Individual industrial and commercial tenant identification method, apparatus and device, medium and program product

Abstract

The invention provides an individual industrial and commercial tenant identification method and device, equipment, a medium and a program product, relates to the field of artificial intelligence, can be used in the field of financial science and technology, and can improve the accuracy of identifying individual industrial and commercial tenants from individual users. The method comprises the steps that first transaction relation dynamic graph data including T first snapshots are obtained, n first users serve as nodes of the T first snapshots, and an attribute connecting edge is set for m transaction relations between the users in the first snapshots respectively; and inputting the first transaction relation dynamic graph data to the individual industrial and commercial tenant identification model to obtain an identification result. The processing process of the individual industrial and commercial tenant identification model comprises the following steps: respectively carrying out graph data processing based on each attribute connecting edge in a first snapshot graph to obtain a first node matrix; fusing m first node matrixes corresponding to the first snapshot graph to obtain a first transaction space feature matrix; and performing time sequence feature processing on the T first transaction space feature matrixes to obtain an identification result.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and can be used in the field of financial technology. More specifically, it relates to a method, device, equipment, medium and program product for identifying individual business owners. Background Art

[0002] Due to their small scale of production and operation and weak risk resistance, individual industrial and commercial households inevitably face many difficulties in their actual business activities. Providing exclusive financial services to individual industrial and commercial households based on their specific needs in production and operation can improve effective support for individual industrial and commercial households.

[0003] In the bank data system, self-employed individuals usually open accounts as individual users, which results in self-employed individuals and ordinary individual users being mixed together in the bank data system. Therefore, how to accurately identify the self-employed individuals of individual users has become the primary issue in providing financial support for self-employed individuals. The relevant machine learning algorithm model mainly identifies the identity of self-employed individuals based on the transaction behavior characteristics of the user himself, and the identification results are not accurate enough. Summary of the invention

[0004] In view of the above problems, the present invention provides a method and device for identifying individual business owners that can improve recognition accuracy, a method and device for constructing an individual business owner identification model, and a device, medium and program product.

[0005] The first aspect of the embodiment of the present invention provides a method for identifying individual business owners. The method comprises: firstly obtaining first transaction relationship dynamic graph data, wherein the first transaction relationship dynamic graph data comprises T first snapshot graphs corresponding to T consecutive time windows, wherein the T first snapshot graphs are all nodes of n first users, and each first snapshot graph comprises m types of attribute edges corresponding to m types of transaction relationships, wherein in each first snapshot graph, for each of the m types of transaction relationships existing between two first users in the time window corresponding to the first snapshot graph, corresponding types of attribute edges are set, and m, n, and T are all integers greater than 1; then, the first transaction relationship dynamic graph data is input into an individual business owner identification model, and the identification result of each first user among the n first users by the individual business owner identification model is obtained, and the individual business owner identification model is a trained machine learning model for classifying users between ordinary users and individual business owners.

[0006] Among them, the process of the individual business household identification model processing the first transaction relationship dynamic graph data includes: performing graph data processing based on each attribute edge in the first snapshot graph to obtain a first node matrix corresponding to each transaction relationship; fusing the m first node matrices corresponding to each first snapshot graph to obtain a first transaction space feature matrix corresponding to each first snapshot graph; and obtaining the identification result based on the time series feature processing of T first transaction space feature matrices corresponding to T first snapshot graphs.

[0007] The second aspect of the embodiment of the present invention provides a method for constructing a model for identifying an individual business owner. The construction method includes: first, obtaining the second transaction relationship dynamic graph data, wherein the second transaction relationship dynamic graph data includes T second snapshot graphs corresponding to T time windows, wherein the T second snapshot graphs are all based on s second users as nodes, and each second snapshot graph includes m types of attribute edges corresponding to m types of transaction relationships, wherein in each second snapshot graph, for each of the m types of transaction relationships between two second users in the time window corresponding to the second snapshot graph, a corresponding type of attribute edge is set, and m, s, and T are all integers greater than 1; then, each second user in the second transaction relationship dynamic graph data is labeled with the identity tag of each second user, wherein the identity tag is an ordinary user or an individual business owner; then, the labeled second transaction relationship dynamic graph data is used to train the individual business owner identification model.

[0008] Among them, the training of the individual business owner identification model includes repeatedly performing the following operations until the model training termination condition is met: performing graph data processing based on each attribute edge in the second snapshot graph to obtain a second node matrix corresponding to each transaction relationship; fusing the m second node matrices corresponding to each second snapshot graph to obtain a second transaction space feature matrix corresponding to each second snapshot graph; obtaining a model recognition result based on the time series characteristics of the T second transaction space feature matrices corresponding to the T second snapshot graphs; and adjusting the parameters of the individual business owner identification model based on the difference between the model recognition result and the identity label of the second user in the second transaction relationship dynamic graph data.

[0009] A third aspect of an embodiment of the present invention provides a device for identifying a self-employed business operator, wherein the device comprises: a first acquisition module and an identification module. A self-employed business operator identification model is deployed in the identification module, and the self-employed business operator identification model is a trained machine learning model for classifying users between ordinary users and self-employed businesses.

[0010] The first acquisition module is used to acquire first transaction relationship dynamic graph data, wherein the first transaction relationship dynamic graph data includes T first snapshot graphs corresponding one-to-one to T consecutive time windows, wherein the T first snapshot graphs all have n first users as nodes, and each first snapshot graph includes m types of attribute edges corresponding one-to-one to m types of transaction relationships, wherein in each first snapshot graph, corresponding types of attribute edges are set for each of the m types of transaction relationships existing between two first users within the time window corresponding to the first snapshot graph, and m, n, and T are all integers greater than 1.

[0011] The identification module is used to input the first transaction relationship dynamic graph data into the individual business owner identification model, and obtain the identification result of the individual business owner identification model for each of the n first users. The individual business owner identification model is a trained machine learning model for classifying users between ordinary users and individual business owners.

[0012] The individual business owner identification model includes a spatial modeling layer, a temporal modeling layer and a node classification layer.

[0013] The spatial modeling layer is used to: perform graph data processing based on each attribute edge in the first snapshot graph to obtain a first node matrix corresponding to each transaction relationship; and fuse the m first node matrices corresponding to each first snapshot graph to obtain a first transaction space feature matrix corresponding to each first snapshot graph.

[0014] The time modeling layer is used to process the time series features of the T first transaction space feature matrices corresponding to the T first snapshot graphs.

[0015] The node classification layer is used to process the output of the time modeling layer to obtain the recognition result

[0016] A fourth aspect of the embodiments of the present invention provides a device for constructing a self-employed business owner identification model, wherein the device comprises a second acquisition module, a labeling module and a training module.

