CN111523678A

CN111523678A - Service processing method, device, equipment and storage medium

Info

Publication number: CN111523678A
Application number: CN202010319159.5A
Authority: CN
Inventors: 南晓杰; 王逸聪; 曹大瀛; 励强超; 仲杉
Original assignee: JD Digital Technology Holdings Co Ltd
Current assignee: JD Digital Technology Holdings Co Ltd
Priority date: 2020-04-21
Filing date: 2020-04-21
Publication date: 2020-08-11

Abstract

According to the service processing method, the service processing device, the service processing equipment and the storage medium, the service processing model is adopted to process the service data corresponding to the service request to be processed, so that the prediction result of the target variable of the service processing model is obtained, the prediction result of the target variable is explained in a preset local explanation mode, and the local service explanation corresponding to the prediction result of the target variable is obtained. Therefore, in the embodiment of the application, the machine learning model is white-boxed by combining the machine learning model with the model interpretation to obtain the relevant business interpretation information about the machine learning model, so that the interpretability of the machine learning model is improved, on one hand, the model reliability is improved, and the utilization rate of the machine learning model in the technical field with higher requirements on the model interpretability is favorably improved; and on the other hand, the service interpretation information is combined to determine the user group suitable for the service to be processed.

Description

Service processing method, device, equipment and storage medium

Technical Field

The embodiment of the application relates to the technical field of computers, in particular to a service processing method, device, equipment and storage medium.

Background

With the development of big data and artificial intelligence, machine learning technology is more and more widely applied.

In the prior art, many machine learning models are black box models, the working state in the machine learning models cannot be sensed, and the interpretability is poor. For the fields with high requirements on model interpretability, such as Internet finance and electronic commerce, the existing black box sub-model has the problem of low model reliability caused by poor interpretability.

Disclosure of Invention

Embodiments of the present application provide a service processing method, apparatus, device, and storage medium, which are used to solve the problem in the prior art that model reliability is low due to poor interpretability of a black box model.

In a first aspect, an embodiment of the present application provides a method for processing a service, including:

acquiring service data related to a service processing model according to a service request to be processed;

processing the service data by adopting the service processing model to obtain a prediction result of a target variable of the service processing model;

interpreting the prediction result of the target variable by adopting a preset local interpretation mode to obtain a local service interpretation corresponding to the prediction result of the target variable; the preset local interpretation mode is a calculation mode used for determining a local service interpretation corresponding to the prediction result.

In one possible implementation, the method further includes:

acquiring an expert explanation and a preset model explanation corresponding to the service to be processed;

and judging whether the local business explanation accords with business logic or not according to the expert explanation and/or the preset model explanation.

In one possible implementation, the method further includes:

when the local business interpretation is determined to accord with the business logic, acquiring the contribution degree of each feature of the business data to the prediction result;

taking the characteristics corresponding to the contribution degrees of the characteristics to the prediction result, which are greater than a preset contribution degree, as optimization characteristics;

and determining a user group related to the service to be processed based on the optimization characteristics.

In one possible implementation, the method further includes:

when the local business interpretation is determined not to accord with the business logic, business data to be acquired related to the business processing model are determined again according to the global business interpretation in the preset model interpretation; the service data to be acquired is used for indicating the type of the service data to be acquired related to the service processing model.

In a possible implementation manner, the obtaining manner of the service processing model includes:

acquiring sample data and preprocessing the sample data;

and training a pre-configured initialized service processing model according to the pre-processed sample data to obtain the service processing model.

In one possible implementation, the method further includes:

interpreting the preprocessed sample data by adopting a preset global interpretation mode, and determining to-be-acquired service data related to the service processing model according to an interpretation result; the service data to be acquired is used for indicating the type of the service data to be acquired related to the service processing model.

In a possible implementation manner, the interpreting the preprocessed sample data in a preset global interpretation manner, and determining to-be-acquired service data related to the service processing model according to an interpretation result includes:

for each feature of the preprocessed sample data, acquiring importance indication information of the feature;

judging whether the interactivity of the features and other features is obvious or not;

when the interaction between the features and other features is not obvious, judging whether the prediction result of the features on the target variables of the business processing model conforms to business logic or not by adopting an accumulated local effect graph or a partial dependency graph;

and when the feature conforms to the service logic for the prediction result, taking the data corresponding to the feature as the service data to be acquired related to the service processing model.

