the subgraph vector representation corresponding to each of the first-order subgraphs. This method can more finely capture the local behavioral characteristics of each user in the transaction network, that is, the local self-attention mechanism can be used to perform deep feature extraction at the local subgraph in a manner that retains structural information. Compared with relying on predefined features, it can more comprehensively reflect the actual behavior patterns of users and improve the expressiveness of features and the generalization ability of the model.

[0039] Secondly, the global self-attention mechanism is used to calculate and update the weights between all user nodes for the sub-graph vector representation that has captured the local structural information, so as to obtain the target node representation corresponding to each of the first-order sub-graphs. Since the global self-attention mechanism can learn the long-distance pairwise dependencies between nodes, it can effectively broaden the receptive field, that is, the perception range of the input data while avoiding over-smoothing or over-compression. Therefore, the present application adopts a combination of local self-attention mechanism and global self-attention mechanism, which can simultaneously capture the structure and feature similarity of user nodes in the sub-graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0041] Figure 1 It is a flowchart of a first embodiment of a method for user identification based on local information and global information provided by the present application;

[0042] Figure 2 It is a schematic diagram of an embodiment of the original diagram provided by this application;

[0043] Figure 3 is a schematic diagram of a first-order root subtree provided by this application;

[0044] Figure 4 is a schematic diagram of a first-order subgraph provided by this application;

[0045] Figure 5 It is a schematic diagram of another embodiment of the original diagram provided by the present application;

[0046] Figure 6 This is a schematic diagram of a first-order subgraph in which the target user provided by the present application is user 2;

[0047] Figure 7 This is a schematic diagram of a first-order subgraph in which the target user provided by the present application is user 6;

[0048] Figure 8 is a schematic diagram of the local self-attention mechanism provided by the present application applied to user 1;

[0049] Fig. 9 is a schematic diagram of the local self-attention mechanism provided by the present application applied to user 2;

[0050] Fig.10 is a schematic diagram of the local self-attention mechanism provided by the present application applied to user 3;

[0051] Fig.11 is a schematic diagram of the local self-attention mechanism provided by the present application applied to user 7;

[0052] Fig.12 It is a structural schematic diagram of an embodiment of a computer-readable storage medium provided by the present application. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some but not all structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0054] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0055] In recent years, in the field of blockchain analysis and account deanonymization, researchers have adopted a variety of methods and technologies to infer and identify account identities. Early methods mainly include the following:

[0056] 1) Focus on address clustering and re-identify users or entities by analyzing transaction patterns and heuristic methods.

[0057] 2) Manual feature engineering.

[0058] Manual feature engineering relies on expert prior knowledge. By designing specific statistical features such as transaction rates, timestamp-related features, or behavioral pattern features, account identity can be inferred. Manual feature engineering plays an important role in account identification.

[0059] 3)Graph analysis method.

[0060] Among them, graph analysis is a method that models and analyzes account identification as a graph classification task. It uses graph topology, graph random walks, and graph embedding techniques to learn account features. Graph analysis methods are widely used in the modeling and analysis of transaction data.

[0061] 4) For emerging public blockchains with smart contract functions, smart contract analysis has become an important method. Smart contract functions can expand the feature space and account identification capabilities. By analyzing features such as call information and contract code style, it is possible to better identify and infer accounts.

[0062] In addition, there are other methods and frameworks for identity recognition, such as graph mining technology, meta-path analysis, and heterogeneous network analysis. These methods can effectively detect abnormal accounts and transactions, providing more options and possibilities for account identification.

[0063] However, although these methods are effective to a certain extent, there are still some limitations and challenges. First, manual feature engineering relies on the prior knowledge of feature designers and cannot fully capture the underlying information in blockchain data, such as transaction patterns, resulting in low utilization and limited expressiveness of features. In addition, due to the technical differences between different blockchain platforms, the reusability of manual features is weak. For example, Ethereum data has different smart contract functions from Bitcoin, which greatly limits the reusability of manual features.

[0064] In addition, existing methods are unable to capture potential transaction patterns in the Ethereum anonymous transaction network and perceive all account information. This means that traditional methods cannot fully understand and analyze transaction patterns in the Ethereum network, and cannot fully perceive all accounts. This limitation may make it difficult to analyze account identities and behaviors, which in turn makes the final recognition results less accurate.