[0017] The second acquisition module is used to obtain second transaction relationship dynamic graph data, wherein the second transaction relationship dynamic graph data includes T second snapshot graphs corresponding one-to-one to T time windows, wherein the T second snapshot graphs all have s second users as nodes, and each second snapshot graph includes m types of attribute edges corresponding one-to-one to m types of transaction relationships, wherein in each second snapshot graph, corresponding types of attribute edges are set for each of the m types of transaction relationships existing between two second users within the time window corresponding to the second snapshot graph, and m, s, and T are all integers greater than 1.

[0018] The labeling module is used to label each second user in the second transaction relationship dynamic graph data with an identity tag of each second user, wherein the identity tag is an ordinary user or an individual business operator.

[0019] The training module is used to train the individual business household identification model using the annotated second transaction relationship dynamic graph data.

[0020] Among them, the training module is used to repeatedly perform the following operations until the model training termination condition is met: based on each attribute edge in the second snapshot graph, graph data processing is performed separately to obtain a second node matrix corresponding to each transaction relationship; the m second node matrices corresponding to each second snapshot graph are merged to obtain a second transaction space feature matrix corresponding to each second snapshot graph; based on the time series characteristics of the T second transaction space feature matrices corresponding to the T second snapshot graphs, a model recognition result is obtained; and based on the difference between the model recognition result and the identity label of the second user in the second transaction relationship dynamic graph data, the parameters of the individual business identification model are adjusted.

[0021] A fifth aspect of an embodiment of the present invention provides an electronic device. The electronic device includes: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method provided in the first aspect or the second aspect.

[0022] The sixth aspect of the embodiments of the present invention further provides a computer-readable storage medium on which a computer program or instructions are stored. When the computer program or instructions are executed by a processor, the steps of the method provided in the first or second aspect are implemented.

[0023] The seventh aspect of the embodiments of the present invention further provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the steps of the method provided in the first or second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above contents and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0025] Figure 1 A flowchart of a method for identifying an individual business operator according to an embodiment of the present invention is schematically shown;

[0026] Figure 2 The structural relationship of the multi-dimensional information contained in the transaction relationship dynamic graph data according to an embodiment of the present invention is schematically shown;

[0027] Figure 3The model architecture of the individual business household identification model in one embodiment of the present invention is schematically shown;

[0028] Figure 4 A flowchart schematically shows a method for constructing a self-employed business identification model according to an embodiment of the present invention;

[0029] Figure 5 The structure of the space modeling layer in the self-employed business identification model according to an embodiment of the present invention is schematically shown;

[0030] Figure 6 The working process of the spatial modeling layer in the self-employed business identification model according to an embodiment of the present invention is schematically shown;

[0031] Figure 7 The internal structure of the time modeling layer in the self-employed business identification model according to an embodiment of the present invention is schematically shown;

[0032] Figure 8 A block diagram schematically shows a device for identifying an individual business operator according to an embodiment of the present invention;

[0033] Fig. 9 A block diagram schematically shows a device for constructing a self-employed business owner identification model according to an embodiment of the present invention; and

[0034] Fig.10 A block diagram of an electronic device suitable for implementing the method for identifying individual business owners or the method for constructing an individual business owner identification model according to an embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0035] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0036] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0037] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner. The terms "first" and "second" herein are used only to distinguish and do not have a limiting effect.

[0038] It should be noted that in the technical solution of the present invention, the user information (including but not limited to user personal information, user accounts, historical transaction information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0039] The inventor found that in actual transaction relationships, individual industrial and commercial households often carry out production and business activities as a family unit, with the characteristics of family aggregation and regional aggregation. For example, the business locations of individual industrial and commercial households are usually concentrated in some specific areas, so trading counterparts (such as consumers) are often concentrated in a large number in a specific area, and due to the single source of goods and other reasons, their suppliers are often limited to specific dealers. Therefore, the user characteristics of individual industrial and commercial households are not only affected by themselves, but also closely related to the characteristic information of trading counterparts. In the related art, when identification is performed only based on the characteristic analysis of the user itself, the correlation between different users is ignored, and it is difficult to use the incremental information generated by the mutual influence between users, which leads to certain limitations on the accuracy of prediction. In addition, the inventor also found that since the business behavior of individual industrial and commercial households will be adjusted with the changes of market environment, business conditions and other factors, the topological structure formed by the transaction relationship between individual industrial and commercial households and other users will change regularly over time. The dynamic changes of these topological structures can be captured by introducing time series analysis to more accurately identify the identity of individual industrial and commercial households.

[0040] In view of this, the embodiments of the present invention provide a method, apparatus, device, medium and program product for identifying individual business owners, which can be based on graph data processing and analysis and processing of time series features. After obtaining the user's authorization to use his or her bank transaction data, based on the user's bank statements and other transaction data, extract the multiple transaction relationship features between the user and other users and the time series features generated by the topological structure formed by these multiple transaction relationships changing over time, thereby identifying the identity of the individual business owner.

[0041] Figure 1The flowchart of the method for identifying an individual business operator according to an embodiment of the present invention is schematically shown.

[0042] like Figure 1 As shown, the method for identifying individual business owners according to this embodiment may include operation S110 and operation S120, wherein operation S120 may specifically include operations S121 to S123.

[0043] First, in operation S110, first transaction relationship dynamic graph data is obtained.

[0044] Reference here Figure 2 The structural relationship of the multi-dimensional information contained in the transaction relationship dynamic graph data is used to illustrate the first transaction relationship dynamic graph data.

[0045] Specifically, the first transaction relationship dynamic graph data may include T first snapshot graphs corresponding to T consecutive time windows, so that the T first snapshot graphs constitute a sequence. The T first snapshot graphs all have n first users as nodes.

[0046] Each first snapshot graph includes m types of attribute edges corresponding to m types of transaction relationships. Specifically, in each first snapshot graph, for each of the m types of transaction relationships between two first users in the time window corresponding to the first snapshot graph, corresponding types of attribute edges are set. The m types of transaction relationships include but are not limited to: consumer payment transactions, settlement transactions, and collection and operation transactions.

[0047] The first transaction relationship dynamic graph data can be constructed in the following manner: when the use authorization of n first users for their transaction data is obtained, the first historical transaction data formed by the n first users in the first historical period is obtained, wherein the first historical period can be divided into T time windows; then the transaction relationship contained in each data of the first historical transaction data is identified (for example, the transaction remarks of each data of the first historical transaction data are processed by natural language processing technology to identify the transaction relationship contained in each data), and then the first historical transaction data is divided into T first sub-data sets according to the T time windows; next, a first snapshot graph is obtained based on each first sub-data set, wherein each first user is mapped to a node in the first snapshot graph, and the data containing the same transaction relationship between two first users are merged and mapped to an attribute edge in the first snapshot graph, so that the same transaction relationship between the two nodes is set with an attribute edge, and different transaction relationships are set with different types of attribute edges; finally, based on the T first snapshot graphs, the first transaction relationship dynamic graph data is obtained. It can be seen that the transaction relationship dynamic graph data is represented by a set of snapshot graphs of multi-period transaction relationships, wherein each snapshot graph is a heterogeneous graph containing multiple attribute edges.

