A model processing method, apparatus and device

Abstract

This specification discloses a model processing method, apparatus, and device. The method includes: constructing a semantic tree for a target business; constructing a business graph corresponding to the target business based on the semantic tree, wherein the business graph consists of nodes and edges; acquiring historical business data generated in the target business; determining a first business subgraph corresponding to the historical business data based on the historical business data; and generating a second business subgraph based on the first business subgraph using preset data augmentation rules, wherein the first and second business subgraphs are subgraphs contained in the business graph respectively; and training a graph structure model based on the business graph corresponding to the target business, using the first and second business subgraphs, and through a contrastive learning method that compares each business subgraph in the first and second business subgraphs with each other, to obtain a trained graph structure model.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to a method, apparatus and device for processing a model. Background Technology

[0002] As data becomes increasingly abundant, a large amount of data is stored in the form of graph structures. The advantage of graph structure storage lies in its ability to highlight relationships between data points, thereby enriching data dimensions and enabling the discovery of high-level, effective features. Furthermore, even with a large amount of data, much of it is unlabeled. Therefore, deep learning models are needed to train and extract the inherent characteristics of the data, providing more convenient support for downstream tasks. However, the same data often has multiple downstream tasks. If each task requires training a deep learning model, it will consume significant computational resources. Therefore, contrastive learning can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data. Considering these factors in contrastive learning mechanisms is challenging. Therefore, a superior contrastive learning technique based on semantic information is needed to improve its performance. Summary of the Invention

[0003] The purpose of the embodiments in this specification is to provide a better contrastive learning technology solution based on semantic information, thereby improving the performance of contrastive learning.

[0004] To achieve the above technical solution, the embodiments in this specification are implemented as follows:

[0005] This specification provides a model processing method, comprising: constructing a semantic tree for a target business; constructing a business graph corresponding to the target business based on the semantic tree, the business graph consisting of nodes and edges; acquiring historical business data generated in the target business; determining a first business subgraph corresponding to the historical business data based on the historical business data; and generating a second business subgraph based on the first business subgraph using preset data augmentation rules, the first and second business subgraphs being subgraphs respectively contained within the business graph; and training a graph structure model based on the business graph corresponding to the target business, using the first and second business subgraphs corresponding to the historical business data, and through a contrastive learning method that compares each business subgraph in the first and second business subgraphs with each of the other business subgraphs, thereby obtaining a trained graph structure model.

[0006] This specification provides a model processing apparatus, comprising: a graph construction module, which constructs a semantic tree for a target business, and constructs a business graph corresponding to the target business based on the semantic tree, the business graph consisting of nodes and edges; a subgraph construction module, which acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph using preset data augmentation rules, the first business subgraph and the second business subgraph being subgraphs respectively contained in the business graph; and a model training module, which, based on the business graph corresponding to the target business, uses the first business subgraph and the second business subgraph corresponding to the historical business data, and trains the graph structure model using a contrastive learning method that compares each business subgraph in the first and second business subgraphs with each of the business subgraphs, to obtain a trained graph structure model.

[0007] This specification provides an embodiment of a model processing device, comprising: a processor; and a memory arranged to store computer-executable instructions, wherein the executable instructions, when executed, cause the processor to: construct a semantic tree for a target service; construct a service graph corresponding to the target service based on the semantic tree, the service graph consisting of nodes and edges; acquire historical service data generated in the target service; determine a first service subgraph corresponding to the historical service data based on the historical service data; and generate a second service subgraph based on the first service subgraph using preset data augmentation rules, wherein the first and second service subgraphs are subgraphs respectively contained in the service graph; and train a graph structure model based on the service graph corresponding to the target service, using the first and second service subgraphs corresponding to the historical service data, and through a contrastive learning method that compares each service subgraph in the first and second service subgraphs with each of the other service subgraphs, to obtain a trained graph structure model.

[0008] This specification also provides a storage medium for storing computer-executable instructions. When executed by a processor, these instructions implement the following process: constructing a semantic tree for a target service; constructing a service graph corresponding to the target service based on the semantic tree, the service graph consisting of nodes and edges; acquiring historical service data generated in the target service; determining a first service subgraph corresponding to the historical service data based on the historical service data; and generating a second service subgraph based on the first service subgraph using preset data augmentation rules, the first and second service subgraphs being subgraphs respectively contained within the service graph; and training a graph structure model based on the service graph corresponding to the target service, using the first and second service subgraphs corresponding to the historical service data, and through a contrastive learning method that compares each service subgraph in the first and second service subgraphs with each of the other service subgraphs, thereby obtaining a trained graph structure model. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram illustrating one method of comparative learning as described in this manual;

[0011] Figure 2 This is an embodiment of a model processing method described in this specification;

[0012] Figure 3A This is a schematic diagram of a semantic tree structure as described in this specification;

[0013] Figure 3B This is a schematic diagram of another semantic tree structure in this specification;

[0014] Figure 4A This is a schematic diagram of the structure of a business map as described in this specification;

[0015] Figure 4B This is a schematic diagram of another business map structure in this specification;

[0016] Figure 5A This is a structural diagram of a first business sub-diagram in this specification;

[0017] Figure 5B This is a structural diagram of a second business sub-diagram in this specification;

[0018] Figure 6 This is an embodiment of another model processing method described in this specification;

[0019] Figure 7 This is a schematic diagram illustrating the processing procedure of one model in this specification;

[0020] Figure 8 This is yet another embodiment of the model processing method described in this specification;

[0021] Figure 9 This is an embodiment of a processing device for a model described in this specification;

[0022] Figure 10 This is an embodiment of a processing device for one model in this specification. Detailed Implementation

[0023] This specification provides a method, apparatus, and device for processing a model.