In one possible implementation, the method further includes:

when the feature is determined to be obvious in interactivity with other features, judging whether the prediction result of the feature on the target variable of the business processing model conforms to business logic or not by adopting a SHAP graph or an accumulated local effect graph;

In one possible implementation, the method further includes:

performing derivation processing on the features to obtain derived new features;

and when the prediction result of the new feature conforms to the service logic, taking the data corresponding to the new feature as the service data to be acquired related to the service processing model.

In one possible implementation, the method further includes:

and when the predicted result of the feature pair is determined not to accord with the business logic, the business processing model is trained again.

In a second aspect, an embodiment of the present application provides a device for processing a service, including:

the first acquisition module is used for acquiring service data related to a service processing model according to a service request to be processed;

the second obtaining module is used for processing the service data by adopting the service processing model to obtain a prediction result of a target variable of the service processing model;

the third obtaining module is used for explaining the prediction result of the target variable by adopting a preset local explanation mode to obtain a local service explanation corresponding to the prediction result of the target variable; the preset local interpretation mode is a calculation mode used for determining a local service interpretation corresponding to the prediction result.

In one possible implementation, the apparatus further includes:

the fourth acquisition module is used for acquiring the expert explanation and the preset model explanation corresponding to the service to be processed;

and the judging module is used for judging whether the local service explanation accords with the service logic or not according to the expert explanation and/or the preset model explanation.

In one possible implementation, the apparatus further includes: a fifth obtaining module;

wherein the fifth obtaining module is configured to:

when the judging module determines that the local service interpretation accords with the service logic, acquiring the contribution degree of each feature of the service data to the prediction result;

In one possible implementation, the apparatus further includes: a sixth obtaining module;

wherein the sixth obtaining module is configured to:

when the judging module determines that the local service interpretation does not accord with the service logic, the service data to be acquired related to the service processing model is determined again according to the global service interpretation in the preset model interpretation; the service data to be acquired is used for indicating the type of the service data to be acquired related to the service processing model.

In one possible implementation, the apparatus further includes: a seventh obtaining module;

wherein the seventh obtaining module is configured to:

acquiring sample data and preprocessing the sample data;

In one possible implementation, the apparatus further includes: an eighth obtaining module;

wherein the eighth obtaining module is configured to:

In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method for processing the service according to any one of the first aspect when executing the computer program.

In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, where computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are executed by a processor, the computer-readable storage medium is configured to implement the service processing method according to any one of the first aspect.

According to the service processing method, the device, the equipment and the storage medium provided by the embodiment of the application, the service data related to the service processing model is obtained according to the service request to be processed; further, the business data is processed by adopting the business processing model to obtain a prediction result of a target variable of the business processing model, and the prediction result of the target variable is explained by adopting a preset local explanation mode to obtain a local business explanation corresponding to the prediction result of the target variable. Therefore, in the embodiment of the application, the machine learning model is white-boxed by combining the machine learning model and the model interpretation to obtain the relevant business interpretation information about the machine learning model, so that the interpretability of the machine learning model is improved, and on one hand, the handling basis of business handling results can be displayed to business personnel or clients according to the handling basis, so that the model reliability is improved, and the utilization rate of the machine learning model in the technical field with higher requirements on model interpretability is improved; on the other hand, the service interpretation information is combined to determine the user group (or client group) suitable for the service to be processed.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.

Fig. 1 is a schematic diagram of an application architecture provided in an embodiment of the present application;

fig. 2 is a schematic flowchart of a service processing method according to an embodiment of the present application;

fig. 3 is a schematic flowchart of a service processing method according to another embodiment of the present application;

fig. 4 is a schematic flowchart of a service processing method according to another embodiment of the present application;

fig. 5 is a schematic flowchart of a service processing method according to another embodiment of the present application;

fig. 6 is a schematic flowchart of a service processing method according to another embodiment of the present application;

fig. 7 is a schematic structural diagram of a service processing apparatus according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

With the foregoing drawings in mind, certain embodiments of the disclosure have been shown and described in more detail below. These drawings and written description are not intended to limit the scope of the disclosed concepts in any way, but rather to illustrate the concepts of the disclosure to those skilled in the art by reference to specific embodiments.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

First, terms referred to in the embodiments of the present application will be explained.

The business processing model related to the embodiment of the application is a machine learning model, and may include but is not limited to: a gradient lifting tree model or a deep learning model. For example, the traffic processing model in the embodiment of the present application may be a Machine learning model such as a Gradient Boosting iterative Decision Tree (GBDT), a Gradient Boosting Decision Tree model (Xgboost), or a Light Gradient Boosting Machine (Light gbm).