[0048] It can be seen that the first transaction relationship dynamic graph data integrates the transaction dynamic information of n first users in the spatial dimension and the time dimension. Among them, the information in the spatial dimension is not only reflected by the node relationship in the graph, but also distinguishes different types of transaction relationships through the types of attribute edges. The information in the time dimension is reflected by the sequence of T snapshot graphs. The information of multiple attribute edges in the spatial dimension can reflect the clustering characteristics of nodes, which corresponds to the family clustering and regional clustering behavior characteristics of individual industrial and commercial households. The sequence changes of the information in the spatial dimension in the time dimension can reflect the changes of the multiple relationship states between nodes over time, which corresponds to the behavior characteristics of individual industrial and commercial households with changes in market environment, business conditions and other factors.

[0049] In one embodiment, each first snapshot graph also includes the node initial features of each node, wherein the node initial features of each node include information on the transaction behavior of the first user corresponding to the node within the time window corresponding to the first snapshot graph. For example, when the user's authorization to use the relevant data is obtained, the node initial features of each first user are formed according to the basic data of each first user (such as personal basic information) and the transaction behavior of the first user in a time window (such as time window 1) (such as transaction frequency, total transaction amount, transaction amount or number of times under various transaction relationships, etc.). In this way, the node initial features of the nodes in the snapshot graphs under different time windows will also change dynamically, thereby supplementing more information on the fact that the behavior of individual industrial and commercial households will show certain regularity and trends as the market environment changes, business conditions and other factors change.

[0050] Next, in operation S120, the first transaction relationship dynamic graph data is input into the self-employed business identification model, and the self-employed business identification model obtains the identification result of each first user among the n first users. The self-employed business identification model is a trained machine learning model for classifying users between ordinary users and self-employed businesses. Specifically, operation S120 may include operations S121 to S123.

[0051] In operation S121, graph data processing is performed based on each attribute edge in the first snapshot graph to obtain a first node matrix corresponding to each transaction relationship.

[0052] In operation S122, the m first node matrices corresponding to each first snapshot image are fused to obtain the first transaction space feature matrix corresponding to each first snapshot image. When training the individual business identification model, the model can be trained to learn the importance of different transaction relationships to the final identification result, so that the m first node matrices can be specifically fused according to the importance.

[0053] In operation S123, based on the time series feature processing of the T first transaction space feature matrices corresponding to the T first snapshot images, the recognition result is obtained. Specifically, the time series feature of the T first transaction space feature matrices can be processed using a neural network that processes time series features to obtain the spatiotemporal coding of each node, and then the node classification is performed based on the spatiotemporal coding of each node to obtain the probability of whether each first user is an individual business operator.

[0054] Since individual business owners have the characteristics of team aggregation and regional aggregation, the transaction space feature matrix that integrates multiple transaction relationships in the embodiment of the present invention can effectively reflect the specific characteristics of users in this aspect. At the same time, as the market environment changes, business conditions and other factors change, the behavior of individual business owners often shows certain regularity and trends. The time series feature processing of T transaction space feature matrices in the embodiment of the present invention can also effectively extract this aspect of features. It can be seen that the feature extraction and architecture design of the individual business owner identification model in the embodiment of the present invention fully considers the spatiotemporal characteristics of the behavior of individual business owners, so that more accurate identification can be performed based on more comprehensive spatiotemporal characteristics.

[0055] In the embodiment of the present invention, the first transaction relationship dynamic graph data input to the model integrates the transaction information of the user in the spatial dimension and the time dimension, wherein the information in the spatial dimension is not only reflected by the node relationship in the graph, but also distinguishes different types of transaction relationships by the types of attribute edges, and the information in the time dimension is reflected by a sequence of T snapshot graphs. At the same time, when the model is identified, the graph data is first processed in the spatial dimension according to each transaction relationship to extract the node spatial features under a single type of transaction relationship, and then the node spatial features under m types of transaction relationships are fused together to form a transaction space feature matrix that integrates m types of transaction relationships, so as to mine the spatial aggregation features between nodes. Then, in the time dimension, the model can analyze the time series features of the transaction space feature matrices that integrate m types of transaction relationships, so as to mine the time series change features of the spatial features of the nodes, thereby obtaining the final time and space features of the nodes. In this way, the time and space features of the nodes of individual industrial and commercial households and ordinary users can be more clearly distinguished, so that the model can more accurately identify the identity of individual industrial and commercial households.

[0056] In addition, in the embodiment of the present invention, the transaction relationship dynamic graph data of the input model is represented by discrete snapshot graphs, which better balances the computational complexity and efficient mining of original information, greatly saves graph storage and computing resources, and ensures that an efficient computing process can be executed while the model predicts classification, thereby having high computational efficiency.

[0057] In one embodiment, the model architecture of the individual business owner identification model is as follows: Figure 3As shown, a spatial modeling layer, a temporal modeling layer and a node classification layer may be included.

[0058] In one embodiment, the spatial modeling layer includes m neighborhood node aggregation modules, wherein the neighborhood node aggregation module is constructed using a graph convolutional network. Accordingly, the above operation S121 may specifically include: first extracting a first hierarchical snapshot graph including only one attribute edge from the first snapshot graph to obtain m first hierarchical snapshot graphs, and then inputting each first hierarchical snapshot graph into a neighborhood node aggregation module, obtaining a first node matrix output by the neighborhood node aggregation module after fusing the node neighborhood relationship in the first hierarchical snapshot graph. In this way, the hierarchical snapshots corresponding to each transaction relationship can be processed separately using m neighborhood node aggregation modules constructed using a graph convolutional network, thereby extracting the importance of the transaction object under the same relationship to each user in the associated network, and aggregating the neighbor information to form a node representation in the node matrix. In this way, the neighborhood node information of each attribute edge of the node can be embedded into the node representation through the convolutional network, and the spatial neighborhood information of multi-order transaction counterparties can be modeled and processed to supplement more information.

[0059] In one embodiment, the spatial modeling layer also includes a multi-relationship view fusion module, in which a trained weight matrix is ​​provided. Accordingly, in the above operation S122, the node representations of the corresponding nodes in the m first node matrices can be integrated through the weight matrix in the multi-relationship view fusion module to obtain a first transaction space feature matrix, thereby differentially fusing user features from multiple transaction scenarios (i.e., transaction relationships).

