CN115545938B

CN115545938B - Method, device, storage medium and equipment for executing risk identification service

Info

Publication number: CN115545938B
Application number: CN202211508342.5A
Authority: CN
Inventors: 赵闻飙; 朱亮; 田胜; 但家旺; 李金膛; 孟昌华
Original assignee: Alipay Hangzhou Information Technology Co Ltd
Current assignee: Alipay Hangzhou Information Technology Co Ltd
Priority date: 2022-11-29
Filing date: 2022-11-29
Publication date: 2023-03-28
Anticipated expiration: 2042-11-29
Also published as: CN115545938A

Abstract

The specification discloses a method, a device, a storage medium and equipment for executing a risk identification service, wherein historical full graph data is disassembled into sub graph data, the weight of a main body in each sub graph data is determined through a trained explanation model, the weight of the main body in the historical full graph data is determined according to the weight of the main body in each sub graph data and is used as a label of the historical full graph data, and a pre-noise reduction model is trained according to the historical full graph data and the label of the historical full graph data. And when the updating condition is met, determining the current full-map data, inputting the current full-map data into the trained pre-noise reduction model to obtain the weight of each main body in the current full-map data, and cutting the main body in the current full-map data according to the obtained weight to obtain the credible map. The current full graph data can be cut based on the weight output by the pre-noise reduction model before the business is executed, the noise-reduced credible graph is obtained, and the risk identification result corresponding to the business data can be determined according to the credible graph when the wind control business request carrying the business data is received.

Description

Method, device, storage medium and equipment for executing risk identification service

Technical Field

The present disclosure relates to the field of data processing technologies, and in particular, to a method, an apparatus, a storage medium, and a device for executing a risk identification service.

Background

People pay more and more attention to private data at present. In some business scenarios, personal data of the user is typically input into the neural network to obtain the business result.

For example, in a wind-controlled service scenario, with users as nodes, edges are determined according to services performed between users, and graph data composed of the nodes and the edges is determined. Since graph data is composed of nodes and edges, the nodes and edges in the graph data are generally referred to as bodies in the graph data. And inputting the Graph data and the personal data of the target user into a Graph Neural Network (GNN) model to obtain a wind control result of the currently executed service of the target user. Or obtaining a wind control result based on the business rule according to the total graph data and the personal data of the target user.

In the conventional method, no matter through a model or a business rule, a business result is obtained according to a full amount of graph data when a business is executed.

However, how to determine graph data for executing a service is an urgent problem to be solved.

Disclosure of Invention

The present specification provides a method, an apparatus, a storage medium, and a device for performing risk identification service, so as to partially solve the above problems in the prior art.

The technical scheme adopted by the specification is as follows:

the present specification provides a method of performing a risk identification service, the method comprising:

splitting the historical full graph data into sub graph data;

determining the weight of each main body in each sub-graph data through a trained interpretation model, wherein the interpretation model takes the sub-graph data corresponding to the historical wind control service as a training sample, the service result of the historical wind control service is obtained by label training, and the weight represents the contribution degree of the main body serving as the input sub-graph data to the obtained labeled service result;

determining the weight of each main body in the historical full graph data according to the weight of each main body in each sub graph data, and using the weight as the label of the historical full graph data;

training a pre-noise reduction model according to the historical full map data and the label of the historical full map data; the pre-noise reduction model is used for outputting the weight of each main body in the full-image data to be subjected to noise reduction;

when the situation that the updating condition is met is determined, determining current full graph data, inputting the current full graph data into the trained pre-noise reduction model, and obtaining the weight of each main body in the current full graph data;

cutting out the main bodies in the current full graph data according to the weight of each main body in the current full graph data to obtain a credible graph;

and when a wind control service request carrying service data is received, identifying a risk identification result of a service corresponding to the service data according to the service data and the credible graph.

The present specification provides an apparatus for performing a risk identification service, the apparatus comprising:

the splitting module is used for splitting the historical full graph data into sub graph data;

the first determining module is used for determining the weight of each main body in each sub-graph data through a trained interpretation model, the interpretation model takes the sub-graph data corresponding to historical wind control service as a training sample, the service result of the historical wind control service is obtained by label training, and the weight represents the contribution degree of the main body of the sub-graph data as input to the obtained labeled service result;

the marking module is used for determining the weight of each main body in the historical full graph data according to the weight of each main body in each sub graph data, and the weight is used as the mark of the historical full graph data;

the first training module is used for training a pre-noise reduction model according to the historical full map data and the label of the historical full map data; the pre-noise reduction model is used for outputting the weight of each main body in the full graph data to be subjected to noise reduction;

the weighting module is used for determining current full graph data when the updating condition is met, inputting the current full graph data into the trained pre-noise reduction model, and obtaining the weight of each main body in the current full graph data;

the cutting module is used for cutting the main bodies in the current full graph data according to the weight of each main body in the current full graph data to obtain a credible graph;

and the execution module is used for identifying the risk identification result of the business corresponding to the business data according to the business data and the credible graph when receiving a wind control business request carrying the business data.

The present specification provides a computer readable storage medium storing a computer program which, when executed by a processor, implements the above-described method of performing a risk identification service.

The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the above method of performing a risk identification service when executing the program.

The technical scheme adopted by the specification can achieve the following beneficial effects:

in the method for executing the risk identification service, the historical full-graph data is divided into the sub-graph data, the weight of the main body in the sub-graph data is determined through the trained interpretation model, the weight of the main body in the historical full-graph data is determined according to the weight of the main body in the sub-graph data and is used as the label of the historical full-graph data, and the pre-noise reduction model is trained according to the historical full-graph data and the label of the historical full-graph data. And when the updating condition is met, determining the current full-map data, inputting the current full-map data into the trained pre-noise reduction model to obtain the weight of each main body in the current full-map data, and cutting the main body in the current full-map data according to the obtained weight to obtain the credible map.