The local service explanation related to the embodiment of the present application means: contribution degree of different characteristics of the business processing model to the prediction result of the target variable corresponding to the business data of a single sample or a single client of the business processing model.

The global service interpretation related to the embodiment of the application refers to: contribution degree of different characteristics of the business processing model to the prediction result of the target variable corresponding to the business data of all samples or all clients of the business processing model.

The expert interpretation corresponding to any service related to the embodiment of the application can be used for judging whether the service interpretation corresponding to the service conforms to a standard interpretation of service logic. Illustratively, expert interpretations may include, but are not limited to: an expert local interpretation, and/or an expert global interpretation.

The preset model interpretation corresponding to any service related to the embodiment of the present application may be an interpretation about the service processing model obtained in a training process of the service processing model corresponding to the service, and is used to determine whether the service interpretation corresponding to the service conforms to another reference interpretation of the service logic. Illustratively, the preset model interpretation may include, but is not limited to: local business interpretations, and/or, global business interpretations.

The service processing method provided by the embodiment of the application can be applied to application scenes such as Internet finance, electronic commerce and the like which have higher requirements on model interpretability; of course, the method can also be applied to other application scenarios, and this is not limited in the embodiments of the present application.

For example, the service processing method of the embodiment of the present application may be applied to a service processing scenario of credit consumption.

Fig. 1 is a schematic diagram of an application architecture provided in the embodiment of the present application. As shown in fig. 1, the application architecture of the embodiment of the present application may include: a terminal 10 and a service processing device 11; the terminal 10 is configured to send a service request to the service processing device 11; the service processing device 11 is configured to execute a service processing method provided in the embodiment of the present application.

Illustratively, the terminal 10 related to the embodiment of the present application may include, but is not limited to, any one of the following: mobile phones, notebook computers, tablet computers, desktop computers.

Illustratively, the service processing device 11 related to the embodiment of the present application may include, but is not limited to, any one of the following: personal computers, medium and large sized computers, computer clusters.

In the prior art, many machine learning models are black box models, the working state in the machine learning models cannot be sensed, and the interpretability is poor. For the fields with high requirements on model interpretability, such as internet finance, electronic commerce and the like, the existing black box model has the problem of low model reliability caused by poor interpretability, so that the utilization rate of the black box model in the technical fields is low.

In view of the above technical problems, the service processing method provided in the embodiment of the present application, by combining the machine learning model and the model interpretation mode, makes the "black box" machine learning model with poor interpretability white-boxed, obtains the relevant service interpretation information about the machine learning model, and improves the interpretability of the machine learning model, so as to provide a handling basis for displaying the service handling result to service personnel or customers according to the basis, thereby improving the model reliability and facilitating to improve the utilization rate of the machine learning model in the technical field with higher requirements on model interpretability.

The following describes the technical solutions of the present invention and how to solve the above technical problems with specific embodiments. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments. Embodiments of the present invention will be described below with reference to the accompanying drawings.

Fig. 2 is a schematic flow chart of a service processing method according to an embodiment of the present application. The execution main body in the embodiment of the present application may be the service processing device or a service processing apparatus in the service processing device (for convenience of description, the execution main body is taken as the service processing device in this embodiment as an example for description). Illustratively, the service processing device may be implemented by software and/or hardware. As shown in fig. 2, a method for processing a service provided in an embodiment of the present application may include:

step S201, acquiring service data related to a service processing model according to a service request to be processed.

In this embodiment of the application, the service processing device may be preset with a trained service processing model and service data to be acquired (for indicating a type of the service data to be acquired and related to the service processing model) related to the service processing model. It should be noted that the business process model may be trained in advance for the business process device, or may be obtained from other devices for the business process device.

In this embodiment of the present application, the service processing device may receive a service request (for example, a service application request for consuming credit) to be processed, where the service request is sent by a service person or a customer through a terminal, where the service request may include: identification (ID) information of the client; of course, the service request may further include other information (for example, service identification information to be processed, etc.), which is not limited in this embodiment of the application.

Illustratively, the identification information of the client may include, but is not limited to, at least one of: the name and identification number of the customer.

In this step, the service processing device may obtain, from a database or other devices, service data related to the service processing model corresponding to the client according to the service data to be obtained related to the service processing model and the identification information of the client in the service request. Wherein the business data associated with the business process model may be used to indicate a plurality of characteristics of the customer.