[0024] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0025] This specification provides an embodiment of a contrastive learning mechanism for spectra. For contrastive learning, such as... Figure 1 As shown, U1, U2, U3, and U4 are four original samples. For each original sample, a corresponding positive sample can be generated by the encoder in contrastive learning, thereby constructing corresponding positive sample pairs (e.g., ...). Figure 1(The two small circles in the right-hand box). In typical contrastive learning mechanisms, the similarity between any positive sample pair corresponding to any original sample in the same batch (i.e., the same batch) and the positive sample pairs corresponding to other original samples is 0. Alternatively, in the same batch (i.e., the same batch), the similarity between any positive sample pair corresponding to any original sample and the positive sample pairs corresponding to original samples of the same category is 1, and the similarity between any positive sample pair corresponding to the remaining original samples is 0. However, the above processing method will result in poor performance of the final contrastive learning, and the accuracy of the prediction of the trained model will be reduced. In addition, considering that graph structure data is mostly unlabeled data, and at the same time, there are often multiple downstream tasks for the same data, if each task... Training a deep learning model for each task would consume significant computational resources. Contrastive learning, however, can effectively serve as an upstream task for feature extraction, especially for graph-structured data. Unlike other image and NLP tasks, graph-structured data not only contains the attribute features of nodes themselves but also rich relational data. To improve the performance of contrastive learning, for each original sample in each batch, the similarity between each original sample's corresponding positive sample pair and each original sample's corresponding positive sample pair in the batch is calculated to determine the corresponding loss information. The similarity calculation can be based on semantic information, making it easier to measure the similarity between nodes. Specific processing details can be found in the following embodiments.

[0026] Example 1

[0027] like Figure 2 As shown in the embodiments of this specification, a method for processing a model is provided. The execution subject of this method can be a terminal device or a server, etc. The terminal device can be a mobile terminal device such as a mobile phone or tablet computer, a computer device such as a laptop or desktop computer, or an IoT device (specifically, a smartwatch, an in-vehicle device, etc.). The server can be a single server or a server cluster composed of multiple servers. The server can be a backend server for financial business or online shopping business, or a backend server for an application, etc. This embodiment uses a server as the execution subject for detailed description. For the case where the execution subject is a terminal device, please refer to the following server case processing, which will not be repeated here. The method may specifically include the following steps:

[0028] In step S202, a semantic tree for the target business is constructed, and a business graph corresponding to the target business is constructed based on the semantic tree. The business graph consists of nodes and edges.

[0029] The target business can be any business, such as risk identification for online transactions or information recommendation, etc. It can be specifically set according to actual circumstances, and this specification does not limit this in the embodiments. The semantic tree can be a tree-like data structure built based on the semantics of data related to the target business. Different semantic trees can be built for different target business-related data. For example, the semantic tree for risk identification of online transactions involving fraud risk can be as follows: Figure 3A As shown, the semantic tree for risk identification in online transactions targeting illegal transactions can be represented as follows: Figure 3B As shown above, Figure 3A The semantic tree and the above Figure 3B The semantic tree is clearly different. A business graph can be a graph constructed from graph-structured data, such as a knowledge graph or a graph. A business graph can be constructed from nodes and edges. Nodes can be predefined entities in the business data, and edges can be used to represent the relationship between two nodes.

[0030] In practice, as data becomes increasingly abundant, a large amount of data is stored in the form of graph-structured data. The advantage of graph-structured data storage lies in its ability to highlight the relationships between data points, thereby enriching data dimensions and enabling the discovery of high-level, effective features. Furthermore, even with a large amount of data, much of it is unlabeled. Therefore, deep learning models are needed to train and extract the inherent characteristics of the data, providing more convenient support for downstream tasks. At the same time, the same data often has multiple downstream tasks. If each task requires training a deep learning model, it will consume significant computational resources. Therefore, contrastive learning can effectively serve as an upstream task for feature extraction, especially for graph-structured data. Unlike other image and NLP tasks, graph-structured data not only contains the attribute features of nodes themselves but also rich relational data. Considering these factors in contrastive learning mechanisms is challenging. Therefore, a better semantic information-based contrastive learning technique is needed to improve its performance. This specification provides an achievable technical solution, which may include the following:

[0031] To better utilize the business data of the target business and considering the advantages of graph structure data, the business data can be transformed into graph architecture data. Based on this graph structure data, a graph structure model of the target business can be trained using comparative learning. Specifically, this can be achieved by combining information such as the nature, attributes, and characteristics of the target business and its data, and constructing a semantic tree for the target business using expert experience or specified rules, as described above. Figure 3A or Figure 3BAs shown. By constructing a semantic tree, we can define the nodes contained in the semantic tree, as well as the attribute information of each node and the upstream and downstream relationships between different nodes, such as... Figure 3B In this system, for online transactions with risks of illegal transactions, nodes such as illegal transaction, user, transaction target, four-party platform, and resource transfer platform can be set. Attribute information for each node can be configured, such as user identifier, transaction target identifier, quantity of resources traded between the user and the transaction target, four-party platform identifier, role or purpose of the four-party platform, and resource transfer platform identifier. Upstream and downstream relationships between different nodes can also be set, specifically as follows: Figure 3B As shown. Since a semantic tree is a tree-like relational graph, in order for the graph structure model to be trained to better identify the relationships between nodes, a graph transformation mechanism can be pre-defined. Based on this mechanism, the constructed semantic tree can be converted into a corresponding graph. Nodes from the semantic tree can be retained in the business graph and used as nodes within the business graph. For example, for... Figure 3B The semantic tree, and the corresponding business graph constructed from it, can be like... Figure 4A As shown, where, Figure 3B All included nodes are in Figure 4A In practical applications, nodes in the semantic tree can also be used to construct nodes in the corresponding business graph. For example, nodes can be... Figure 3B The nodes containing "illegal transactions" are removed, and the remaining nodes are used as nodes in the corresponding business graph. The constructed business graph can be as follows: Figure 4B As shown, the specific settings can be customized according to actual circumstances. Using the above method, a business graph corresponding to the target business can be obtained.

[0032] It should be noted that the semantic tree constructed above is a complete semantic tree for the target business, that is, the semantic tree contains all the nodes that may appear in the target business and the attribute information of the nodes. Similarly, the business graph constructed above is a complete graph for the target business, that is, the business graph contains all the nodes that may appear in the target business, the attribute information of the nodes, and the relationship information between the nodes.

[0033] In step S204, historical business data generated in the target business is obtained. Based on the historical business data, a first business subgraph corresponding to the historical business data is determined. Based on the first business subgraph, a second business subgraph is generated through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs contained in the business graph, respectively.

[0034] Historical business data can be set according to different target businesses. For example, if the target business is risk identification for online transactions, the historical business data may include the accounts of both parties, the transaction location, the transaction time, the transaction amount, the product information, and the delivery method (e.g., express delivery or self-pickup). If the target business is information recommendation, the historical business data may include the location, time, content, type, and source of the recommended information. The specific details can be set according to the actual situation, and this embodiment does not limit this. Data augmentation rules can include various types. For example, they may include rules for pruning nodes and / or edges in the first business subgraph, or rules for performing specified deformation processing on the first business subgraph. The specific details can be set according to the actual situation, and this embodiment does not limit this.

[0035] In implementation, when a user performs a target service, the business data generated during the performance can be recorded. When the business data of the target service is needed, the recorded business data can be retrieved to obtain historical business data. In practical applications, business data generated within a specified time period can be selected from the recorded business data as the aforementioned historical business data. Alternatively, business data generated within a specified time period can be retrieved from the database used to store business data generated during the performance of the target service, and the retrieved business data can be used as the aforementioned historical business data. In addition to obtaining historical business data generated in the target service through the above methods, various other methods can also be used to obtain historical business data, which can be set according to the actual situation.

[0036] Each historical business data point can be analyzed. Based on the analysis results, key information contained in the historical business data can be determined. Then, a corresponding graph can be constructed based on this key information. The nodes in this graph are contained within the nodes of the aforementioned business graph, and the edges in this graph are also contained within the edges of the aforementioned business graph. This graph can be used as the first business subgraph corresponding to the historical business data. Based on a contrastive learning mechanism, using the first business subgraph as the base graph, a second business subgraph can be generated through preset data augmentation rules. Specifically, for example, using the aforementioned... Figure 4A For example, the first business sub-graph can be as follows: Figure 5A As shown, this includes two nodes: illegal transactions and users. Based on the first business subgraph mentioned above, the second business subgraph generated through preset data augmentation rules can be as follows: Figure 5B As shown, it includes three nodes: illegal transaction, transaction target party, and resource transfer platform, and corresponding edges. The first business subgraph is a subgraph contained in the business graph, and the second business subgraph is a subgraph contained in the business graph.

[0037] In the above way, each historical business data will correspond to one

[0038] In step S206, based on the business graph corresponding to the target business, the first business subgraph and the second business subgraph corresponding to the historical business data are used, and the graph structure model is trained by a comparative learning method that compares each business subgraph in the first business subgraph and the second business subgraph with each other, so as to obtain the trained graph structure model.

[0039] The graph structure model can include various types, such as a graph structure model that can be constructed using a GNN model. The specific model can be set according to the actual situation, and the embodiments in this specification do not limit it.