As can be seen from the above, the method for executing a risk identification service provided in this specification may cut current full graph data based on a weight output by a pre-noise reduction model before executing a service to obtain a noise-reduced reliability graph, and may execute the service according to the reliability graph when receiving a wind-controlled service request carrying service data, where the calculation pressure is small when executing the service, the calculation resource waste is less, and a more accurate service result may be obtained when executing the service based on the reliability graph.

Drawings

The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:

FIG. 1 is a schematic flow chart of a method for performing a risk identification service in the present specification;

FIG. 2 is a schematic diagram of determining sub-graph data provided in this specification;

FIG. 3 is a schematic diagram of an apparatus for performing a risk identification service provided herein;

fig. 4 is a schematic diagram of an electronic device provided in this specification.

Detailed Description

To determine graph data for performing a risk identification service, the present specification provides a method of performing a risk identification service. Based on the method, a pre-noise reduction model for generating the credible graph can be trained. The reliability map is the map data obtained by reducing the noise of the entire map data and removing the noise. After the full amount of graph data (hereinafter referred to as full graph data) is input into the pre-noise reduction model, the weight of each main body (such as node or edge) in the full graph data can be determined. And (4) clipping the main bodies in the full graph data based on the weight of each main body in the full graph data, so that the credible graph can be determined. The reliability map can be used for executing downstream services, and for example, the service result of the downstream target service can be accurately output through the downstream GNN model according to the reliability map. Or, according to the credible graph and the business rules, the business result of the downstream target business can be accurately obtained.

The weight of the main body represents the contribution degree of the main body to the obtained service result, and based on the weight of the main body, part of the main body in the whole image data can be cut to realize noise reduction. For example, when the weight of a subject is 1, it may be determined that the subject is not noise, and the subject may be retained. When the weight of the subject is 0, it may be determined that the subject does not contribute to the service result, the subject is noise, and the subject may be cut.

Thus, the weight of one body can be used to represent one of the operation of retaining the body and the operation of clipping the body.

It should be noted that, when the risk identification service is executed according to the reliability graph and the service result is determined, the characteristics of the nodes or the characteristics of the edges in the reliability graph are relied on. For example, the nodes may be different users, and the edges may represent historical traffic, such as transactions, between the different users. The characteristics of the edge can be transaction data and the like, and the characteristics of the node can be user data, such as basic information of the user's age and the like, historical accumulated transaction amount, complained times and the like.

Due to the particularity and importance of the wind control service, the wind control service executed based on the credible graph can be applied to various service scenes, such as wind control for payment service, transfer service, recommendation service and the like. That is, the wind control service can be specifically classified into a type of wind control service that performs wind control on a payment service, a type of wind control on a ledger service, a type of wind control on a recommendation service, and the like.

Of course, the wind control service may be a separate service (hereinafter referred to as a pure wind control service) specially controlled by the user, that is, a service that is not executed at the time of payment, transfer, recommendation to the user, and the like. For example, the method can be a wind control service executed for the user after the user is complained, or a wind control service performed periodically. And the anti-money laundering service can be used as a special wind control service.

In a scenario of performing wind control on a recommendation service, the nodes may include two types: users and goods. The side between the user and the product represents the purchase record of the user and the product, and the side is characterized by order data corresponding to the purchase record and the like. Of course, the edges between users represent the associative relationship between users, for example, the characteristics of the edges may include which same items the users purchase, and the like. The edges between the commodities represent the association relationship between the commodities, for example, the characteristics of the edges may include the commodities connected by the edges and the users who purchased the commodities.

In the present specification, the generation of the reliability map is decoupled from the downstream business, that is, the cutting of the whole map data is not performed in real time when the risk identification business is executed, but is generated in advance before the risk identification business is executed, and when the risk identification business needs to be executed, the risk identification business can be executed directly based on the generated reliability map. Therefore, occupation of computing resources during execution of the risk identification service can be reduced, the problems that due to noise, the service result obtained by executing the risk identification service is inaccurate, and the computing resources are wasted are avoided, the computing pressure during execution of the risk identification service can be reduced, and the accuracy of the determined service result is guaranteed.

It should be noted that, in this specification, the full graph data may be determined based on the nodes and the edges formed by the historical traffic between the nodes. Historical traffic as edges is different from the managed traffic that is executed based on the reliability graph. For example, the history service for generating the full graph data may be a transfer service and/or a payment service between nodes (users), and the like. When the service executed based on the credible graph is a wind control service performed on a recommendation service, the historical service for generating the full graph data may be an inter-node purchase order, that is, an order for a user node to purchase a commodity node.

For the purpose of differentiation, the wind control service executed based on the credible graph is subsequently taken as the target service.

In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without any creative effort belong to the protection scope of the present specification.

The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.

Fig. 1 is a schematic flow chart of a method for executing a risk identification service in this specification, which specifically includes the following steps:

s100: and splitting the historical full graph data into sub graph data.

In this specification, the method for performing risk identification service may be performed by a server, and specifically, may be a single server, or may be a server cluster, and this specification is not limited herein.

Because the reliability map is obtained based on the weight of the full map data output by the pre-noise reduction model, the pre-noise reduction model can be trained firstly.