For example, if the pending service request is a service application request for credit consumption, the service data related to the service processing model may include, but is not limited to, at least one of the following: attribute feature data (e.g., gender, age, academic history, occupation, marital status, address, etc.) of the customer, payment behavior feature data (e.g., payment amount or payment stroke number, etc.), financial financing feature data (e.g., fund, insurance financing, securities trader financing, etc.), historical repayment behavior feature data, consumption behavior feature data (e.g., online shopping consumption information, coupon usage information, shopping browsing information, risk orders, category preferences, ordering behavior, comment text data, etc.), social relationship network feature data, loan credit investigation feature data (e.g., loan data, multi-headed credit data, etc.).

Of course, the service processing device may obtain the service request to be processed in other manners, which is not limited in this embodiment of the application.

Step S202, the business data is processed by adopting the business processing model, and a prediction result of a target variable of the business processing model is obtained.

The business processing model related to the embodiment of the present application is, for example, a machine learning model, and may include, but is not limited to: a gradient lifting tree model or a deep learning model. For example, the traffic processing model in the embodiment of the present application may be a machine learning model such as GBDT, Xgboost, or lightGBM.

In this step, the service processing device may input the service data related to the service processing model acquired in step S201 to the service processing model, and may acquire a prediction result of a target variable of the service processing model corresponding to the service data, so as to process the service request according to the prediction result of the target variable.

It should be understood that the service processing device may also return the prediction result of the target variable of the service processing model and the service transaction result of the service request to the terminal; the service transaction result of the service request may include, but is not limited to: and approving the service request or rejecting the service request.

For example, if the pending service request is a service application request for credit consumption, the target variable of the service processing model may be a risk probability of a customer breach. For example, if the prediction result of the target variable is 0, the service processing device may convert the prediction result (for example, 0) of the target variable into a credit score, and then the credit score is higher, so as to approve the service request; if the prediction result of the target variable is 1, the service processing device may convert the prediction result (for example, 1) of the target variable into a credit score, and then the credit rating of the target variable is lower, so as to reject the service request.

Step S203, a preset local interpretation mode is adopted to interpret the prediction result of the target variable, and a local service interpretation corresponding to the prediction result of the target variable is obtained.

Illustratively, the preset local interpretation mode related to the embodiment of the present application refers to a calculation mode for determining a local business interpretation corresponding to the prediction result of the target variable.

Illustratively, the local service interpretation related to the embodiments of the present application refers to: and the contribution degree of different characteristics of the business processing model to the prediction result of the target variable corresponding to the business data of a single sample or a single client of the business processing model.

In this step, the service processing device may use a preset local interpretation manner to interpret the prediction result of the target variable obtained in step S202, so as to obtain a local service interpretation corresponding to the prediction result of the target variable, that is, the contribution (or influence degree) of different features of the service processing model to the prediction result of the target variable corresponding to the service data of the client.

It should be understood that, the service processing device may further interpret the local service, and return the interpreted local service to the terminal after being packaged, so that on one hand, a service person or a client can objectively obtain a transaction basis of a service transaction result related to the service request; on the other hand, the local service interpretation information is combined to determine the user group (or client group) suitable for the service to be processed.

Optionally, the service processing device may use a preset local interpretation manner, such as local interpretable Model-adaptive interpretation (LIME), sharey value additive interpretation (swap), or anchor point (Anchors), that is not related to the local interpretable Model to interpret the prediction result of the target variable, so as to obtain a local service interpretation corresponding to the prediction result of the target variable.

For example, if the preset local interpretation mode is LIME, the business processing device may locally train an interpretable proxy model (for example, an interpretable model such as a linear regression model or a decision tree), and then use the proxy model to perform interpretation processing on the prediction result of the target variable, so as to obtain a local business interpretation corresponding to the prediction result of the target variable. It should be noted that, a specific implementation manner of interpreting the prediction result of the target variable in the LIME manner may refer to an implementation manner of the LIME manner provided in the related art.

As another example, if the preset local interpretation mode is the SHAP, the service processing apparatus may obtain the local service interpretation corresponding to the prediction result of the target variable by separately calculating a Shaply value (Values) of each feature of the service processing model and then linearly combining the Shaply Values of each feature; wherein the Shaply value of each feature is an average marginal contribution of the feature in the feature sequence selected (or corresponding) by the business process model. It should be noted that, a specific implementation manner of interpreting the prediction result of the target variable in the SHAP manner may refer to an implementation manner of the SHAP manner provided in the related art.