[0060] In one embodiment, the time modeling layer is constructed using a long short-term memory network. The node classification layer can be a binary classifier model, such as one constructed using algorithms such as a multilayer perceptron. Accordingly, the above operation S123 may specifically include: first, the T first transaction space feature matrices are formed into a first input sequence in the order of the time windows corresponding to the T first snapshots, and then the first input sequence is input into the time modeling layer, and the first spatiotemporal feature matrix output by the time modeling layer is obtained; next, the node representation of each node in the first spatiotemporal feature matrix is ​​input into the node classification layer, and the classification result of each node output by each node classification layer is obtained, so as to obtain the recognition result of the first user corresponding to each node according to the classification result of each node. Among them, the node representation in the spatiotemporal feature matrix output by the time modeling layer is the final representation of each node, which integrates the influence of the information of the user node in different time windows on the spatiotemporal node representation of the final user, so that the spatiotemporal information of the classification decision is more comprehensive and the classification is more accurate.

[0061] Figure 4 The flowchart schematically shows a method for constructing a self-employed business owner identification model according to an embodiment of the present invention.

[0062] like Figure 4 As shown, according to this embodiment, the method for constructing an individual business owner identification model may include operations S210 to S250.

[0063] Operation S210, collecting historical bank statement data of the target customer group, and establishing transaction relationship dynamic graph data based on the historical bank statement data, wherein the target customer group includes a plurality of second users with known identity tags, wherein the identity tag of each second user is an ordinary user or an individual business operator.

[0064] First, data collection and preprocessing are carried out, and then transaction relationship dynamic graph data is established based on the bank statement data collected for a certain historical period (such as the whole year).

[0065] Combination Figure 2 ,The formula definition of transaction relationship dynamic graph data can be expressed as follows:

[0066]

[0067] Wherein, V is a node set. For example, for the first transaction relationship dynamic graph data, the node set V includes n first users mapped to n nodes.

[0068] is an adjacency matrix; in one embodiment, It can be represented as a combination of m sub-adjacency matrices, where each sub-adjacency matrix is ​​an adjacency matrix of a hierarchical snapshot graph formed by extracting an attribute edge from the first snapshot graph; in another embodiment, Each element in can be an m-dimensional vector, where each dimension of the m-dimensional vector corresponds to an attribute edge;

[0069] is the initial feature of the node;

[0070] Store the identity tag information corresponding to each node; t is the identifier of the time window, with a value of 1~T, which means that the entire historical period (such as one year) corresponding to the historical bank statement data is divided into T time windows. Different time windows have the same length and do not overlap with each other. For example, a year can be divided into 12 months according to the natural month, and each month is a time window;

[0071] Among them, the adjacency matrix and node initial features It changes dynamically in different time windows.

[0072] It can be seen that the nodes in the transaction relationship dynamic graph data are all users in the target customer group, and the edges include multi-attribute edges, which correspond to the transaction relationships such as consumption payment, settlement transfer, and business collection generated or continuously existing between users, corresponding to transactions in different transaction scenarios. The transaction relationship dynamic graph data is represented by a set of multi-period transaction relationship group graphs, where each snapshot graph is a heterogeneous graph containing multi-attribute edges.

[0073] Operation S220, through the spatial modeling layer of the graph convolutional network, integrates the information of multi-hop transaction objects in different transaction scenarios (transaction relationships), and outputs a sequence consisting of T transaction space feature matrices.

[0074] Specifically, the constructed transaction relationship dynamic graph data is input into the spatial modeling layer to generate a user spatial embedding representation that integrates multi-hop transaction counterparties under different transaction relationships, and realizes the mining and embedding of neighborhood transaction object features under each time window. Among them, the spatial modeling layer can use a multi-layer graph convolutional network to spatially encode the snapshot graph under each time window to obtain the transaction space feature matrix.

[0075] The output of the spatial modeling layer is represented in the form of a sequence, which is a collection of node spatial feature representations of multiple second users under multiple time windows, and is defined as follows:

[0076]

[0077] in, is the transaction space feature matrix corresponding to the t-th time window, which is obtained by connecting the spatial encodings of n nodes in the t-th time window, and represents the node space feature matrix in time window t.

[0078] like Figure 5 As shown, the spatial modeling layer may include a neighborhood node aggregation module and a multi-relationship view fusion module. In other embodiments, considering that the proportion of self-employed individuals in individual users of a bank is relatively small, the spatial modeling layer may be further configured to include a subgraph generation module during model training, wherein the subgraph generation module is used to sample nodes in each snapshot graph based on an unbalanced strategy, and then construct a subgraph using the sampled nodes as input for subsequent model learning, thereby solving the problem of user sample imbalance in the target user group and increasing the proportion of self-employed individuals samples that the model can learn.

[0079] The subgraph generation module, the neighborhood node aggregation module, and the multi-relation view fusion module together constitute the spatial feature extractor on a single time snapshot. The working process of the spatial modeling layer includes the following steps 2.1 to 2.4, combined with Figure 5 and Figure 6 Instructions are as follows.

[0080] Step 2.1: Input the snapshot of the transaction relationship dynamic graph data into the subgraph generation module. The subgraph generation module samples the nodes in each snapshot graph based on the imbalance strategy, and then constructs a subgraph using the sampled nodes, and uses the subgraph as the input of the subsequent neighborhood node aggregation module.

[0081] In one embodiment, the sampling probability of each node in a snapshot graph in an unbalanced strategy is positively correlated with the complexity of the connection relationship between the node and other nodes in the snapshot graph, and negatively correlated with the proportion of the identity tag of the second user corresponding to the node in the identity tags in the overall data set, wherein the complexity of the connection relationship between a node and other nodes is obtained based on the number and type of attribute edges between the node and other nodes. In this way, when constructing a subgraph, nodes with high complexity of relationship with other nodes and belonging to minority classes can be retained as much as possible.

[0082] In one embodiment, the formula for calculating the sampling probability assigned to each node based on the unbalanced strategy is as follows:

[0083]

[0084] The denominator represents the proportion of the node's label in the entire data set; the numerator is the square value of the norm of the node column vector in the normalized adjacency matrix, which can be used to measure the complexity of the connection relationship between a node and other nodes. The normalized adjacency matrix can be obtained by For example, the adjacency matrix Different attribute edges are assigned different weights (or all are assigned weight 1) and then added up.

[0085] The subgraph generation module can effectively improve the participation of these minority class nodes in the training process, thereby helping to alleviate the problem of unbalanced sample label distribution.

[0086] Step 2.2: Input the subgraphs generated by the subgraph generation module into m neighborhood node aggregation modules respectively. Each neighborhood node aggregation module is used to learn the neighborhood information of a node under a transaction relationship and output a node matrix corresponding to the transaction relationship.