First, the server may determine historical full graph data according to each historical service as a training sample for training the pre-noise reduction model. Since the full graph data is composed of nodes and edges connecting the nodes, the server can determine the business object of each historical business, take the business object as a node and take the historical business as an edge. The business object of a historical business can comprise an active party and a passive party. For example, in a recommendation scenario, the active party may be a user purchasing goods and the passive party may be the goods being purchased. Of course, the recommendation scene is not limited to recommendation of a product, and may be music, an article, or the like. Or, in a pure wind control scenario, the active party and the passive party may be both parties of a transaction.

After determining the historical full graph data, the server may split the historical full graph data into sub-graph data.

Specifically, the server can split the historical full graph data in a sampling or dividing mode to obtain each sub graph data. When each sub-graph data is obtained through division, the sub-graph data can be divided evenly. Alternatively, the partitioning may be non-uniform, so that each main body can be split into one of the sub-graph data.

In one or more embodiments of the present specification, when obtaining sub-graph data through sampling, specifically, the server may determine each node included in the historical full-graph data as each target node, and determine, for each target node, a node connected to the target node in a first order as a first node according to the historical full-graph data. Then, for each first node, a node first-order connected to the first node is determined as a second node. And finally, determining sub-graph data corresponding to the target node from the historical full-graph data according to the first node and the second node.

Taking the wind control service as an example, it is assumed that the graph data is formed by using users as nodes and using services executed by the users as edges. The server can take each user in the historical full graph data as a target node. And aiming at each user, taking the user as a center, determining a node of the first-order connection of the user as a first node. Then, for each first node, a node first-order connected to the first node is determined as a second node. And finally, determining sub-graph data corresponding to the target node from the historical full-graph data according to the first node and the second node.

Fig. 2 is a schematic diagram of determining sub-graph data provided in this specification. As shown, the top largest rectangle in fig. 2 represents a full graph data, wherein circles represent nodes and arrows represent edges. It can be seen that the full graph data is composed of a number of nodes and edges connecting the nodes, and the edges have directions. For example, when the nodes are different users and the edge is transfer service, the direction of the edge may represent the transfer direction. Ellipses in the full graph data represent omissions of some of the nodes and edges. As can be seen from the bold arrows under the full graph data, a plurality of sub-graph data can be obtained by sampling the full graph data, and one sub-graph data is represented in each smaller rectangle under the bold arrows. Of course, partial sub-graph data is omitted from the figure, and only 3 sub-graph data are shown. In the sub-graph data, circles filled with diagonal lines represent target nodes, circles filled with grids represent first nodes, and circles without filling represent second nodes.

S102: determining the weight of each main body in each sub-graph data through a trained interpretation model, wherein the interpretation model takes the sub-graph data corresponding to the historical wind control service as a training sample, the service result of the historical wind control service is obtained by label training, and the weight represents the contribution degree of the main body serving as the input sub-graph data to the obtained labeled service result.

In one or more embodiments of the present specification, the weights of the subjects in each sub-graph data can be output through a pre-trained interpretation model to determine the labels of the historical full graph data.

Therefore, the server can determine the weight of each main body in each sub-graph data through the trained interpretation model. That is, for each sub-graph data, the sub-graph data is input to the interpretation model, and the weight corresponding to each main body in the sub-graph data is obtained. And the weight representation of the main body determines the contribution degree of the main body to the service result of the target service when the target service is executed.

Wherein the interpretation model can output the weight of at least one of the two subjects of the node and the edge. The interpretation model takes sub-graph data corresponding to historical wind control services as training samples, and service results of the historical wind control services are obtained by label training. And the weight represents the contribution degree of the main body of the input sub-graph data to the business result of the label.

In one or more embodiments of the present disclosure, the server inputs the sub-graph data into the interpretation model, determines a weight of a subject in the sub-graph data according to the interpretation model, weights the sub-graph data according to the weight of the subject, and determines a risk prediction result based on the weighted sub-graph data. Therefore, when the interpretation model is trained, the server can take the service result of the historical wind control service as a label, the sub-graph data corresponding to the historical wind control service as a training sample, and the interpretation model is trained by taking the minimum difference between the risk prediction result output by the interpretation model and the label as an optimization target. The interpretation model can learn from historical risk business which main body in the sub-graph data has high importance for obtaining an accurate risk prediction result (namely a business result), the higher the weight is, the greater the influence on the risk prediction result is, and the greater the contribution degree of the interpretation model to outputting the risk prediction result which is the same as the labeled business result is.

Therefore, by taking the business result as a label and the sub-graph data as a training sample, an interpretation model of the risk prediction result is trained and output, and the weight of each main body in the sub-graph data can be given. And the weight represents the main body of the input sub-graph data, and the contribution degree of the risk prediction result which is the same as the labeled business result is obtained for the interpretation model. When the predicted risk result output by the interpretation model is more accurate, the more accurate the weight of the subject obtained by the interpretation model is.

S104: and determining the weight of each main body in the historical full graph data according to the weight of each main body in each sub graph data, wherein the weight is used as the label of the historical full graph data.

In one or more embodiments of the present specification, after obtaining the weight of the subject in each sub-graph data, the weight of each subject in the historical full graph data may be determined according to the weight of each subject in each sub-graph data, and the weight is used as the label of the historical full graph data.

When each sub-graph data is obtained through segmentation, as the main bodies in different sub-graph data are not repeated, and the main bodies in the sub-graph data correspond to the main bodies in the historical full-graph data one by one, the server can directly use the weight of each main body in each sub-graph data as the weight of each main body in the historical full-graph data.

When each sub-graph data is obtained through sampling, the main bodies contained in different sub-graph data may repeat, and the same main body in the historical whole-graph data may belong to different sub-graph data at the same time, so that one main body can also correspond to a plurality of weights determined based on different sub-graph data input interpretation models. The server can determine, for each subject, a weight corresponding to the subject in the historical full graph data from the weights output by the interpretation model corresponding to the subject.