[0040] In implementation, through the above processing, each historical business data corresponds to a specific business subgraph. Simultaneously, new business data and corresponding business subgraphs can be generated through data augmentation rules. Thus, the business subgraph (or its corresponding semantic subtree) corresponding to the historical business data can be used as weak labeling information for the historical business data. Similarly, the business subgraph (or its corresponding semantic subtree) corresponding to the new business data generated through data augmentation rules can be used as weak labeling information for the new business data. Since the business subgraph already contains the key data from the corresponding business data, it can be directly used as input data into the graph structure model, ignoring the actual business data. In practical applications, business data and its corresponding business subgraph can also be input into the graph structure model for training.

[0041] Specifically, the first business subgraph can be input into the graph structure model to determine its corresponding representation information (referred to as the first representation information for ease of description). Similarly, the second business subgraph can be input into the graph structure model to determine its corresponding representation information (referred to as the second representation information for ease of description). A preset similarity algorithm can be used to calculate the similarity between the first and second representation information. Based on this similarity, the similarity of each node can be determined. Based on the obtained similarity, a preset loss function can be used to calculate the corresponding loss information. The model parameters of the graph structure model can then be adjusted based on this loss information. In addition, sample pairs can be formed using the first business subgraph and other business subgraphs determined by historical business data. These business subgraphs can be input into the graph structure model to determine their corresponding representation information. Then, the similarity is obtained by combining this with the business graph corresponding to the target business. A preset loss function is used to calculate the corresponding loss information, which can be used to adjust the model parameters of the graph structure model. Thus, for historical business data in a batch, each business subgraph in the first and second business subgraphs can be compared with each other (i.e., similarity calculation is performed). Then, it is determined whether the graph structure model has converged. If it has not converged, historical business data can be acquired, and the graph structure model can be trained using steps S204 and S206. By training the graph structure model through contrastive learning, a trained graph structure model can be obtained.

[0042] This specification provides a model processing method through embodiments. It involves constructing a semantic tree for a target business, building a business graph corresponding to the target business based on the semantic tree (the business graph consists of nodes and edges), acquiring historical business data generated in the target business, determining a first business subgraph based on the historical business data, and generating a second business subgraph based on the first business subgraph using preset data augmentation rules. The first and second business subgraphs are subgraphs contained within the business graph, respectively. Finally, based on the business graph corresponding to the target business, the first and second business subgraphs are used, and each business subgraph in the first and second business subgraphs is compared with each other. The contrastive learning approach trains a graph structure model, resulting in a trained graph structure model. A semantic tree is then defined to describe the hierarchical relationships of node attributes, allowing for the comparison of a node's own attributes with those of other nodes. Furthermore, graph structure similarity is used to determine the similarity between nodes. This semantic-based contrastive learning approach can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data, which is fully utilized in contrastive learning, reducing computational resource consumption and improving its performance.

[0043] Example 2

[0044] like Figure 6 As shown in the embodiments of this specification, a method for processing a model is provided. The execution subject of this method can be a terminal device or a server, etc. The terminal device can be a mobile terminal device such as a mobile phone or tablet computer, a computer device such as a laptop or desktop computer, or an IoT device (specifically, a smartwatch, an in-vehicle device, etc.). The server can be a single server or a server cluster composed of multiple servers. The server can be a backend server for financial business or online shopping business, or a backend server for an application, etc. This embodiment uses a server as the execution subject for detailed description. For the case where the execution subject is a terminal device, please refer to the following server case processing, which will not be repeated here. The method may specifically include the following steps:

[0045] In step S602, a semantic tree for the target business is constructed, and a business graph corresponding to the target business is constructed based on the semantic tree. The business graph consists of nodes and edges.

[0046] In step S604, historical business data generated in the target business is obtained. Based on the historical business data, a first business subgraph corresponding to the historical business data is determined. Based on the first business subgraph, a second business subgraph is generated through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs contained in the business graph, respectively.

[0047] The data augmentation rules may include one or more of the following: rules for flipping the first business subgraph, rules for shifting nodes in the first business subgraph, rules for pruning nodes and / or edges in the first business subgraph, and rules for deforming the first business subgraph.

[0048] In step S606, based on the business graph corresponding to the target business, the first business subgraph and the second business subgraph corresponding to the historical business data are used. The graph structure model is trained by comparing each business subgraph in the first business subgraph and the second business subgraph with each other using a contrastive learning method. The graph structure model is trained using a preset loss function of the contrastive learning method. The preset loss function includes the InfoNCE loss function.

[0049] Among them, such as Figure 7 As shown, the graph structure model can be a model constructed from a graph neural network (GNN) and a multilayer perceptron (MLP). The preset loss function can be constructed based on the first business subgraph, the second business subgraph, the first business subgraph, and the second business subgraph, as well as preset hyperparameters and a similarity matrix based on comparing each business subgraph in the first and second business subgraphs with each other. The preset loss function (i.e., the InfoNCE loss function) can be seen in the following formula.

[0050]

[0051] Where q is a business subgraph, k + Let q be the positive sample corresponding to q. i Let q represent the positive and negative sample pairs, τ be a hyperparameter, k represent the number of data points in the positive and negative sample pairs, and TopN(S) represent the N top-ranked samples selected from the calculated similarity scores based on the similarity matrix S, where S = α*S f +(1-α)*S g S f S represents feature similarity. g This represents the similarity between business subgraphs, where α is a hyperparameter.