[0087] Specifically, a node-level attention mechanism can be introduced into the neighborhood node aggregation module, so that the neighborhood node aggregation module can learn the importance of the transaction object under the same transaction relationship to each user in the associated network, and aggregate the neighbor information to form the node hidden layer representation, in order to prevent too much noise from being mixed in the process of neighborhood aggregation, and avoid aggregating the characteristics of a large number of ordinary individual users into individual businesses. In the neighborhood node aggregation module, weighted aggregation is performed by calculating the standardized attention coefficient, which realizes the differential aggregation of noise information and normal information.

[0088] In one embodiment, in the neighborhood node aggregation module In the layer, the neighborhood vector is obtained through aggregation and update functions to represent the neighborhood information of the node, which is fused with the hidden state from the previous layer. The formula of the above process is as follows:

[0089]

[0090] in, Represents an aggregate function; represents the update function; Indicated in Layer, the node embedding representation of all nodes under the transaction relationship r in the t-th time window; It represents the set of all neighbor nodes of the node in the transaction relationship r in the tth time window.

[0091] The node-level attention calculation method based on the transaction relationship r includes the following steps 2.2.1 to 2.2.3.

[0092] Step 2.2.1: Obtain the importance of all neighbor nodes to each other through the following formula: :

[0093]

[0094] in, is the parameter vector of the node-level attention mechanism under transaction relationship r. Given a node pair connected by transaction relationship r, using node-level attention Learn the importance of mutual influence between nodes.

[0095] Step 2.2.2: As input, through Normalize to the interval [0,1] to get the standardized aggregation weight coefficient , as shown below:

[0096]

[0097] Step 2.2.3: Use attention coefficient As the weight, the neighborhood information is weighted and aggregated as follows:

[0098]

[0099] in is the node i in The feature embedding of the transaction relation r at the layer.

[0100] Step 2.3: Input the outputs of the m neighborhood node aggregation modules to the multi-relation view fusion module. Specifically, after obtaining the node representations that aggregate neighborhood information on each transaction relationship through the m neighborhood node aggregation modules, the multi-relation view fusion module fuses the embedding vectors of each node under different transaction relationships.

[0101] The multi-relationship view fusion module assigns importance coefficients to different transaction relationships by training a specific weight matrix, thereby achieving differential integration of information in multiple transaction relationships.

[0102] Specifically, in the multi-relation view fusion module Layer, multi-relation aggregation is calculated as follows:

[0103]

[0104] in, represents the Hadamard product, which is the product of the corresponding elements of two matrices or elements. is a trainable model parameter vector used to measure the different contributions of m transaction relationships to node representation. represents the tth time window in the multi-relation view fusion module The node feature representation of the layer means that each layer integrates the feature representation results of each transaction relationship.

[0105] Step 2.4: The multi-relational view fusion module obtains the embedded representation of each layer output by stacking multiple layers of graph convolutional networks. Finally, the layers with different smoothness are embedded and fused into the final node space feature representation In order to increase the difference between node representations and avoid over-smoothing problems.

[0106] Specifically, as the number of GNN layers increases, the GNN can gradually accumulate information from deeper neighboring nodes, but the differences between nodes are also weakened, especially between nodes that are close to each other in the same connected component, which is called over-smoothing. The over-smoothing problem can easily lead to performance degradation when performing classification tasks. In order to maintain the differences between nodes while taking into account the deeper features in the graph structure, the following operators are defined to comprehensively consider feature information at different levels:

[0107]

[0108] in, Function Input Indicates that a single node in time window t is The output features in the layer. The output of the function is recorded as .

[0109] In the sequence finally output by the spatial modeling layer, By n nodes The concatenation results in a transaction space feature matrix under time window t, from which the transaction feature vector of the node in each time window can be obtained.

[0110] Operation S230: The sequence of T transaction space feature matrices output by the space modeling layer is , and sent to the time modeling layer, which extracts time series features from the sequence composed of T transaction space feature matrices.

[0111] The temporal modeling layer can use long short-term memory networks to implement time series analysis and capture the dynamic changes in user transaction characteristics, thereby further learning the different temporal characteristics between individual business owners and ordinary individual users.

[0112] The internal structure of the temporal modeling layer is as follows Figure 7 In the time modeling layer, taking node u in the t-th layer network as an example, the time embedding representation calculation is as follows:

[0113]

[0114]

[0115]

[0116]

[0117]

[0118] in, , , , represents a trainable weight parameter, and tanh represents the hyperbolic tangent function. In the gating mechanism, the input gate Used to control the ratio of information that needs to be saved from the current candidate state, the forget gate Used to control the internal state of the previous time step The forgetting rate of information, output gate Used to control the output ratio of the current internal state information. Corresponding to the time window t, the LSTM unit receives the current spatial embedding and the time embedding of the previous time window t-1 and As input, output The embedding contains the temporal information of node u up to time window t.

[0119] Through the time modeling layer, the influence of the spatial information of the nodes in different time windows on the final user's spatiotemporal node representation is integrated, and the nodes in the spatiotemporal feature matrix output by the time modeling layer are used as the spatiotemporal embedding representation of the user nodes.

[0120] In this way, the spatiotemporal information in the original data is effectively extracted, and the information of multi-hop counterparties in different transaction scenarios (transaction relationships) is spatially integrated, and the dynamic time series features in different time windows are temporally integrated.

[0121] Operation S240: convert the spatiotemporal feature matrix output by the temporal modeling layer into The node representation in is connected to the node classification layer, which classifies the node representation of each node and outputs the classification of all users, that is, the probability score of belonging to the category of individual business owners and ordinary users. The node classification layer can be constructed using a multilayer perceptron (MLP), and its formula is as follows:

[0122]

[0123] in, represents the input of the node classification layer, is the predicted label, which indicates the probability that the model predicts that the u-th node belongs to an individual business owner. Represents a multi-layer perceptron function, which is used to implement the binary classification task of the node.

[0124] Operation S250, based on the predicted probability of the user by the node classification layer and the actual label of the user, the overall optimization target of the model is established based on the cross entropy loss function, the model parameters are adjusted, and iterative training is performed. In one embodiment, the overall optimization target calculation formula of the model is as follows:

[0125]

[0126] The loss function consists of two parts: cross entropy loss and regularization term. is the true label of node u, It is the prediction result of the model, which indicates the probability that the u-th node belongs to the category of individual business owners.

[0127] Using cross entropy loss to measure model prediction probability With the true label The inconsistency between them quantifies the difference between the true label distribution and the model prediction distribution. is the set of parameters of the model, Represents the regularization coefficient of the model. By adding the regularization term, that is, the square of the L2 norm of the model weight, the risk of overfitting of the model is reduced.