For another example, if the preset local interpretation manner is Anchors (local interpretation is performed on the prediction result of the target variable based on the perturbation technology), the business processing device may perform local interpretation on the prediction result of the target variable based on the obtained if-then rule, so as to obtain a local business interpretation corresponding to the prediction result of the target variable. It should be noted that, a specific implementation manner of the process of interpreting the prediction result of the target variable by the anchor manner may refer to an implementation manner of the anchor manner provided in the related art.

To sum up, in the embodiment of the present application, service data related to a service processing model is obtained according to a service request to be processed; further, the business data is processed by adopting the business processing model to obtain a prediction result of a target variable of the business processing model, and the prediction result of the target variable is explained by adopting a preset local explanation mode to obtain a local business explanation corresponding to the prediction result of the target variable. Therefore, in the embodiment of the application, the machine learning model is white-boxed by combining the machine learning model and the model interpretation to obtain the relevant business interpretation information about the machine learning model, so that the interpretability of the machine learning model is improved, and on one hand, the handling basis of business handling results can be displayed to business personnel or clients according to the handling basis, so that the model reliability is improved, and the utilization rate of the machine learning model in the technical field with higher requirements on model interpretability is improved; on the other hand, the user group (or client group) suitable for the service to be processed can be determined by combining the service interpretation information.

Fig. 3 is a schematic flow diagram of a service processing method according to another embodiment of the present application, and as shown in fig. 3, on the basis of the foregoing embodiment, after the service processing device executes step S203, the service processing device may further include:

step S301, acquiring expert explanation and preset model explanation corresponding to the service to be processed.

Illustratively, the expert interpretation corresponding to any service related to the embodiment of the present application may be used to determine whether the service interpretation corresponding to the service conforms to a reference interpretation of service logic. Illustratively, expert interpretations may include, but are not limited to: an expert local interpretation, and/or an expert global interpretation.

For example, the preset model interpretation corresponding to any service related to the embodiment of the present application may be an interpretation about the service processing model obtained in a training process of the service processing model corresponding to the service, and is used to determine whether the service interpretation corresponding to the service conforms to another reference interpretation of the service logic. Illustratively, the preset model interpretation may include, but is not limited to: local business interpretations, and/or, global business interpretations.

In this step, the service processing device may obtain, from a database or other devices, an expert interpretation and a preset model interpretation corresponding to the service to be processed; of course, the expert explanation and the preset model explanation corresponding to the service to be processed may also be obtained in other manners, which is not limited in the embodiment of the present application.

Step S302, judging whether the local business explanation accords with business logic according to the expert explanation and/or the preset model explanation.

In this step, the service processing device may determine whether the local service interpretation obtained in step S203 conforms to the service logic according to the expert local interpretation in the expert interpretation and/or the local service interpretation in the preset model interpretation.

For example, the service processing device may determine whether the local service interpretation obtained in step S203 conforms to the service logic according to an expert local interpretation in the expert interpretation.

For example, if the local service interpretation obtained in step S203 conflicts with an expert local interpretation in the expert interpretations, the service processing device may determine that the local service interpretation obtained in step S203 does not conform to the service logic. If the local service explanation obtained in step S203 does not conflict with the expert local explanation in the expert explanation, the service processing device may determine that the local service explanation obtained in step S203 conforms to the service logic.

For another example, the service processing device may determine whether the local service interpretation obtained in step S203 conforms to the service logic according to the local service interpretation in the preset model interpretation.

For example, if the local service interpretation obtained in step S203 conflicts with the local service interpretation in the preset model interpretation, the service processing device may determine that the local service interpretation obtained in step S203 does not conform to the service logic. If the local service interpretation obtained in step S203 does not conflict with the local service interpretation in the preset model interpretation, the service processing device may determine that the local service interpretation obtained in step S203 conforms to the service logic.

For another example, the service processing device may determine whether the local service interpretation obtained in step S203 conforms to the service logic according to an expert local interpretation in the expert interpretation and a local service interpretation in the preset model interpretation.

For example, if the local service interpretation obtained in step S203 conflicts with an expert local interpretation in the expert interpretations, and/or the local service interpretation obtained in step S203 conflicts with a local service interpretation in the preset model interpretation, the service processing device may determine that the local service interpretation obtained in step S203 does not conform to the service logic. If the local service explanation obtained in the step S203 does not conflict with the expert local explanation in the expert explanation, and the local service explanation obtained in the step S203 does not conflict with the local service explanation in the preset model explanation, the service processing device may determine that the local service explanation obtained in the step S203 conforms to the service logic.