[0052] The similarity matrix can be calculated using the following formula.

[0053]

[0054] Where Sim represents the similarity matrix, and A and B represent the sets of semantic subtrees corresponding to different business subgraphs. For example, the semantic subtree of historical business data 1 is {illegal transaction, user}, and the semantic subtree of the business data generated by data augmentation rules is {illegal transaction, transaction target party, resource transfer platform}. Then the similarity matrix between the two can be as follows:

[0055]

[0056] In implementation, such as Figure 7 As shown, the first business subgraph and the second business subgraph can be input into the graph structure model respectively. The graph structure model is trained by comparing each business subgraph in the first and second business subgraphs with each other, using the loss function of the above-mentioned comparative learning method.

[0057] For cases where the graph structure model is used to identify accounts with preset risks in historical transactions, after obtaining the trained graph structure model through the above method, the trained graph structure model can be used for risk identification processing. For details, please refer to the processing of steps S608 to S612 below.

[0058] In step S608, historical transaction data generated within a preset period is obtained.

[0059] The preset period can be set according to actual conditions, such as 7 days or 1 month. Historical transaction data may include the accounts of both parties, transaction location, transaction time, transaction amount, product information, and delivery method (such as express delivery or self-pickup). The specific settings can be customized according to actual conditions; this embodiment does not limit this.

[0060] In step S610, based on historical transaction data, a target business subgraph corresponding to the historical transaction data is determined. The target business subgraph includes nodes constructed from account information and edges constructed from transaction relationships between different accounts.

[0061] In step S612, the target business subgraph is input into the trained graph structure model to obtain the accounts with preset risks in the historical transaction data.

[0062] In implementation, the preset risks can include various types, such as fraud risk or illegal transaction risk, which can be set according to the actual situation. The target business subgraph can be input into the trained graph structure model to obtain accounts with preset risks in historical transaction data. This can result in either a single account with a preset risk or a set of accounts with multiple preset risks, which can include the relationships between different accounts.

[0063] This specification provides a model processing method through embodiments. It involves constructing a semantic tree for a target business, building a business graph corresponding to the target business based on the semantic tree (the business graph consists of nodes and edges), acquiring historical business data generated in the target business, determining a first business subgraph based on the historical business data, and generating a second business subgraph based on the first business subgraph using preset data augmentation rules. The first and second business subgraphs are subgraphs contained within the business graph, respectively. Finally, based on the business graph corresponding to the target business, the first and second business subgraphs are used, and each business subgraph in the first and second business subgraphs is compared with each other. The contrastive learning approach trains a graph structure model, resulting in a trained graph structure model. A semantic tree is then defined to describe the hierarchical relationships of node attributes, allowing for the comparison of a node's own attributes with those of other nodes. Furthermore, graph structure similarity is used to determine the similarity between nodes. This semantic-based contrastive learning approach can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data, which is fully utilized in contrastive learning, reducing computational resource consumption and improving its performance.

[0064] Example 3

[0065] The following describes in detail the processing method of the model provided in the embodiments of this specification, with specific application scenarios. The target business is the risk control business of transactions. The graph structure model is used to identify the risk of a specified transaction in the current process or to identify accounts with preset risks in historical transactions. In this embodiment, the graph structure model is used to identify accounts with preset risks in historical transactions as an example. The historical business data in the above embodiment can be represented by first historical transaction data and second historical transaction data. Both the first historical transaction data and the second historical transaction data can represent relevant data of transactions between different accounts, which may include user behavior information, account information of the two parties to the transaction, transaction amount, transaction time and transaction location, etc.

[0066] like Figure 8 As shown in the embodiments of this specification, a method for processing a model is provided. The execution subject of this method can be a terminal device or a server, etc. The terminal device can be a mobile terminal device such as a mobile phone or tablet computer, a computer device such as a laptop or desktop computer, or an IoT device (specifically, a smartwatch, an in-vehicle device, etc.). The server can be a single server or a server cluster composed of multiple servers. The server can be a backend server for financial business or online shopping business, or a backend server for an application, etc. This embodiment uses a server as the execution subject for detailed description. For the case where the execution subject is a terminal device, please refer to the following server case processing, which will not be repeated here. The method may specifically include the following steps:

[0067] In step S802, a semantic tree for risk prevention and control services of transactions is constructed, and a business graph corresponding to the risk prevention and control services of transactions is constructed based on the semantic tree. The business graph consists of nodes and edges.

[0068] In step S804, first historical transaction data is obtained, and based on the first historical transaction data, a first business subgraph corresponding to the first historical transaction data is determined. Based on the first business subgraph, a second business subgraph is generated through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs contained in the business graph, respectively.

[0069] In step S806, based on the above business graph, the first business subgraph and the second business subgraph corresponding to the first historical transaction data are used. The graph structure model is trained by using the InfoNCE loss function through a comparative learning method that compares each business subgraph in the first and second business subgraphs with each other, and the trained graph structure model is obtained.

[0070] The graph structure model can be a model constructed from graph neural networks and multilayer perceptrons. The preset loss function is constructed based on the first business subgraph, the second business subgraph, the first business subgraph, and the second business subgraph, as well as preset hyperparameters and a similarity matrix constructed by comparing each business subgraph in the first and second business subgraphs with each other.