S106: training the pre-noise reduction model according to the historical full map data and the label of the historical full map data; the pre-noise reduction model is used for outputting the weight of each main body in the full-image data to be subjected to noise reduction.

In one or more embodiments of the present description, the server may train the pre-noise reduction model according to the historical full graph data and the label of the historical full graph data.

Similar to the interpretation model, the pre-noise reduction model is used to output the weight of each subject in the full graph data. But compared with the explanation model, the pre-noise reduction model is a lighter model with less model parameters. For example, the interpretation model may be one of a Graph Neural Network interpreter (GNNExplainer) and a PG interpreter (PGExplainer). The pre-noise reduction model may be a tree model or a shallow neural network model, or may be other models capable of outputting a subject weight in graph data, which is not limited herein.

The pre-noise reduction model is used for outputting the weight of each main body in the full-image data to be subjected to noise reduction. The whole graph data refers to all the whole graph data to be denoised, and does not refer to any whole graph data.

S108: and when the updating condition is determined to be met, determining the current full-map data, and inputting the current full-map data into the trained pre-noise reduction model to obtain the weight of each main body in the current full-map data.

In one or more embodiments of the present disclosure, the full graph data determined at different times is different because new historical traffic is continuously generated over time. Therefore, the server can determine the current full-map data according to the updating condition. And when the updating condition is determined to be met, determining the current full-map data, and inputting the current full-map data into the trained pre-noise reduction model to obtain the weight of each main body in the current full-map data.

In one or more embodiments of the present specification, when a historically determined reliability map exists, the server may determine a latest historically determined reliability map as a historical reliability map, and determine a newly added historical service after the historical reliability map is generated, so as to determine the current full map data according to the historical reliability map and the newly added historical service. When the historically determined credible graph does not exist, the server can determine the current full graph data according to the current historical services.

The update condition may be set as needed, for example, the update condition may include an update condition based on the number of the newly added historical services, an update condition based on a time interval, and the like.

In one or more embodiments of the present description, the server may determine the current full map data at preset time intervals. When the server determines that the interval between the current time and the last time the current full map data was historically determined reaches the time interval based on the time interval, it determines that the update condition is satisfied.

In one or more embodiments of the present specification, the server may further determine the current full graph data according to a preset number of newly added historical services. For example, the server may set a preset newly added historical traffic amount as the target amount. The server may determine that the update condition is satisfied when the number of the newly added historical traffics reaches the target number. Namely, when the server determines that the number of the newly added history services reaches the target number since the current full graph data is determined last time in history, the server determines that the update condition is met.

In one or more embodiments of the present specification, the server may further update the current full-graph data in real time when determining a travel-added historical business, so as to ensure integrity of the current full-graph data used for determining the credible graph.

S110: and cutting the main bodies in the current full graph data according to the weight of each main body in the current full graph data to obtain the credible graph.

In one or more embodiments of the present disclosure, after determining the weight of each subject in the current full map data, the subject in the current full map data may be cut according to the weight of each subject in the current full map data, so as to obtain the reliability map.

As described above, the weight of a principal may be used to represent one of an operation to retain the principal and an operation to crop the principal. For example, the weight may be 0 or 1. When the weight of one body is 0, the weight represents an operation of clipping the body, and when the weight of the body is 1, the weight represents an operation of retaining the body.

Alternatively, the weight may be a numerical value between 0 and 1. Whether the weight represents an operation of retaining a subject or an operation of clipping a subject may be determined based on the weight and a preset weight threshold.

For example, a weight larger than a preset weight threshold value may be used as a weight representing an operation of retaining a subject, and a weight not larger than the weight threshold value may be used as a weight representing an operation of clipping a subject. Assuming that the weight threshold is 0.55, when the weight of the determined subject is less than 0.55, the subject is regarded as noise, and the weight is determined as an operation representing a clipped subject. When the weight of the determined subject is not less than 0.55, the weight may be determined as an operation indicating that the subject is left, the subject is determined not to be noise, and the subject is not clipped.

And cutting off the main body corresponding to the noise in the determined current full image data to obtain the credible image.

S112: and when a wind control service request carrying service data is received, identifying a risk identification result of a service corresponding to the service data according to the service data and the credible graph.

In one or more embodiments of the present description, when a wind-controlled service request carrying service data is received, the server may execute a wind-controlled service according to the determined reliability map, and obtain a risk identification result of a service corresponding to the service data.

For example, when the wind-controlled service is a wind-controlled service for wind-controlling a payment service, the service data carried by the wind-controlled service request may be payment data of the payment service (at least part of data of the payment initiator, the receiver, the amount, the payment account, and behavior data of the payment initiator). The server can determine a risk identification result, namely a wind control result, of the payment service based on the service data and the credibility graph.

Of course, for different types of wind control services, different pre-noise reduction models can be trained for outputting an interpretation model for training the labels of the pre-noise reduction models, and training can also be performed based on the historical wind control services corresponding to different service types and the service results thereof.

Thus, the service request may also carry the service type. The server can also determine a corresponding credible graph according to the service type carried by the service request when receiving the service request, and execute the service corresponding to the service request according to the credible graph corresponding to the service type.

For example, when the service type is the wind control service corresponding to the payment service, the reliability map corresponding to the service type is obtained by cutting the current full map data corresponding to the payment service according to the weight output by the pre-noise reduction model obtained by training the historical full map data determined based on the payment service. The method comprises the steps of training a pre-noise reduction model obtained based on historical full-map data corresponding to payment services, namely, for an explanation model for outputting labels of the pre-noise reduction model corresponding to the payment services, taking sub-graph data determined by the historical full-map data as a training sample of the explanation model, taking a wind control result of historical wind control services corresponding to the training sample as a service result, determining the service result as a label, and then training the service result based on the training sample and the labels.