Optionally, when the service processing device determines that the local service interpretation obtained in step S203 conforms to the service logic according to the expert local interpretation in the expert interpretation and/or the local service interpretation in the preset model interpretation, it may further determine whether the local service interpretation obtained in step S203 conforms to the service logic by further combining the expert global interpretation in the expert interpretation and/or the global service interpretation in the preset model interpretation.

Further, when it is determined that the local service interpretation conforms to the service logic, step S303 is executed; upon determining that the local business interpretation does not conform to the business logic, step S304 is performed.

Step S303, obtaining the contribution degrees of the features of the service data to the prediction result of the target variable, taking the feature corresponding to each contribution degree greater than a preset contribution degree in the contribution degrees of the features to the prediction result as an optimization feature, and determining a user group related to the service to be processed based on the optimization feature.

In this step, the business processing device may obtain, according to the local business interpretation obtained in step S203, a contribution degree of each feature of the business data of the client to a prediction result of the target variable. Then, the business processing device may use, as an optimization feature (i.e., a feature having a large influence on the preset result of the target variable), a feature corresponding to each contribution degree greater than a preset contribution degree in the contribution degrees of the features to the prediction result. Then, the service processing device may determine, based on the optimization characteristics, a user group related to the service to be processed.

For example, if the pending service request is a service application request for credit, the optimization features may include, but are not limited to: the service processing equipment can also screen and screen a user group (or a customer group) suitable for the service to be processed based on the optimization characteristics, so as to provide strategic support for the development of the service to be processed.

Step S304, re-determining the business data to be acquired related to the business processing model according to the global business interpretation in the preset model interpretation.

In this step, the service processing device may re-determine the service data to be acquired related to the service processing model according to the global service interpretation in the preset model interpretation corresponding to the service to be processed, so that the service data more related to the service processing model may be acquired according to the service request to be processed next time, which is not only beneficial to improving the accuracy of the prediction result of the target variable of the service processing model, but also beneficial to improving the accuracy of the local service interpretation; on the other hand, the service data more relevant to the service processing model can be adopted when the service processing model is updated next time, so that the model training efficiency can be improved.

Fig. 4 is a flowchart illustrating a service processing method according to another embodiment of the present application. On the basis of the foregoing embodiments, in the embodiments of the present application, a manner of acquiring the service processing model is described. As shown in fig. 4, the method of the embodiment of the present application may include:

step S401, sample data is obtained and preprocessed.

For example, the sample data in the embodiment of the present application may include business data corresponding to a plurality of samples (or a plurality of clients); the service data corresponding to any client may include, but is not limited to: the identification information of the client, the business data related to the business processing model corresponding to the client and the real result of the target variable of the business processing model corresponding to the client. For example, for an application scenario where credit is consumed, the business data related to the business process model corresponding to the customer may include, but is not limited to, at least one of: the system comprises client attribute characteristic data, client payment behavior characteristic data, financial management characteristic data, historical repayment behavior characteristic data (used for indicating a real result of a target variable corresponding to the client), consumption behavior characteristic data, social relationship network characteristic data and loan credit investigation characteristic data.

In this step, the service processing device may obtain the sample data from a database or other devices, and preprocess the sample data so as to improve the efficiency of model training, where the preprocessing may include, but is not limited to: classification processing, feature derivation processing and screening processing.

For example, the service processing device may first divide the service data corresponding to each sample (or client) in the sample data into at least one of the following categories: categorical characteristic data (e.g., gender, occupation, academic calendar, etc. of the customer), sequential category characteristic data (e.g., age, number of paid strokes, etc. of the customer), textual category characteristic data (e.g., address, comment text data, etc.), relational network category characteristic data (e.g., social relational network characteristic data, etc.), browsing sequence characteristic data (e.g., shopping browsing information, ordering behavior, etc.).

Further, the service processing device may perform feature derivation processing on each classified data, so that the feature data corresponding to each sample (or client) in the sample data is more complete.

1) For the classification feature data: the business processing device may perform feature combination on the classification feature data through violence derivation, occupation ratio calculation, one-hot (one-hot) coding, and/or evidence Weight (WOE) coding.