[0071] In step S808, the second historical transaction data generated within a preset period is obtained.

[0072] In step S810, based on the second historical transaction data, a target business subgraph corresponding to the second historical transaction data is determined. The target business subgraph includes nodes constructed from account information and edges constructed from transaction relationships between different accounts.

[0073] In step S812, the target business subgraph is input into the trained graph structure model to obtain the accounts with preset risks in the second historical transaction data.

[0074] This specification provides a model processing method through embodiments. It involves constructing a semantic tree for a target business, building a business graph corresponding to the target business based on the semantic tree (the business graph consists of nodes and edges), acquiring historical business data generated in the target business, determining a first business subgraph based on the historical business data, and generating a second business subgraph based on the first business subgraph using preset data augmentation rules. The first and second business subgraphs are subgraphs contained within the business graph, respectively. Finally, based on the business graph corresponding to the target business, the first and second business subgraphs are used, and each business subgraph in the first and second business subgraphs is compared with each other. The contrastive learning approach trains a graph structure model, resulting in a trained graph structure model. A semantic tree is then defined to describe the hierarchical relationships of node attributes, allowing for the comparison of a node's own attributes with those of other nodes. Furthermore, graph structure similarity is used to determine the similarity between nodes. This semantic-based contrastive learning approach can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data, which is fully utilized in contrastive learning, reducing computational resource consumption and improving its performance.

[0075] Example 4

[0076] The above describes the model processing method provided in the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide a model processing device, such as... Figure 9 As shown.

[0077] The processing unit of the model includes: a graph construction module 901, a subgraph construction module 902, and a model training module 903, wherein:

[0078] Graph construction module 901 constructs a semantic tree for the target business, and constructs a business graph corresponding to the target business based on the semantic tree. The business graph consists of nodes and edges.

[0079] The subgraph construction module 902 acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs respectively contained in the business graph.

[0080] The model training module 903, based on the business graph corresponding to the target business, uses the first business subgraph and the second business subgraph corresponding to the historical business data, and trains the graph structure model by comparing each business subgraph in the first business subgraph and the second business subgraph with each business subgraph, to obtain the trained graph structure model.

[0081] In the embodiments described in this specification, the graph structure model is a model constructed from graph neural networks and multilayer perceptrons.

[0082] In the embodiments of this specification, the data augmentation rules include one or more of the following: rules for flipping the first service subgraph, rules for shifting nodes in the first service subgraph, rules for clipping nodes and / or edges in the first service subgraph, and rules for deforming the first service subgraph.

[0083] In this embodiment of the specification, the model training module 903, based on the business graph corresponding to the target business, uses the first business subgraph and the second business subgraph corresponding to the historical business data, and trains the graph structure model using a comparative learning method that compares each business subgraph in the first and second business subgraphs with each of the business subgraphs, and employs a preset loss function of the comparative learning method to obtain the trained graph structure model. The preset loss function includes the InfoNCE loss function.

[0084] In the embodiments of this specification, the preset loss function is constructed based on the first service subgraph, the second service subgraph, the first service subgraph and the second service subgraph, as well as preset hyperparameters and a similarity matrix based on comparing each service subgraph in the first service subgraph and the second service subgraph with each of the service subgraphs.

[0085] In this embodiment of the specification, the similarity matrix is ​​constructed from the similarity scores of each business subgraph in the first and second business subgraphs compared with each other, based on similarity scores greater than a similarity threshold.

[0086] In this embodiment of the specification, the target business is the risk control business of transactions, and the graph structure model is used to identify risks in the current specified transactions or to identify accounts with preset risks in historical transactions.

[0087] In the embodiments described in this specification, the device further includes:

[0088] The transaction data acquisition module acquires historical transaction data generated within a preset period.

[0089] The target subgraph construction module determines the target business subgraph corresponding to the historical transaction data based on the historical transaction data. The target business subgraph includes nodes constructed from account information and edges constructed from transaction relationships between different accounts.

[0090] The risk account identification module inputs the target business subgraph into the trained graph structure model to obtain accounts with preset risks in the historical transaction data.

[0091] This specification provides a model processing apparatus that constructs a semantic tree for a target business, builds a business graph corresponding to the target business based on the semantic tree, and the business graph consists of nodes and edges. Then, it acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph using preset data augmentation rules. The first and second business subgraphs are subgraphs contained within the business graph, respectively. Finally, based on the business graph corresponding to the target business, it uses the first and second business subgraphs, and compares each business subgraph in the first and second business subgraphs with each other. The contrastive learning approach trains a graph structure model, resulting in a trained graph structure model. A semantic tree is then defined to describe the hierarchical relationships of node attributes, allowing for the comparison of a node's own attributes with those of other nodes. Furthermore, graph structure similarity is used to determine the similarity between nodes. This semantic-based contrastive learning approach can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data, which is fully utilized in contrastive learning, reducing computational resource consumption and improving its performance.

[0092] Example 5

[0093] The above describes the model processing apparatus provided in the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide a model processing device, such as... Figure 10 As shown.