[0128] It can be seen that the spatial modeling layer, the temporal modeling layer and the node classification layer constitute the overall structure of the individual business identification model. The overall goal of model training is to minimize the loss function by adjusting the parameters of the model. .

[0129] According to the construction method of some embodiments of the present invention, the ability of the individual business identification model to process complex and nonlinear data can be improved. Specifically, by integrating the spatial modeling layer and the temporal modeling layer, potential spatiotemporal information can be mined from different angles. The spatial modeling layer constructed using a graph convolutional network can effectively capture the impact of neighborhood spatial features on user identity recognition. The temporal modeling layer constructed using LSTM can effectively analyze the dynamic changes of user transaction behavior over time, and further learn the differentiated characteristics of individual business owners and ordinary individual users in the time dimension. This achieves a multi-level representation and effective fusion of transaction spatiotemporal information.

[0130] According to the construction method of some embodiments of the present invention, the robustness of the self-employed business identification model can be improved. Among them, the subgraph is constructed by sampling with an unbalanced strategy, which improves the recognition and generalization performance of the model for minority class samples and solves the class imbalance problem of low proportion of self-employed business customers. At the same time, the graph attention mechanism is used to compatibly process the noise information in the aggregation process, thereby improving the stability and adaptability of the model.

[0131] According to the construction method of some embodiments of the present invention, the accuracy of user identification of the individual business owner identification model can be improved. Specifically, the embodiments of the present invention make full use of the incremental information generated by the mutual influence of nodes on the graph, and compared with the machine learning algorithm that only uses the user's own information (such as the initial features of the node) for identification and classification, its performance is more outstanding and more accurate classification results can be obtained.

[0132] Figure 8 The block diagram of a device for identifying an individual business operator according to an embodiment of the present invention is schematically shown.

[0133] like Figure 8 As shown, the self-employed business identification device 40 of this embodiment includes a first acquisition module 41 and an identification module 42. The identification module 42 is deployed with a trained self-employed business identification model.

[0134] The first acquisition module 41 is used to acquire the first transaction relationship dynamic graph data, wherein the first transaction relationship dynamic graph data includes T first snapshot graphs corresponding to T consecutive time windows, wherein the T first snapshot graphs all have n first users as nodes, and each first snapshot graph includes m types of attribute edges corresponding to m types of transaction relationships, wherein in each first snapshot graph, for each of the m types of transaction relationships existing between two first users in the time window corresponding to the first snapshot graph, a corresponding type of attribute edge is set, and m, n, and T are all integers greater than 1. In one embodiment, the first acquisition module 41 can perform the operation S110 described above.

[0135] The identification module 42 is used to input the first transaction relationship dynamic graph data into the self-employed business identification model, and obtain the identification result of the self-employed business identification model for each of the n first users. The self-employed business identification model is a trained machine learning model for classifying users between ordinary users and self-employed businesses. In one embodiment, the identification module 42 can perform the operation S120 described above.

[0136] The individual business owner identification model includes a spatial modeling layer, a temporal modeling layer and a node classification layer.

[0137] The spatial modeling layer is used to process graph data based on each attribute edge in the first snapshot graph to obtain a first node matrix corresponding to each transaction relationship; and to fuse the m first node matrices corresponding to each first snapshot graph to obtain a first transaction space feature matrix corresponding to each first snapshot graph.

[0138] The time modeling layer is used to process the time series features of the T first transaction space feature matrices corresponding to the T first snapshot graphs.

[0139] The node classification layer is used to process the output of the temporal modeling layer to obtain the recognition result.

[0140] According to some other embodiments of the present invention, the first acquisition module 41 further includes a transaction acquisition submodule, a transaction relationship identification submodule, a data segmentation submodule and a graph data generation submodule.

[0141] The transaction acquisition submodule is used to acquire first historical transaction data generated by n first users in a first historical period when the use authorization of the transaction data of n first users is obtained, and the first historical period is divided into T time windows.

[0142] The transaction relationship identification submodule is used to identify the transaction relationship contained in each data of the first historical transaction data, wherein the first historical transaction data includes m types of transaction relationships, and the transaction relationship contained in each data is one of the m types of transaction relationships.

[0143] The data segmentation submodule is used to segment the first historical transaction data into T first sub-data sets according to T time windows.

[0144] The graph data generation submodule is used to: first obtain a first snapshot graph based on each first sub-data set, wherein each first user is mapped to a node in the first snapshot graph, and the data containing the same transaction relationship between two first users are merged and mapped to an attribute edge in the first snapshot graph; then, based on T first snapshot graphs, obtain the first transaction relationship dynamic graph data.

[0145] The individual business identification device 40 can perform reference Figure 1 The method for identifying individual industrial and commercial households described in detail is introduced in the previous article and will not be repeated here.

[0146] Fig. 9 A block diagram schematically shows a device for constructing a self-employed business owner identification model according to an embodiment of the present invention.

[0147] like Fig. 9 As shown, the construction device 50 according to this embodiment includes a second acquisition module 51 , a labeling module 52 and a training module 53 .

[0148] The second acquisition module 51 is used to obtain second transaction relationship dynamic graph data, wherein the second transaction relationship dynamic graph data includes T second snapshot graphs corresponding one-to-one to T time windows, wherein the T second snapshot graphs all have s second users as nodes, and each second snapshot graph includes m types of attribute edges corresponding one-to-one to m types of transaction relationships, wherein in each second snapshot graph, for each of the m types of transaction relationships existing between two second users within the time window corresponding to the second snapshot graph, a corresponding type of attribute edge is set, and m, s, and T are all integers greater than 1.

[0149] The marking module 52 is used to mark each second user in the second transaction relationship dynamic graph data with the identity tag of each second user, wherein the identity tag is a common user or an individual business operator.

[0150] The training module 53 is used to train the individual business household identification model using the annotated second transaction relationship dynamic graph data.

[0151] Specifically, the training module 53 is used to repeatedly perform the following operations until the model training termination condition is met: based on each attribute edge in the second snapshot graph, the graph data is processed separately to obtain a second node matrix corresponding to each transaction relationship; the m second node matrices corresponding to each second snapshot graph are merged to obtain the second transaction space feature matrix corresponding to each second snapshot graph; and based on the time series characteristics of the T second transaction space feature matrices corresponding to the T second snapshot graphs, the model recognition result is obtained; and based on the difference between the model recognition result and the identity label of the second user in the second transaction relationship dynamic graph data, the parameters of the individual business identification model are adjusted. Among them, the training termination condition can be the maximum number of loop iterations, or the overall optimization target of the model established based on the cross entropy loss function no longer changes or the difference between the two changes before and after is less than a threshold, etc.