2) For the above continuous class feature data: the service processing device may obtain the statistical description characteristics corresponding to the continuous characteristic data by taking a maximum value, a minimum value, a quantile (e.g., a 50% quantile, a 75% quantile, etc.), time slicing, summarizing, averaging, and/or calculating a ratio.

3) For the text type feature data: the service processing device can perform word vector processing on the text feature data.

4) For the above relational network class feature data: the business processing equipment can analyze the relational network type characteristic data and mine characteristic information related to the target variable. For example, for an application scenario of credit consumption, the business processing device may analyze the relational network class feature data, and mine default risk information related to the customer, such as how many people in the main related people are overdue and the like.

5) For the browsing sequence feature data: the service processing device may perform feature mining on the browsing sequence feature data in an embedding (embedding) manner, a drawing embedding manner, or the like.

Further, the service processing device may perform a screening process on derived sample data (which may include original feature data and derived feature data obtained by the deriving process) obtained after the feature deriving process to obtain preprocessed sample data, where the screening process may include, but is not limited to, at least one of the following: uniqueness processing, missing value processing and extreme value processing. It should be understood that the above derived sample data may exist in the form of a wide table in the above service processing device, and of course, may also exist in other forms, which is not limited in this embodiment of the application.

1) And (3) uniqueness processing: the method refers to deleting the business data corresponding to the repeated other samples in the derived sample data according to the identification information of the samples (or clients). For example, for the same client, the derived sample data includes a set of service data corresponding to the client.

2) Missing value processing may include: transverse missing value processing and longitudinal missing value processing. The horizontal missing value processing refers to deleting the business data corresponding to the sample (or client) lacking the key features. For example, if key feature data such as gender, age, and academic calendar are missing in the business data corresponding to a certain sample (or customer), the business data corresponding to the sample (or customer) may be deleted from the derived sample data. The longitudinal missing value processing refers to deleting the feature data of which the missing rate is greater than a preset missing rate in the derivative sample data, and filling the feature data of which the missing rate is not greater than the preset missing rate in the derivative sample data according to a preset rule (for example, a preset filling rule and the like). For example, when the missing rate of the number of foreign consumption times in the derived sample data is greater than a preset certainty rate, the list of feature data of the number of foreign consumption times in the derived sample data may be deleted.

3) Extreme value processing: setting data which is greater than a preset upper threshold value in the derivative sample data as the preset upper threshold value.

S402, training a pre-configured initialized service processing model according to the preprocessed sample data to obtain the service processing model.

In this step, the service processing device may determine, according to the service data to be obtained related to the service processing model, service data related to the service processing model from the pre-processed sample data obtained in step S401, and then train the pre-configured initialized service processing model according to the service data related to the service processing model until the model prediction index meets the preset service index, so as to obtain the service processing model, so that when a service request to be processed is received, a prediction result of a target variable corresponding to the service data related to the service processing model may be accurately determined according to the service processing model. Illustratively, the model predictors may include, but are not limited to, at least one of: accuracy index, Area Under ROC Curve (AUC) index, KS (Kolmogorov-Smirnov) test index, etc.

For example, on one hand, the business processing device may select the first feature according to interactivity between features of the business processing model, a degree of distinguishing the features from a target variable, and/or stability of the features in sample data; on the other hand, the service processing device may cross the first feature with the feature importance ranking corresponding to the service processing model to obtain a second feature, so as to use data corresponding to the second feature as service data to be acquired related to the service processing model.

For another example, the service processing device may further determine, through a global service interpretation in the preset model interpretation, to-be-acquired service data related to the service processing model, or may further determine, through a global service interpretation corresponding to a preset global interpretation mode according to an embodiment of the present application, to-be-acquired service data related to the service processing model.

Of course, the service processing device may also determine the service data to be acquired, which is related to the service processing model, in other manners, which is not limited in this embodiment of the application.

Further, on the basis of the above embodiment, the service processing device may further perform interpretation processing on the preprocessed sample data in a preset global interpretation manner, and determine the service data to be acquired, which is related to the service processing model, according to an interpretation result. On one hand, the service processing equipment can carry out targeted training on the pre-configured initialized service processing model or further carry out updating training on the service processing model according to the service data corresponding to the service data to be acquired in the pre-processed sample data, so that the training efficiency is improved; on the other hand, the service processing equipment can acquire service data more relevant to the service processing model according to the service request to be processed, and the accuracy of the prediction result of the target variable of the service processing model is improved.

For example, if the global interpretation indicates that the degree of contribution of the feature data 1 to the feature data 10 to the prediction result of the target variable of the business process model is large, the business process device may determine the feature data 1 to the feature data 10 as the business data to be acquired related to the business process model.