[0094] The processing device for the model can be a terminal device or a server, as described in the above embodiments.

[0095] The processing device of the model can vary considerably due to differences in configuration or performance, and may include one or more processors 1001 and memory 1002. Memory 1002 may store one or more application programs or data. Memory 1002 may be temporary or persistent storage. The application programs stored in memory 1002 may include one or more modules (not shown), each module including a series of computer-executable instructions for the model's processing device. Furthermore, processor 1001 may be configured to communicate with memory 1002 and execute the series of computer-executable instructions in memory 1002 on the model's processing device. The model's processing device may also include one or more power supplies 1003, one or more wired or wireless network interfaces 1004, one or more input / output interfaces 1005, and one or more keyboards 1006.

[0096] Specifically, in this embodiment, the processing device for the model includes a memory and one or more programs, wherein one or more programs are stored in the memory, and one or more programs may include one or more modules, and each module may include a series of computer-executable instructions for the processing device of the model, and is configured to be executed by one or more processors. The one or more programs include computer-executable instructions for performing the following:

[0097] Construct a semantic tree for the target business, and construct a business graph corresponding to the target business based on the semantic tree. The business graph consists of nodes and edges.

[0098] Historical business data generated in the target business is obtained. Based on the historical business data, a first business subgraph corresponding to the historical business data is determined. Based on the first business subgraph, a second business subgraph is generated through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs respectively contained in the business graph.

[0099] Based on the business graph corresponding to the target business, the first business subgraph and the second business subgraph corresponding to the historical business data are used, and the graph structure model is trained by a comparative learning method that compares each business subgraph in the first business subgraph and the second business subgraph with each business subgraph, to obtain the trained graph structure model.

[0100] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for processing the model are relatively simple in description because they are fundamentally similar to the method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0101] This specification provides a model processing device that constructs a semantic tree for a target business, builds a business graph corresponding to the target business based on the semantic tree, and the business graph consists of nodes and edges. Then, it acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph using preset data augmentation rules. The first and second business subgraphs are subgraphs contained within the business graph, respectively. Finally, based on the business graph corresponding to the target business, it uses the first and second business subgraphs, and compares each business subgraph in the first and second business subgraphs with each other. The contrastive learning approach trains a graph structure model, resulting in a trained graph structure model. A semantic tree is then defined to describe the hierarchical relationships of node attributes, allowing for the comparison of a node's own attributes with those of other nodes. Furthermore, graph structure similarity is used to determine the similarity between nodes. This semantic-based contrastive learning approach can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data, which is fully utilized in contrastive learning, reducing computational resource consumption and improving its performance.

[0102] Example 6

[0103] Furthermore, based on the above Figures 2 to 8 The method shown in this specification, along with one or more embodiments, also provides a storage medium for storing computer-executable instruction information. In one specific embodiment, the storage medium can be a USB flash drive, optical disc, hard disk, etc. When the computer-executable instruction information stored in the storage medium is executed by a processor, it can achieve the following process:

[0104] Construct a semantic tree for the target business, and construct a business graph corresponding to the target business based on the semantic tree. The business graph consists of nodes and edges.

[0105] Historical business data generated in the target business is obtained. Based on the historical business data, a first business subgraph corresponding to the historical business data is determined. Based on the first business subgraph, a second business subgraph is generated through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs respectively contained in the business graph.

[0106] Based on the business graph corresponding to the target business, the first business subgraph and the second business subgraph corresponding to the historical business data are used, and the graph structure model is trained by a comparative learning method that compares each business subgraph in the first business subgraph and the second business subgraph with each business subgraph, to obtain the trained graph structure model.

[0107] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the above-described storage medium embodiment is basically similar to the method embodiment, so the description is relatively simple; relevant parts can be referred to the description of the method embodiment.

[0108] This specification provides a storage medium that constructs a semantic tree for a target service, builds a service graph corresponding to the target service based on the semantic tree, and the service graph consists of nodes and edges. Then, it acquires historical service data generated in the target service, determines a first service subgraph corresponding to the historical service data based on the historical service data, and generates a second service subgraph based on the first service subgraph using preset data augmentation rules. The first and second service subgraphs are subgraphs contained within the service graph, respectively. Finally, based on the service graph corresponding to the target service, the first and second service subgraphs are used, and each service subgraph in the first and second service subgraphs is compared with each other. The contrastive learning approach trains the graph structure model, resulting in a trained graph structure model. A semantic tree is defined to describe the hierarchical relationships of node attribute information, allowing for the comparison of a node's own attributes with those of other nodes. Furthermore, graph structure similarity is used to determine the similarity between nodes. This semantic-based contrastive learning approach can effectively serve as an upstream task for feature extraction, especially for graph structure data. Unlike other image and NLP tasks, graph structure data not only contains the attribute features of nodes themselves but also rich relational data, which is fully utilized in contrastive learning, reducing computational resource consumption and improving its performance.