[0065] Graph analysis relies on transaction graphs constructed from large-scale transaction data, which results in high computational and time costs when applying graph random walks or graph neural networks (GNNs). At the same time, as the number of transactions on the blockchain continues to increase, the update frequency of nodes and edges in the transaction graph also increases, which is not conducive to full graph learning.

[0066] Therefore, in order to solve the above technical problems, the present application provides a user identity recognition method based on local information and global information. Please refer to the following embodiments for details.

[0067] See also Figure 1 , Figure 1 This is a flow chart of the first embodiment of the user identity identification method based on local information and global information provided by the present application. The method includes:

[0068] Step 110: Obtain the original image to be processed.

[0069] The original graph in this embodiment is composed of an account relationship interaction graph, based on the Ethereum user relationship, and a lightweight account transaction graph is obtained through interaction merging and feature construction operations. In the lightweight account transaction graph, nodes represent accounts, and edges represent transaction information and contract callbacks.

[0070] The original graph includes multiple target user nodes, for example, 13 target user nodes, namely user 1, user 2, user 3, user 4, user 5, user 6, user 7, user 8, user 9, user 10, user 11, user 12, and user 13.

[0071] Step 120: Extract the first-order subgraph corresponding to each target user node from the original graph.

[0072] Among them, the first-order subgraph is a local subgraph centered on the target user node and extracting neighboring nodes with a distance of 1 from it.

[0073] Step 130: Determine the sub-graph vector representation corresponding to each first-order sub-graph based on the local self-attention mechanism.

[0074] Among them, the local self-attention mechanism usually refers to introducing a restriction in the self-attention mechanism so that the attention weight is calculated only within a local neighborhood. The self-attention mechanism is a mechanism that calculates the correlation between different nodes of the input to determine the importance of each node to other nodes. In this example, the local self-attention mechanism can be used to perform deep feature extraction in a way that preserves structural information at the local subgraph, update the feature vectors of all nodes on the first-order subgraph, and finally obtain the subgraph vector representation corresponding to each first-order subgraph.

[0075] Step 140: Use the global self-attention mechanism to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph.

[0076] Among them, the global self-attention mechanism can learn long-distance pairwise dependencies between nodes, which can effectively broaden the receptive field, that is, the perception range of the input data, while avoiding over-smoothing or over-compression.

[0077] Step 150: Use the trained multilayer perceptron to predict the target node representation and obtain the prediction result output by the multilayer perceptron.

[0078] in,

[0079] Step 160: Determine the identity category information of the target user based on the prediction result.

[0080] The prediction result includes the probability value of the identity category, and the probability value is between 0 and 1. The identity category information of the target user includes normal users and abnormal users;

[0081] Step 160 may include:

[0082] 1) If the probability value is greater than or equal to the preset value, the target user is an abnormal user;

[0083] 2) If the probability value is less than the preset value, the target user is a normal user.

[0084] Specifically, for example, the preset values ​​are 0.5, 0.6, 0.7, etc. Taking the preset value of 0.5 as an example, if the probability value output by the multilayer perceptron is 0.9 which is greater than 0.5, it means that the target user is very likely to be involved in fraudulent behavior, and the target user can be classified as an abnormal user; if the probability value output by the multilayer perceptron is 0.1 which is less than 0.5, it means that the target user's behavior is normal, and the target user can be classified as a normal user.

[0085] On the contrary, if the probability value is greater than or equal to the preset value, the target user is a normal user; if the probability value is less than the preset value, the target user is an abnormal user, which can be determined according to the specific meaning of the set label.

[0086] This embodiment extracts the first-order subgraph of the target user node and first uses the local self-attention mechanism to calculate the attention weight of a local neighborhood such as the first-order subgraph, and determines the subgraph vector representation corresponding to each of the first-order subgraphs. This method can more finely capture the local behavioral characteristics of each user in the transaction network, that is, the local self-attention mechanism can be used to perform deep feature extraction in the local subgraph in a manner that retains structural information. Compared with relying on predefined features, it can more comprehensively reflect the actual behavior patterns of users and improve the expressiveness of features and the generalization ability of the model.