When the wind control result corresponding to the training sample is determined, a target node or a target edge corresponding to the training sample (the training sample is obtained by sampling or cutting from historical full-map data by taking the target node or the target edge as a center) can be determined, and the wind control result of the historical wind control service of the target node or the target edge is used as the wind control result corresponding to the training sample.

It should be noted that, before the server receives the service request, steps S108 to S110 may be repeatedly executed until the service request is received, and a risk identification result of the service corresponding to the service data carried in the service request is identified according to the credible map determined latest in step S110.

Of course, the reception of the service request does not mean the end of the update of the reliability map, and the server may continue to update the reliability map through steps S108 to S110.

According to the method for executing the risk recognition service shown in fig. 1, historical full graph data is divided into sub graph data, the weight of a main body in each sub graph data is determined through a trained interpretation model, the weight of the main body in the historical full graph data is determined according to the weight of the main body in each sub graph data and is used as the label of the historical full graph data, and a pre-noise reduction model is trained according to the historical full graph data and the label of the historical full graph data. And when determining that the updating condition is met, determining current full-map data, inputting the current full-map data into the trained pre-noise reduction model to obtain the weight of each main body in the current full-map data, cutting the main body in the current full-map data according to the obtained weight to obtain a credible map, and when receiving a wind control service request carrying service data, identifying a risk identification result of a service corresponding to the service data according to the service data and the credible map.

According to the method, the pre-noise reduction model can be trained based on the historical full-map data and the label determined by the weight of the main body in the historical full-map data output by the interpretation model, so that before the risk identification service is executed, the current full-map data can be cut based on the weight output by the pre-noise reduction model obtained through training to obtain the credible map after noise reduction, and when a service request carrying service data is received, the risk identification service can be executed according to the service data and the credible map to obtain the risk identification result of the service corresponding to the service data. Due to the fact that the risk identification result is obtained by cutting the credible graph, calculation pressure is small when risk identification business is executed based on the credible graph, waste of calculation resources is little, and a more accurate risk identification result can be obtained through noise reduction.

In addition, in one or more embodiments of the present specification, in step S104, when determining the weight of each subject in the historical full view data according to the weight of each subject in each sub-view data, specifically, the server may set, for each subject in the historical full view data, the weight corresponding to the operation indicating the retention subject as the weight of the subject in the historical full view data when the subject corresponds to the weight determined based on the plurality of sub-view data and the operations indicated by the weights corresponding to the subjects are different.

For example, when a subject is sampled into 3 pieces of sub-graph data and the weight of the subject in each piece of sub-graph data is 1, 0, or 0, the weight 1 corresponding to the operation indicating the retention subject may be set as the weight of the subject in the history whole graph data.

When the weight is a numerical value between 0 and 1, and there are multiple weights indicating the operation of the retention subject, for example, the weight of the retention subject in each sub-graph data is 0.9, 0.8, and 0.45, and if the weight threshold is 0.5, 0.9 and 0.8 larger than 0.5 are weights indicating the operation of the retention subject, and the server may determine the weight of the retention subject in the historical full graph data according to the weights 0.9 and 0.8 corresponding to the operation indicating the retention subject.

Specifically, the server may randomly determine a weight from weights corresponding to operations representing retention subjects as the weight of the subject in the historical full graph data. Alternatively, the weight of the subject in the historical full map data may be determined according to one of a maximum value, a minimum value, a median, and a mode of the weight corresponding to the operation representing the retention subject.

Taking a pure wind control service as an example, for the pre-noise reduction model, the weight of each main body in the sub-graph data determined by the interpretation model is, and the main bodies respectively contribute to obtaining a wind control result in the sub-graph data range. When one main body corresponds to the weight corresponding to a plurality of sub-graph data, one weight is determined from the weight as the weight of the main body in the historical full graph data, so that the weight of the main body in the historical full graph data can be more reasonably determined by combining the weight of the main body in different sub-graph data based on different bases. The weights output by the pre-noise reduction model obtained through training are more accurate based on the marking of the weights output by the interpretation model, and the obtained credible graph is more sufficient for obtaining an accurate wind control result after the current full graph data is cut based on the weights output by the pre-noise reduction model.

In addition, in order to increase the number of training samples for training the pre-noise reduction model to obtain a better pre-noise reduction model, in one or more embodiments of the present specification, the server may further determine historical full-map data corresponding to different historical times as each second sample, and split the second sample of the historical time into several sub-map data for the second sample corresponding to each historical time. And determining the weight of each subject in each sub-graph data of the second sample through the trained interpretation model, and determining the weight of each subject in the second sample according to the weight of each subject in each sub-graph data to be used as the label of the historical full-graph data. And then, training the pre-noise reduction model according to the second sample and the label of the second sample.

Therefore, before determining the current full map data, the server may determine historical full map data corresponding to different historical time instants as second samples, so as to split the second sample of each historical time instant into several sub-map data for the second sample corresponding to the historical time instant. And then, aiming at each sub-graph data of the historical moment, inputting the sub-graph data into a pre-trained interpretation model to obtain the weight corresponding to each main body in the sub-graph data as each first weight.

And then, determining the label of the second sample corresponding to the historical time according to the first weight of each main body in the second sample determined based on each sub-graph data. And inputting the second sample into a pre-noise reduction model to be trained, and then obtaining the weight corresponding to each main body in the second sample as each second weight. Then, the server may determine a difference between each second weight of the second sample and the label of the second sample, and train the pre-noise reduction model with a target of minimizing the determined difference corresponding to the second sample at each historical time.