[0109] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0110] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using a hardware physical module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program a digital system themselves to "integrate" it onto a PLD, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0111] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0112] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0113] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0114] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] Embodiments in this specification are described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable parallel device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable parallel device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0116] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable fraud device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0117] These computer program instructions can also be loaded onto a computer or other programmable device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0118] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0119] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0120] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0121] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0122] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] One or more embodiments of this specification can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a particular task or implement a particular abstract data type. One or more embodiments of this specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0124] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0125] The above description is merely an embodiment of this specification and is not intended to limit this application. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for processing a model, the method comprising: Construct a semantic tree for the target business, and construct a business graph corresponding to the target business based on the semantic tree. The business graph consists of nodes and edges, and the target business is risk identification for online transactions. The system acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph using preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs respectively contained in the business graph. The historical business data includes the accounts of both parties to the transaction, the transaction location, the transaction time, the transaction amount, the transaction product information, and the product handover method. Based on the business graph corresponding to the target business, the first business subgraph and the second business subgraph corresponding to the historical business data are used. A contrastive learning method is adopted, which compares each business subgraph in the first and second business subgraphs with each other. The graph structure model is trained using a preset loss function of the contrastive learning method to obtain the trained graph structure model. The preset loss function is constructed based on the first business subgraph, the second business subgraph, the first business subgraph, and the second business subgraph, as well as preset hyperparameters and a similarity matrix based on the comparison of each business subgraph in the first and second business subgraphs with each other. The preset loss function includes the InfoNCE loss function, which introduces the first N values ​​in the similarity matrix into the denominator of the InfoNCE loss function in the form of multiplicative coefficients.

2. The method according to claim 1, wherein the graph structure model is a model constructed by a graph neural network and a multilayer perceptron.

3. The method according to claim 2, wherein the data augmentation rules include one or more of the following: rules for flipping the first service subgraph, rules for shifting nodes in the first service subgraph, rules for clipping nodes and / or edges in the first service subgraph, and rules for deforming the first service subgraph.

4. The method according to claim 1, wherein the similarity matrix is ​​constructed from the similarity scores of each business subgraph in the first business subgraph and the second business subgraph compared with each other, based on similarity scores greater than a similarity threshold.

5. The method according to claim 4, wherein the target business is the risk control business of transactions, and the graph structure model is used to identify risks in the currently ongoing specified transactions or to identify accounts with preset risks in historical transactions.

6. The method according to claim 5, further comprising: Acquire historical transaction data generated within a preset period; Based on the historical transaction data, a target business subgraph corresponding to the historical transaction data is determined. The target business subgraph includes nodes constructed from account information and edges constructed from transaction relationships between different accounts. The target business subgraph is input into the trained graph structure model to obtain the accounts with preset risks in the historical transaction data.

7. A model processing apparatus, the apparatus comprising: The graph construction module constructs a semantic tree for the target business, and builds a business graph corresponding to the target business based on the semantic tree. The business graph consists of nodes and edges, and the target business is risk identification for online transactions. The subgraph construction module acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph through preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs respectively contained in the business graph. The historical business data includes the accounts of the two parties to the transaction, the transaction location, the transaction time, the transaction amount, the transaction product information, and the product handover method. The model training module, based on the business graph corresponding to the target business, uses the first business subgraph and the second business subgraph corresponding to the historical business data, and trains the graph structure model using a contrastive learning method that compares each business subgraph in the first and second business subgraphs with each other. The trained graph structure model is obtained by using a preset loss function of the contrastive learning method. The preset loss function is constructed based on the first business subgraph, the second business subgraph, the first business subgraph, and the second business subgraph, as well as preset hyperparameters and a similarity matrix that compares each business subgraph in the first and second business subgraphs with each other. The preset loss function includes the InfoNCE loss function, in which the first N values ​​in the similarity matrix are introduced into the denominator of the InfoNCE loss function as multiplicative coefficients.

8. A model processing device, the model processing device comprising: processor; as well as A memory configured to store computer-executable instructions, which, when executed, cause the processor to: Construct a semantic tree for the target business, and construct a business graph corresponding to the target business based on the semantic tree. The business graph consists of nodes and edges, and the target business is risk identification for online transactions. The system acquires historical business data generated in the target business, determines a first business subgraph corresponding to the historical business data based on the historical business data, and generates a second business subgraph based on the first business subgraph using preset data augmentation rules. The first business subgraph and the second business subgraph are subgraphs respectively contained in the business graph. The historical business data includes the accounts of both parties to the transaction, the transaction location, the transaction time, the transaction amount, the transaction product information, and the product handover method. Based on the business graph corresponding to the target business, the first business subgraph and the second business subgraph corresponding to the historical business data are used. A contrastive learning method is adopted, which compares each business subgraph in the first and second business subgraphs with each other. The graph structure model is trained using a preset loss function of the contrastive learning method to obtain the trained graph structure model. The preset loss function is constructed based on the first business subgraph, the second business subgraph, the first business subgraph, and the second business subgraph, as well as preset hyperparameters and a similarity matrix based on the comparison of each business subgraph in the first and second business subgraphs with each other. The preset loss function includes the InfoNCE loss function, which introduces the first N values ​​in the similarity matrix into the denominator of the InfoNCE loss function in the form of multiplicative coefficients.

Images