[0087] Secondly, the global self-attention mechanism is used to calculate and update the weights between all user nodes for the sub-graph vector representation that has captured the local structural information, so as to obtain the target node representation corresponding to each of the first-order sub-graphs. Since the global self-attention mechanism can learn the long-distance pairwise dependencies between nodes, it can effectively broaden the receptive field, that is, the perception range of the input data while avoiding over-smoothing or over-compression. Therefore, the present application adopts a combination of local self-attention mechanism and global self-attention mechanism, which can simultaneously capture the structure and feature similarity of user nodes in the sub-graph.

[0088] In the existing related technologies, the GNN model is usually used to iteratively aggregate the features of neighboring nodes to a central node, obtain a node representation that encodes a rooted subtree around the central node, and perform analysis and prediction based on the node representation of the rooted subtree. However, the Ethereum account interaction graph has many circle structures, and the GNN model that encodes the rooted subtree is not conducive to capturing the potential transaction behavior patterns of the account.

[0089] Based on this, the present application provides a method of using a subgraph around a node for analysis and prediction to capture potential transaction behavior patterns.

[0090] For example, assuming the original image is Figure 2 As shown, the user's first-order root subtree can be found in Figure 3 , the user's first-order subgraph can be found in Figure 4 ,in, Figure 3 and Figure 4 All of them are based on the analysis of the target user being User 2.

[0091] Combination Figure 3 and Figure 4 It can be seen that the first-order root subtree only shows the transaction relationship between the target user, user 2, and other users, but does not show the transaction relationship between other users (such as user 3 and user 5).

[0092] The first-order subgraph extracted by this application can represent the transaction relationship between other users (such as user 3 and user 5), that is, it can capture potential transaction behaviors.

[0093] Refer to the flowchart of the second embodiment of the user identity recognition method based on local information and global information provided by this application. The method includes:

[0094] Step 210: Obtain the original image to be processed.

[0095] The original graph includes multiple target user nodes.

[0096] Step 220: Extract the first-order subgraph corresponding to each target user node from the original graph.

[0097] The first-order subgraph includes multiple first-order user nodes corresponding to the target user.

[0098] For example, Figure 5 As shown, the original graph includes 13 target user nodes, namely user 1, user 2, user 3, user 4, user 5, user 6, user 7, user 8, user 9, user 10, user 11, user 12, and user 13.

[0099] Then the first-order subgraphs corresponding to user 2 and user 6 can be expressed as Figure 6 and Figure 7 shown.

[0100] It should be noted that other target user nodes such as user 3, user 4, user 5, etc. also extract first-order subgraphs by referring to this method, and this application does not provide examples here.

[0101] Specifically, Figure 6User 2 in is the target user, and user 1, user 3, and user 7 are the first-order user nodes of user 2.

[0102] For example, Figure 7 User 6 is the target user, and user 1, user 7, and user 10 are the first-order user nodes of user 6.

[0103] Step 230: Use the local self-attention mechanism to preliminarily update the initial vector representation of each first-order user node to obtain the intermediate vector representation corresponding to each first-order user node.

[0104] Wherein, step 230 includes the following:

[0105] Step 231: Obtain the user nodes to be updated from the first-order subgraph in sequence.

[0106] for example Figure 6 In the example, user 1, user 2, user 3, and user 7 can be taken in turn as user nodes to be updated.

[0107] Similarly, you can Figure 7 User 1, user 6, user 7, and user 10 are sequentially used as user nodes to be updated.

[0108] Step 232: sequentially determine the related first-order user nodes corresponding to the user nodes to be updated.

[0109] for example Figure 6 , if the user node to be updated is user 1, then the related first-order user node corresponding to user 1 is user 2; if the user node to be updated is user 2, then the related first-order user nodes corresponding to user 2 are user 1, user 3, and user 7; if the user node to be updated is user 3, then the related first-order user nodes corresponding to user 3 are user 2 and user 7; if the user node to be updated is user 7, then the related first-order user nodes corresponding to user 7 are user 2 and user 3.

[0110] Step 233: Use the local self-attention mechanism to determine the weight ratios between the user node to be updated and the related first-order user nodes in turn.

[0111] That is, the local self-attention mechanism is applied independently to each user node in the first-order subgraph. For details, please refer to Figure 8-Figure 11 ,in, Figure 8-Figure 11 The local self-attention calculations are performed on user 1, user 2, user 3, and user 7 respectively.