In addition, since the historical full graph data may be heterogeneous graph data corresponding to multiple types of services, for different types of services, the contribution degrees of the historical full graph data to the service result of the target service are different, and therefore, a pre-noise reduction model can be trained for each type of service. Therefore, before determining the full graph data, the server may further determine, for each type of service in the historical full graph data, a third sample corresponding to the type of service according to the historical full graph data, and determine several sub graph data corresponding to the third sample, when the historical full graph data is heterogeneous graph data corresponding to multiple types of services. And then, respectively inputting each sub-image data into the trained interpretation model to obtain the weight corresponding to each main body in each sub-image data as each first weight.

And then, determining the label of a third sample according to the first weight of each main body in each sub-graph data, inputting the third sample into the pre-noise reduction model to be trained, and determining the weight corresponding to each main body in the third sample output by the pre-noise reduction model to be trained as each second weight.

The server can determine the difference between the label of the third sample and the second weight of each main body of the third sample, and then train the pre-noise reduction model by taking the determined difference as a target to obtain the pre-noise reduction model corresponding to the type of service.

Since for a principal, the weight of the principal includes one of an operation representing preserving the principal and an operation representing clipping the principal, the determination of the weight for a principal is a two-classification problem. Therefore, the loss can be determined based on the cross entropy loss function according to the difference between the labeling of the sample and the weight of each main body output by the pre-noise reduction model, and the pre-noise reduction model is trained by taking the minimum loss as a target. Of course, training may also be performed based on other loss functions, and the description is not limited herein.

In one or more embodiments of the present specification, when determining the third sample corresponding to the type of service according to the historical full graph data, specifically, the server may determine a node and an edge corresponding to the type of service from the historical full graph data, and determine the graph data corresponding to the type of service according to the determined node and edge of the type of service, as the third sample corresponding to the type of service.

The present specification also provides a method of training an interpretation model. The training of the interpretation model can also be performed by a server, and the server can determine each sub-graph data according to the historical full-graph data and take each sub-graph data as a training sample of the interpretation model. The interpretation model takes graph data as input, and takes weights in the graph data and business results of the target business determined based on the weights as output. The historical full map data used for training the interpretation model may be the same as or different from the historical full map data used for training the pre-noise reduction model.

As described above, the pre-noise reduction model is used to output the weight of each subject in the full-map data, similar to the interpretation model. However, unlike the interpretation model, the pre-noise reduction model is not used to output the business results, but only to output the weights of the subject.

The interpretation model may be trained specifically using the following method, wherein:

firstly, the server can determine each sub-graph data as each first sample, and determine the label corresponding to each first sample according to the service result of each historical wind control service executed by the target node in the first sample. And then, inputting the first sample as an input into an interpretation model to be trained to obtain weights corresponding to all subjects in the first sample, weighting the first sample according to the determined weights, and outputting a prediction result of a service corresponding to the first sample based on the weighted first sample. And finally, training the interpretation model by taking the minimum difference between the label of the first sample and the prediction result output by the interpretation model as a target. The weights output by the interpretation model are used to characterize the degree of contribution of the subject to the derived prediction (i.e., predicted traffic outcome).

In addition, in order to increase the number of training samples for training the interpretation model to obtain a better interpretation model, in one or more embodiments of the present specification, the server may further determine historical full-graph data corresponding to different historical time instants, and then, for the historical full-graph data corresponding to each historical time instant, split the historical full-graph data into several sub-graph data as each first sample corresponding to the historical time instant. And determining a label corresponding to each first sample according to a service result of each historical wind control service executed at the historical moment by a target node in the first sample, taking the first sample as input, inputting an interpretation model to be trained to obtain weights respectively corresponding to each main body in the first sample, weighting the first sample according to the determined weights, and outputting a prediction result of the service corresponding to the first sample based on the weighted first sample. And determining the difference between the label of the first sample and the predicted result of the first sample output by the interpretation model. Finally, the server may train the interpretation model with the goal of minimizing the difference between the label of the first sample at the historical time and the prediction result of the first sample output by the interpretation model.

For the interpretation model, the loss can also be determined based on the cross entropy loss function according to the label of the first sample and the prediction result of the first sample output by the interpretation model, and the interpretation model is trained with the minimum loss as the target.

It should be noted that the interpretation model and the pre-noise reduction model may also be trained jointly. The interpretation model may also be trained in an unsupervised or an auto-supervised manner. And when the downstream obtains a business result by inputting the reliability map into the GNN model, the pre-noise reduction model and the downstream GNN model can be trained jointly or separately.

Also, the method of performing risk identification services provided herein may be used in, but is not limited to: in search, recommendation, advertising, social, wind control, financial, etc. scenarios.

Fig. 3 is a schematic diagram of an apparatus for performing risk identification service provided in the present specification, the apparatus including:

the splitting module 200 is used for splitting the historical full graph data into sub graph data;

the first determining module 201 is configured to determine, through a trained interpretation model, a weight of each principal in each sub-graph data, where the interpretation model uses sub-graph data corresponding to a historical wind control service as a training sample, a service result of the historical wind control service is obtained by label training, and the weight represents a contribution degree of the principal as input sub-graph data to obtain a labeled service result;

the labeling module 202 is configured to determine, according to the weight of each subject in each sub-graph data, the weight of each subject in the historical full-graph data as a label of the historical full-graph data;

the first training module 203 is configured to train the pre-denoising model according to the historical full map data and the label of the historical full map data; the pre-noise reduction model is used for outputting the weight of each main body in the full-image data to be subjected to noise reduction;

the weighting module 204 is configured to determine current full map data when it is determined that an update condition is met, and input the current full map data into the trained pre-denoising model to obtain weights of all subjects in the current full map data;

the cutting module 205 is configured to cut the main bodies in the current full map data according to the weight of each main body in the current full map data to obtain a reliability map;

and the execution module 206 is configured to, when a wind control service request carrying service data is received, identify a risk identification result of a service corresponding to the service data according to the service data and the reliability map.