[0152] According to an embodiment of the present invention, when training the individual business identification model, the training module 53 performs graph data processing based on each attribute edge in the second snapshot graph to obtain a second node matrix corresponding to each transaction relationship, which may specifically include: sampling the nodes in the second snapshot graph based on an unbalanced strategy; constructing a subgraph using the sampled nodes; and performing graph data processing based on each attribute edge in the subgraph to obtain a second node matrix corresponding to each transaction relationship. The subgraph is constructed by unbalanced sampling, and the subsequent model is trained based on the subgraph to increase the proportion of minority class samples that the model can learn.

[0153] The construction device 50 may execute reference Figure 4 The construction method of the individual business household identification model introduced here can be referred to the previous introduction for details.

[0154] In one embodiment, the construction device 50 may also be integrated into the self-employed business owner identification device 40 to be used for training and optimizing the self-employed business owner identification model in the self-employed business owner identification device 40 .

[0155] According to an embodiment of the present invention, any multiple modules of the first acquisition module 41, the identification module 42, the second acquisition module 51, the labeling module 52 and the training module 53 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the first acquisition module 41, the identification module 42, the second acquisition module 51, the labeling module 52 and the training module 53 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in a suitable combination of any of them. Alternatively, at least one of the first acquisition module 41, the recognition module 42, the second acquisition module 51, the labeling module 52 and the training module 53 may be at least partially implemented as a computer program module, which may perform corresponding functions when executed.

[0156] Fig.10 A block diagram of an electronic device suitable for implementing the method for identifying individual business owners or the method for constructing an individual business owner identification model according to an embodiment of the present invention is schematically shown.

[0157] like Fig.10 As shown, the electronic device 900 according to an embodiment of the present invention includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 to a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0158] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present invention by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present invention by executing the program stored in the one or more memories.

[0159] According to an embodiment of the present invention, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the input / output (I / O) interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage portion 908 as needed.

[0160] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0161] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0162] The embodiment of the present invention also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present invention.

[0163] The computer program executes the above functions defined in the system / device of the embodiment of the present invention when it is executed by the processor 901. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0164] In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0165] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present invention are performed. According to the embodiment of the present invention, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0166] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0167] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0168] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present invention may be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features described in the various embodiments of the present invention may be combined and / or combined in various ways. All of these combinations and / or combinations fall within the scope of the present invention.

[0169] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A method for identifying individual business owners, characterized in that: The method comprises: Acquire first transaction relationship dynamic graph data, wherein the first transaction relationship dynamic graph data includes T first snapshot graphs corresponding one-to-one to T consecutive time windows, wherein the T first snapshot graphs all have n first users as nodes, and each first snapshot graph includes m types of attribute edges corresponding one-to-one to m types of transaction relationships, wherein in each first snapshot graph, a corresponding type of attribute edge is set for each of the m types of transaction relationships between two first users within the time window corresponding to the first snapshot graph, and m, n, and T are all integers greater than 1; and Inputting the first transaction relationship dynamic graph data into a self-employed business identification model, and obtaining an identification result of the self-employed business identification model for each of the n first users, wherein the self-employed business identification model is a trained machine learning model for classifying users between ordinary users and self-employed businesses; The process of the individual business identification model processing the first transaction relationship dynamic graph data includes: Based on each attribute edge in the first snapshot graph, graph data processing is performed respectively to obtain a first node matrix corresponding to each transaction relationship; Merging the m first node matrices corresponding to each first snapshot graph to obtain a first transaction space feature matrix corresponding to each first snapshot graph; and The recognition result is obtained based on the time series feature processing of the T first transaction space feature matrices corresponding to the T first snapshot images.

2. The method according to claim 1, characterized in that The individual business identification model includes m neighborhood node aggregation modules, wherein the neighborhood node aggregation module is constructed using a graph convolutional network, wherein the graph data is processed based on each attribute edge in the first snapshot graph, and a first node matrix corresponding to each transaction relationship is obtained, including: Extracting a first hierarchical snapshot graph including only one attribute edge from the first snapshot graph to obtain m first hierarchical snapshot graphs; and Each first hierarchical snapshot graph is input into a neighborhood node aggregation module, and the first node matrix output by the neighborhood node aggregation module after fusing the node neighborhood relations in the first hierarchical snapshot graph is obtained.

3. The method according to claim 1, characterized in that The individual business identification model includes a multi-relationship view fusion module, and the fusion of the m first node matrices corresponding to each first snapshot graph to obtain the first transaction space feature matrix corresponding to each first snapshot graph includes: The first transaction space feature matrix is ​​obtained by integrating the node representations of the corresponding nodes in the m first node matrices through the weight matrix trained in the multi-relation view fusion module.

4. The method according to claim 1, characterized in that The individual business identification model includes a time modeling layer and a node classification layer. The time modeling layer is constructed using a long short-term memory network. The identification result obtained based on the time series feature processing of the T first transaction space feature matrices corresponding to the T first snapshot graphs includes: Combining the T first transaction space feature matrices into a first input sequence according to the order of the time windows corresponding to the T first snapshot graphs; Inputting the first input sequence into a time modeling layer, and obtaining a first spatiotemporal feature matrix output by the time modeling layer; Inputting the node representation of each node in the first spatiotemporal feature matrix to a node classification layer, and obtaining the classification result of each node output by each node classification layer; and According to the classification result of each node, the identification result of the first user corresponding to each node is obtained.

5. The method according to claim 1, characterized in that Each first snapshot graph also includes node initial features of each node, wherein the node initial features of each node include information on transaction behavior of a first user corresponding to the node within a time window corresponding to the first snapshot graph.

6. The method according to claim 1, characterized in that The obtaining of the first transaction relationship dynamic graph data comprises: In the case of obtaining authorization from n first users to use their transaction data, obtaining first historical transaction data generated by the n first users in a first historical period, where the first historical period is divided into T time windows; Identifying a transaction relationship contained in each data of the first historical transaction data, wherein the first historical transaction data includes m types of transaction relationships, and the transaction relationship contained in each data is one of the m types of transaction relationships; Divide the first historical transaction data into T first sub-data sets according to T time windows; A first snapshot graph is obtained based on each first sub-data set, wherein each first user is mapped to a node in the first snapshot graph, and data containing the same transaction relationship between two first users is merged and mapped to an attribute edge in the first snapshot graph; Based on T first snapshot graphs, the first transaction relationship dynamic graph data is obtained.

7. The method according to claim 6, characterized in that The identifying of the transaction relationship contained in each data of the first historical transaction data includes: The natural language processing technology is applied to process the transaction remarks of each data of the first historical transaction data to identify the transaction relationship contained in each data.