[0112] For example, user 1 has vector A = [3.2, 4.6], user 2 has vector B = [2.6, 5.6], user 3 has vector C = [4.2, 2.5], and user 7 has vector D = [2.4, 3.8].

[0113] For example, for the user node to be updated, i.e., user 1, AA=0.3 and AB=0.7 can be obtained; for the user node to be updated, i.e., user 2, BB=0.2, AB=0.3, BC=0.4 and BD=0.1 can be obtained; for the user node to be updated, i.e., user 3, CC=0.2, BC=0.4 and CD=0.4 can be obtained; for the user node to be updated, i.e., user 7, DD=0.2, BD=0.4 and CD=0.4 can be obtained.

[0114] Step 234: Based on the weight ratios, the intermediate vector representations corresponding to the user nodes to be updated are obtained in sequence.

[0115] For example, taking AA=0.3 and AB=0.7 as an example, the updated intermediate vector represents A'=0.3*[3.2,4.6]+0.7*[2.6,5.6]=[2.78,5.3].

[0116] Similarly, the updated intermediate vector representations B', C', and D' can be obtained in sequence according to this method, and this application will not elaborate on them here.

[0117] Step 240: Perform pooling processing on each intermediate vector representation to obtain a sub-graph vector representation corresponding to the first-order sub-graph.

[0118] Among them, pooling processing can adopt maximum pooling, average pooling, classification / clustering pooling, weighted pooling, differentiated pooling and the like.

[0119] Preferably, this embodiment adopts an average pooling processing method to obtain a sub-image vector representation corresponding to a first-order sub-image. The specific average pooling processing can be referred to as follows.

[0120] Among them, the intermediate vector representation includes the first dimension value and the second dimension value, and the sub-graph vector representation also includes the first dimension value and the second dimension value; then the first dimension values ​​in all the intermediate vector representations are averaged to obtain the first dimension value of the sub-graph vector representation; the second dimension values ​​in all the intermediate vector representations are averaged to obtain the second dimension value of the sub-graph vector representation; based on the first dimension value and the second dimension value of the sub-graph vector representation, the sub-graph vector representation corresponding to the first-order sub-graph is obtained.

[0121] For example, the intermediate vector represents A'=[x1,y1], the intermediate vector represents B'=[x2,y2], the intermediate vector represents C'=[x3,y3], the intermediate vector represents D'=[x4,y4], and the subgraph vector corresponding to the final first-order subgraph is represented by J=[X,Y].

[0122] Among them, each intermediate vector of x1, x2, x3, and x4 represents the corresponding first dimension value; each intermediate vector of y1, y2, y3, and y4 represents the corresponding second dimension value; X' is the first dimension value represented by the sub-graph vector, and Y' is the second dimension value represented by the sub-graph vector.

[0123] The subgraph vector corresponding to the final first-order subgraph is expressed as J=[X,Y], then X=(x1+x2+x3+x4) / 4, Y=(y1+y2+y3+y4) / 4.

[0124] Step 250: Use the global self-attention mechanism to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph.

[0125] Step 250 may include the following:

[0126] Step 251: using the global self-attention mechanism to sequentially calculate the relationship weights between each sub-graph vector representation and all other sub-graph vector representations;

[0127] Step 252: Based on the relationship weights, the target node representations corresponding to the first-order subgraphs are obtained in turn.

[0128] For example, the subgraph of the first-order subgraph of user 1 is represented by JI, the subgraph of the first-order subgraph of user 2 is represented by J2, the subgraph of the first-order subgraph of user 3 is represented by J3, the subgraph of the first-order subgraph of user 4 is represented by J4... and the subgraph of the first-order subgraph of user 13 is represented by J13.

[0129] For JI, the relationship weights between JI and JI, J2, J3, J4, J5.......J13 can be calculated to obtain JIJI, JIJ2, JIJ3, JIJ4, JIJ5.......JIJ13. Based on JIJI, JIJ2, JIJ3, JIJ4, JIJ5.......JIJ13, the globally updated target node representation JI' can be obtained.

[0130] Similarly, for J2, the relationship weights between J2 and JI, J2, J3, J4, J5.......J13 can be calculated to obtain JIJ2, J2J2, J2J3, J2J4, J2J5.......J2J13. Based on JIJ2, J2J2, J2J3, J2J4, J2J5.......J2J13, the globally updated target node representation J2' can be obtained.