Optionally, the splitting module 200 is further configured to determine each node included in the historical full graph data, to serve as each target node, determine, according to the historical full graph data, a node connected to the target node in a first order as a first node for each target node, determine, for each first node, a node connected to the first node in a first order as a second node, and determine, according to the first node and the second node, sub-graph data corresponding to the target node from the historical full graph data.

Optionally, the apparatus further comprises:

the second training module 207 is configured to determine sub-graph data as first samples, determine, for each first sample, a label corresponding to the first sample according to a service result of each historical wind control service executed by a target node in the first sample, input the first sample as an input to an interpretation model to be trained, obtain weights corresponding to each principal in the first sample, weight the first sample according to the determined weights, output a prediction result of the service corresponding to the first sample based on the weighted first sample, and train the interpretation model with a goal of minimizing a difference between the label of the first sample and the prediction result output by the interpretation model.

Optionally, the apparatus further comprises:

the third training module 208 is further configured to determine historical full-graph data corresponding to different historical moments, divide the historical full-graph data into a plurality of sub-graph data for the historical full-graph data corresponding to each historical moment, serve as each first sample corresponding to the historical moment, determine, for each first sample corresponding to the historical moment, a label corresponding to the first sample according to a service result of each historical wind control service executed by a target node in the first sample by the historical moment, input the first sample as an input to the interpretation model to be trained, obtain weights corresponding to each main body in the first sample, weight the first sample according to the determined weights, output a prediction result of the service corresponding to the first sample based on the weighted first sample, determine a difference between the label of the first sample and the prediction result of the first sample output by the interpretation model, so as to minimize the label of the first sample at the historical moment, output the difference between the label of the first sample and the prediction result of the first sample output by the interpretation model as a training target for the interpretation model.

Optionally, the weight is used to represent one of an operation of retaining a main body and an operation of clipping the main body, and the labeling module 202 is further configured to, for each main body of the historical full graph data, when the main body corresponds to a weight determined based on a plurality of sub graph data and the operations represented by the weights corresponding to the main body are different, use the weight corresponding to the operation representing the retained main body as the weight of the main body in the historical full graph data.

Optionally, the apparatus further comprises:

a fourth training module 209, configured to determine historical full-graph data corresponding to different historical moments, as each second sample, split the second sample at the historical moment into a plurality of sub-graph data for the second sample corresponding to each historical moment, input the sub-graph data into a pre-trained interpretation model for each sub-graph data at the historical moment, obtain weights corresponding to each main body in the sub-graph data, as each first weight, determine a label of the second sample corresponding to the historical moment according to the first weight of each main body in the second sample determined based on each sub-graph data, input the second sample into a pre-noise reduction model to be trained, obtain weights corresponding to each main body in the second sample, as each second weight, determine a difference between each second weight of the second sample and the label of the second sample, and train the pre-noise reduction model with the determined difference corresponding to each historical moment being minimized.

Optionally, the apparatus further comprises:

a fifth training module 210, configured to, when historical full-graph data is heterogeneous graph data corresponding to multiple types of services, determine, according to the historical full-graph data, a third sample corresponding to the type of service according to the historical full-graph data, determine a plurality of sub-graph data corresponding to the third sample, input each sub-graph data into a trained interpretation model, obtain a weight corresponding to each main body in each sub-graph data, as each first weight, determine, according to the first weight of each main body in each sub-graph data, determine a label of the third sample, input the third sample into a pre-noise reduction model to be trained, determine a weight corresponding to each main body in the third sample output by the pre-noise reduction model to be trained, as each second weight, determine a difference between the label of the third sample and the second weight of each main body of the third sample, and train the pre-noise reduction model with the difference minimized, to obtain the pre-noise reduction model corresponding to the type of service.

The present specification also provides a computer readable storage medium having stored thereon a computer program operable to execute the above-described method of performing a risk identification service.

This specification also provides a schematic block diagram of the electronic device shown in fig. 4. As shown in fig. 4, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to realize the method for executing the risk identification service. Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.

In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development, but the original code before compiling is also written in a specific Programming Language, which is called Hardware Description Language (HDL), and the HDL is not only one kind but many kinds, such as abll (Advanced boot Expression Language), AHDL (alternate hard Description Language), traffic, CUPL (computer universal Programming Language), HDCal (Java hard Description Language), lava, lola, HDL, PALASM, software, rhydl (Hardware Description Language), and vhul-Language (vhyg-Language), which is currently used in the field. It will also be apparent to those skilled in the art that hardware circuitry for implementing the logical method flows can be readily obtained by a mere need to program the method flows with some of the hardware description languages described above and into an integrated circuit.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, 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, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in purely computer readable program code means, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.

The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

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

The memory may include forms of volatile memory in a computer readable medium, random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, 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 technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises that element.

As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.

This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.