8. The method according to claim 1, characterized in that The training process of the individual business household identification model is as follows: Acquire second transaction relationship dynamic graph data, wherein the second transaction relationship dynamic graph data includes T second snapshot graphs corresponding to T consecutive time windows, wherein the T second snapshot graphs each have s second users as nodes, and each second snapshot graph includes the m types of attribute edges, wherein in each second snapshot graph, a corresponding type of attribute edge is set for each of the m types of transaction relationships between two second users within the time window corresponding to the second snapshot graph, and s is an integer greater than 1; Using the identity tag of each second user to mark each second user in the second transaction relationship dynamic graph data, wherein the identity tag is a common user or an individual business operator; Using the annotated second transaction relationship dynamic graph data to train the individual business household identification model; The training of the individual business owner identification model includes repeatedly performing the following operations until a model training termination condition is met: Based on each attribute edge in the second snapshot graph, graph data processing is performed respectively to obtain a second node matrix corresponding to each transaction relationship; Merging the m second node matrices corresponding to each second snapshot graph to obtain a second transaction space feature matrix corresponding to each second snapshot graph; and Obtaining a model recognition result based on time series feature processing of the T second transaction space feature matrices corresponding to the T second snapshot images; and Based on the difference between the model recognition result and the identity label of the second user in the second transaction relationship dynamic graph data, the parameters of the individual business owner recognition model are adjusted.

9. The method according to claim 8, characterized in that The performing graph data processing on each attribute edge in the second snapshot graph to obtain a second node matrix corresponding to each transaction relationship also includes: Sampling nodes in the second snapshot graph based on an unbalanced strategy; Constructing a subgraph using the nodes selected by sampling; and Graph data processing is performed based on each attribute edge in the subgraph to obtain a second node matrix corresponding to each transaction relationship.

10. The method according to claim 9, characterized in that The sampling probability of each node in the unbalanced strategy is positively correlated with the complexity of the connection relationship between the node and other nodes in the second snapshot graph, and is negatively correlated with the proportion of the identity tag of the second user corresponding to the node in the identity tags of s second users, wherein the complexity of the connection relationship between a node and other nodes is obtained based on the number and type of attribute edges between the node and other nodes.

11. The method according to claim 8, characterized in that Each second snapshot graph also includes node initial features of each second node, wherein the node initial features of the second node include information on transaction behavior of a second user corresponding to the second node within a time window corresponding to the second snapshot graph.

12. A method for constructing a self-employed business identification model, characterized in that: The construction method comprises: Acquire second transaction relationship dynamic graph data, wherein the second transaction relationship dynamic graph data includes T second snapshot graphs corresponding to T time windows one by one, wherein the T second snapshot graphs all have s second users as nodes, and each second snapshot graph includes m types of attribute edges corresponding to m types of transaction relationships one by one, wherein in each second snapshot graph, a corresponding type of attribute edge is set for each of the m types of transaction relationships existing between two second users within the time window corresponding to the second snapshot graph, and m, s, and T are all integers greater than 1; Using the identity tag of each second user to mark each second user in the second transaction relationship dynamic graph data, wherein the identity tag is a common user or an individual business operator; Using the annotated second transaction relationship dynamic graph data to train the individual business household identification model; The training of the individual business owner identification model includes repeatedly performing the following operations until a model training termination condition is met: Based on each attribute edge in the second snapshot graph, graph data processing is performed respectively to obtain a second node matrix corresponding to each transaction relationship; The m second node matrices corresponding to each second snapshot graph are merged to obtain a second transaction space feature matrix corresponding to each second snapshot graph; Obtaining a model recognition result based on the time series characteristics of the T second transaction space feature matrices corresponding to the T second snapshot images; and Based on the difference between the model recognition result and the identity label of the second user in the second transaction relationship dynamic graph data, the parameters of the individual business owner recognition model are adjusted.

13. A device for identifying individual business owners, characterized in that: The device comprises: A first acquisition module is used to acquire first transaction relationship dynamic graph data, wherein the first transaction relationship dynamic graph data includes T first snapshot graphs corresponding to T consecutive time windows, wherein the T first snapshot graphs all have n first users as nodes, and each first snapshot graph includes m types of attribute edges corresponding to m types of transaction relationships, wherein in each first snapshot graph, a corresponding type of attribute edge is set for each of the m types of transaction relationships between two first users in the time window corresponding to the first snapshot graph, and m, n, and T are all integers greater than 1; and an identification module, comprising an individual business identification model, the identification module being used to input the first transaction relationship dynamic graph data into the individual business identification model, and obtain an identification result of the individual business identification model for each of the n first users, the individual business identification model being a trained machine learning model for classifying users between ordinary users and individual business owners; The individual business identification model includes a spatial modeling layer, a temporal modeling layer and a node classification layer, wherein: The spatial modeling layer is used to: perform graph data processing based on each attribute edge in the first snapshot graph to obtain a first node matrix corresponding to each transaction relationship; and fuse the m first node matrices corresponding to each first snapshot graph to obtain a first transaction space feature matrix corresponding to each first snapshot graph; The time modeling layer is used to: process the time series features of the T first transaction space feature matrices corresponding to the T first snapshot graphs; and The node classification layer is used to process the output of the time modeling layer to obtain the recognition result.

14. A device for constructing a self-employed business identification model, characterized in that: The construction device comprises: A second acquisition module is used to acquire second transaction relationship dynamic graph data, wherein the second transaction relationship dynamic graph data includes T second snapshot graphs corresponding to T time windows one by one, wherein the T second snapshot graphs all have s second users as nodes, and each second snapshot graph includes m types of attribute edges corresponding to m types of transaction relationships one by one, wherein in each second snapshot graph, for each of the m types of transaction relationships existing between two second users in the time window corresponding to the second snapshot graph, a corresponding type of attribute edge is set respectively, and m, s, and T are all integers greater than 1; a labeling module, configured to label each second user in the second transaction relationship dynamic graph data with an identity tag of each second user, wherein the identity tag is a common user or an individual business operator; A training module, used to train the individual business household identification model using the annotated second transaction relationship dynamic graph data; The training module is used to repeatedly perform the following operations until the model training termination condition is met: Based on each attribute edge in the second snapshot graph, graph data processing is performed respectively to obtain a second node matrix corresponding to each transaction relationship; The m second node matrices corresponding to each second snapshot graph are merged to obtain a second transaction space feature matrix corresponding to each second snapshot graph; Obtaining a model recognition result based on the time series characteristics of the T second transaction space feature matrices corresponding to the T second snapshot images; and Based on the difference between the model recognition result and the identity label of the second user in the second transaction relationship dynamic graph data, the parameters of the individual business owner recognition model are adjusted.

15. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 12.

16. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

17. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images