[0131] The target node representations corresponding to the remaining first-order subgraphs are also calculated in this manner, which will not be described in detail in this application.

[0132] Step 260: Use the trained multi-layer perceptron to predict the target node representation and obtain the prediction result output by the multi-layer perceptron.

[0133] For example, the target nodes J1', J2', .....J13' are sequentially input into the trained multi-layer perceptron for prediction, and the prediction results of the multi-layer perceptron outputting users 1 to 13 are obtained.

[0134] It should be noted that, according to actual needs, the target node representation corresponding to a specific user to be predicted can be input into a trained multilayer perceptron for prediction to obtain a prediction result about the specific user output by the multilayer perceptron.

[0135] Step 270: Determine the identity category information of the target user based on the prediction result.

[0136] Among them, step 210 and step 260 have the same or similar technical solutions as the above-mentioned embodiment, and are not described in detail here.

[0137] Based on the above embodiments, the present application also provides a user identity recognition system based on local information and global information. The user identity recognition system includes a local structure perception module and a global information perception module.

[0138] Among them, the local structure perception module mainly performs the following steps:

[0139] S1: Extract the first-order subgraph around the target user node from the original graph;

[0140] S2: Apply the local self-attention mechanism independently to each node in the first-order subgraph to obtain the final node representation;

[0141] The final node representation here refers to the intermediate vector representation obtained after updating using the local self-attention mechanism;

[0142] S3: Use the subgraph pooling layer to aggregate the final node representation into a subgraph representation, i.e., a subgraph vector representation.

[0143] For the specific operation of the global information perception module, please refer to the introduction of the above embodiment, and this application will not go into details here.

[0144] In addition, corresponding to the user identity recognition method based on local information and global information in the above embodiment, the user identity recognition system provided by the present application also includes a unit for executing the above method, such as an acquisition unit, an extraction unit, an analysis unit and a prediction unit.

[0145] The acquisition unit is used to acquire an original graph to be processed, wherein the original graph includes a plurality of target user nodes;

[0146] The extraction unit is used to extract the first-order subgraph corresponding to each target user node from the original graph;

[0147] The analysis unit is used to determine the sub-graph vector representation corresponding to each first-order sub-graph based on the local self-attention mechanism;

[0148] The global self-attention mechanism is used to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph;

[0149] The prediction unit is used to predict the target node representation using the trained multi-layer perceptron to obtain the prediction result output by the multi-layer perceptron;

[0150] The identity category information of the target user is determined based on the prediction results.

[0151] It can be understood that the above-mentioned units are also used to implement the technical solution of any embodiment in the present application.

[0152] The present application also provides a computer device, the computer device comprising a memory and a processor, wherein the memory stores a computer program; and the processor is used to implement the following method steps when executing the computer program:

[0153] Acquire an original graph to be processed, wherein the original graph includes a plurality of target user nodes;

[0154] Extract the first-order subgraph corresponding to each target user node from the original graph;

[0155] Determine the sub-graph vector representation corresponding to each first-order sub-graph based on the local self-attention mechanism;

[0156] The global self-attention mechanism is used to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph;

[0157] Use the trained multi-layer perceptron to predict the target node representation and obtain the prediction result output by the multi-layer perceptron;

[0158] The identity category information of the target user is determined based on the prediction results.

[0159] It can be understood that the processor is also used to implement the technical solution of any embodiment of the present application when executing a computer program.

[0160] See also Fig.12 , Fig.12 The structure of an embodiment of a computer-readable storage medium provided by the present application is shown in FIG. 7 . The computer-readable storage medium 70 is used to store a computer program 71. When the computer program 71 is executed by a processor, it is used to implement the following method steps:

[0161] Acquire an original graph to be processed, wherein the original graph includes a plurality of target user nodes;

[0162] Extract the first-order subgraph corresponding to each target user node from the original graph;

[0163] Determine the sub-graph vector representation corresponding to each first-order sub-graph based on the local self-attention mechanism;

[0164] The global self-attention mechanism is used to update the vector representation of each subgraph to obtain the target node representation corresponding to each first-order subgraph;

[0165] Use the trained multi-layer perceptron to predict the target node representation and obtain the prediction result output by the multi-layer perceptron;

[0166] The identity category information of the target user is determined based on the prediction results.