Claims

1. A method of performing a risk identification service, the method comprising:

splitting the historical full graph data into sub graph data;

determining the weight of each main body in each sub-graph data through a trained interpretation model, wherein the interpretation model takes the sub-graph data corresponding to the historical wind control service as a training sample, the service result of the historical wind control service is obtained by label training, and the weight represents the contribution degree of the main body serving as the input sub-graph data to the obtained labeled service result, wherein the main body in the sub-graph data comprises nodes and edges, the service object is taken as the node, and the historical service is taken as the edge;

according to the weight of each main body in the current full map data, cutting the main body in the current full map data to obtain a credible map;

when a wind control service request carrying service data is received, identifying a risk identification result of a service corresponding to the service data according to the service data and the credible graph;

when the interpretation model is trained, sub-graph data corresponding to historical wind control business are used as training samples, business results of the historical wind control business are used as labels, the training samples are input into the interpretation model, the weight of a main body in the sub-graph data corresponding to the training samples output by the interpretation model is determined, the sub-graph data are weighted according to the weight of the main body, then a risk prediction result is determined based on the weighted sub-graph data, and the interpretation model is trained by taking the minimum difference between the risk prediction result and the labels as an optimization target.

2. The method of claim 1, wherein the splitting of the historical full graph data into sub graph data specifically comprises:

determining each node contained in the historical full graph data as each target node;

for each target node, determining a node connected with the target node in a first order as a first node according to the historical full graph data;

for each first node, determining a node connected with the first node in a first order as a second node;

and determining sub-graph data corresponding to the target node from the historical full-graph data according to the first node and the second node.

3. The method of claim 2, the interpretation model is trained using a method wherein:

determining each sub-graph data as each first sample;

for each first sample, determining a label corresponding to the first sample according to a service result of each historical wind control service executed by a target node in the first sample;

inputting the first sample as input into an interpretation model to be trained to obtain weights corresponding to each main body in the first sample, weighting the first sample according to the determined weights, and outputting a prediction result of a service corresponding to the first sample based on the weighted first sample;

and training the interpretation model by taking the minimum difference between the label of the first sample and the prediction result output by the interpretation model as a target.

4. The method of claim 2, the interpretation model is trained using a method wherein:

determining historical full-map data corresponding to different historical moments;

according to the historical full graph data corresponding to each historical moment, dividing the historical full graph data into a plurality of sub graph data to serve as first samples corresponding to the historical moments;

determining a label corresponding to each first sample corresponding to the historical time according to the service result of each historical wind control service executed by the target node in the first sample at the historical time;

determining the difference between the label of the first sample and the predicted result of the first sample output by the interpretation model;

and training the interpretation model by taking the difference between the label of the first sample at the minimized historical moment and the prediction result of the first sample output by the interpretation model as a target.

5. The method of claim 1, wherein the weight is used to represent one of an operation to reserve a body and an operation to crop a body;

determining the weight of each subject in the historical full graph data according to the weight of each subject in each sub graph data, which specifically comprises the following steps:

and regarding each main body of the historical full graph data, when the main body corresponds to the weight determined based on the plurality of sub graph data and the operation represented by each weight corresponding to the main body is different, regarding the weight corresponding to the operation representing the reserved main body as the weight of the main body in the historical full graph data.

6. The method of claim 1, prior to determining current full map data, the method further comprising:

determining historical full-map data corresponding to different historical moments as second samples;

for a second sample corresponding to each historical time, splitting the second sample of the historical time into a plurality of sub-graph data;

inputting the sub-graph data into a pre-trained interpretation model aiming at each sub-graph data of the historical moment to obtain weights respectively corresponding to all main bodies in the sub-graph data as all first weights;

determining labels of second samples corresponding to the historical moments according to the first weights of all subjects in the second samples determined based on all the sub-graph data;

inputting the second sample into a pre-noise reduction model to be trained, and obtaining weights corresponding to all main bodies in the second sample as all second weights;

determining a difference between each second weight of the second sample and the label of the second sample;

and training the pre-noise reduction model by taking the minimized difference corresponding to the second sample at each historical moment as a target.

7. The method of claim 1, prior to determining current full graph data, the method further comprising:

when the historical full graph data is heterogeneous graph data corresponding to multiple types of services, determining a third sample corresponding to the type of service according to the historical full graph data and determining a plurality of sub-graph data corresponding to the third sample for each type of service in the historical full graph data;

respectively inputting each sub-image data into the trained interpretation model to obtain the weight corresponding to each main body in each sub-image data as each first weight;

determining the label of the third sample according to the first weight of each main body in each sub-graph data;

inputting the third sample into a pre-noise reduction model to be trained, and determining weights respectively corresponding to all main bodies in the third sample output by the pre-noise reduction model to be trained as all second weights;

determining a difference between the label of the third sample and a second weight of each subject of the third sample;

and training the pre-noise reduction model by taking the minimized difference as a target to obtain the pre-noise reduction model corresponding to the type of service.

8. An apparatus to perform a risk identification service, the apparatus comprising:

the first determining module is used for determining the weight of each main body in each sub-graph data through a trained interpretation model, the interpretation model takes the sub-graph data corresponding to historical wind control service as a training sample, the service result of the historical wind control service is obtained by label training, and the weight represents the contribution degree of the main body of the sub-graph data used as input to the obtained labeled service result, wherein the main body in the sub-graph data comprises nodes and edges, a service object is used as a node, and the historical service is used as an edge;

the weighting module is used for determining current full-image data when the updating condition is determined to be met, inputting the current full-image data into the trained pre-noise reduction model, and obtaining the weight of each main body in the current full-image data;

the cutting module is used for cutting the main bodies in the current full map data according to the weight of each main body in the current full map data to obtain a credible map;

the execution module is used for identifying a risk identification result of a service corresponding to the service data according to the service data and the credible graph when a wind control service request carrying the service data is received;

9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any one of claims 1 to 7 when executing the program.