[0167] It can be understood that when the computer program 71 is executed by the processor, it is also used to implement the technical solution of any embodiment of the present application.

[0168] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only illustrative, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as two units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0169] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on two network units. Some or all of the units may be selected according to actual needs to achieve the purpose of this embodiment.

[0170] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0171] If the integrated units in the above other embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0172] The above are only implementation methods of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A user identification method based on local information and global information, characterized in that: The method comprises: Acquire an original graph to be processed, wherein the original graph includes a plurality of target user nodes; Extracting a first-order subgraph corresponding to each target user node from the original graph; Determine a sub-graph vector representation corresponding to each of the first-order sub-graphs based on a local self-attention mechanism; Using a global self-attention mechanism to update the vector representation of each subgraph, and obtain the target node representation corresponding to each first-order subgraph; Using a trained multi-layer perceptron to predict the target node representation, and obtaining a prediction result output by the multi-layer perceptron; Determine the identity category information of the target user based on the prediction result; Wherein, the first-order subgraph includes a plurality of first-order user nodes corresponding to the target user; The determining the subgraph vector representation corresponding to each of the first-order subgraphs based on the local self-attention mechanism includes: Using the local self-attention mechanism to preliminarily update the initial vector representation of each of the first-order user nodes to obtain an intermediate vector representation corresponding to each of the first-order user nodes; Performing pooling processing on each of the intermediate vector representations to obtain a sub-graph vector representation corresponding to the first-order sub-graph; The method of using the local self-attention mechanism to preliminarily update the initial vector representation of each of the first-order user nodes to obtain the intermediate vector representation corresponding to each of the first-order user nodes includes: Obtaining user nodes to be updated from the first-order subgraph in sequence; Determine in sequence the related first-order user nodes corresponding to the user nodes to be updated; Determine the weight ratio between the user node to be updated and the related first-order user node in sequence by using the local self-attention mechanism; Based on the weight ratios, intermediate vector representations corresponding to the user nodes to be updated are obtained in sequence; The step of performing pooling processing on each of the intermediate vector representations to obtain a subgraph vector representation corresponding to the first-order subgraph includes: Performing average pooling processing on each of the intermediate vector representations to obtain a sub-graph vector representation corresponding to the first-order sub-graph; The intermediate vector representation includes a first dimension value and a second dimension value, and the sub-graph vector representation also includes a first dimension value and a second dimension value; The step of performing average pooling processing on each of the intermediate vector representations to obtain a sub-graph vector representation corresponding to the first-order sub-graph includes: averaging the first dimension values ​​in all the intermediate vector representations to obtain the first dimension value of the sub-graph vector representation; averaging the second dimension values ​​in all the intermediate vector representations to obtain the second dimension value of the sub-graph vector representation; A sub-graph vector representation corresponding to the first-order sub-graph is obtained according to the first dimension value and the second dimension value of the sub-graph vector representation.

2. The user identification method according to claim 1, characterized in that: The method of updating each subgraph vector representation by using a global self-attention mechanism to obtain a target node representation corresponding to each first-order subgraph includes: The global self-attention mechanism is used to sequentially calculate the relationship weights between each sub-graph vector representation and all other sub-graph vector representations; Based on the relationship weights, the target node representations corresponding to the first-order subgraphs are obtained in turn.

3. The user identification method according to claim 1, characterized in that: The prediction result includes a probability value of an identity category, and the probability value is between 0 and 1, and the identity category information of the target user includes a normal user and an abnormal user; The determining the identity category information of the target user based on the prediction result includes: If the probability value is greater than or equal to a preset value, the target user is the abnormal user; If the probability value is less than the preset value, the target user is the normal user.

4. A user identification system based on local information and global information, characterized in that: The user identification system comprises a unit for executing the method as claimed in any one of claims 1-3.

5. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the user identity recognition method according to any one of claims 1 to 3 when executing the computer program.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program data, and when the program data is executed by a processor, it is used to implement the user identity identification method according to any one of claims 1 to 3.

Citation Information

Patent Citations

  • New technology identification method based on dynamic graph anomaly detection

    CN116561688A

  • Marine target detection method under visible light based on improved YOLOv7 algorithm

    CN116863293A

  • Dynamic global semantic fusion-based common sense question and answer method and system

    CN117474089A

  • Systems and methods for machine-learned optimization of wireless networks

    WO2022258846A1