Fig. 5 is a schematic flow chart of a service processing method according to another embodiment of the present application, and based on the foregoing embodiment, the present application introduces an implementation manner for explaining processing and determining service data to be acquired. As shown in fig. 5, the method of the embodiment of the present application may include:

step S501, for each feature of the preprocessed sample data, obtaining importance indication information of the feature.

In this step, for each feature of the preprocessed sample data, the service processing device may calculate importance indication information of the feature (for indicating the importance of the feature with respect to the service processing model), so that whether the interactivity of each feature with other features is obvious or not may be subsequently determined according to the importance indication information in an order from high to low, and/or service interpretation information of the service processing model (for example, the local service interpretation and/or the global service interpretation) may be output. For example, the importance indication information may include, but is not limited to: importance information and/or weight assignment information.

For each feature (or referred to as feature data) of the preprocessed sample data, the service processing device may obtain importance indication information of the feature by way of substitution (persistence) or SHAP.

For another example, if the business process model is a tree model, for each feature (or referred to as feature data) of the preprocessed sample data, the business process device may obtain the importance indication information of the feature by using an average kini index, the number of times of feature division, or the number of samples of feature division.

And step S502, judging whether the interactivity of the features and other features is obvious.

In a possible implementation manner, the traffic processing device may calculate correlations between the feature and other features by means of a correlation coefficient, variance analysis, or H Statistic (statistical), and then determine whether the interactivity between the feature and other features is obvious by comparing the correlations between the feature and other features with a preset correlation threshold. If the correlation between the upper feature and the other features is greater than a preset correlation threshold, determining that the interactivity between the above feature and the other features is obvious; if the correlation between the above feature and other features is not greater than the preset correlation threshold, it may be determined that the interactivity between the above feature and other features is not significant.

In another possible implementation manner, the business processing device may determine whether the interactivity between the feature and the other feature is obvious through feature interaction (FeatureInteraction) or a multi-dimensional map drawn by the SHAP.

For example, if a two-dimensional graph is drawn according to feature interaction, the abscissa of the two-dimensional graph may be a value range of the feature, and the ordinate may be a contribution degree of the feature to a prediction result of the target variable; if the three-dimensional graph is drawn according to the feature interaction, the x coordinate of the three-dimensional graph can be the value range of the feature, the y coordinate can be the value range of one other feature, and the z coordinate can interact with the feature as the contribution degree of the prediction result of the target variable.

For another example, if a multi-dimensional map (or referred to as a SHAP map) is drawn according to the SHAP, the SHAP map is used to indicate the degree of contribution of the interaction between the above-mentioned feature and other features to the prediction result of the target variable, and the abscissa may be a value range of the above-mentioned feature, and the ordinate may include two coordinates, where one ordinate is a value range of one other feature and the other ordinate is a SHAP value of the interaction feature.

Of course, the service processing device may also determine whether the interactivity between the feature and the other feature is obvious through other manners, which is not limited in this embodiment of the application.

Further, when it is determined that the interactivity between the feature and other features is not obvious, step S503 is executed; when it is determined that the interactivity of the feature with other features is obvious, step S506 is executed.

Step S503, using an Accumulated Local Effects (ALE) or a Partial dependency graph (PDP), to determine whether the prediction result of the target variable of the feature to the service processing model conforms to the service logic.

The accumulated local effect graph related in the embodiment of the present application may be a graph obtained by solving a difference between prediction results of corresponding target variables by replacing an upper bound and a lower bound in each value interval of the feature, then accumulating and averaging the difference between the prediction results to obtain an influence of a variation of the feature in the value interval on the prediction result of the target variable, and then accumulating the influence of each value interval. For example, the abscissa of the cumulative local effect graph may be a value range of the feature, and the ordinate may be a contribution degree of the feature to the prediction result of the target variable.

Illustratively, the partial dependency graph is used to indicate a marginal effect of the feature on the predicted outcome of the target variable. For example, the abscissa of the partial dependency graph may be a value range of the feature, and the ordinate may be a contribution degree of the feature to the prediction result of the target variable.

In this step, the service processing device may determine, according to the accumulated local effect graph or the partial dependency graph, a global service interpretation corresponding to the prediction result of the feature on the target variable, and then may determine, in combination with an expert interpretation, whether the global service interpretation corresponding to the prediction result of the feature on the target variable conforms to a service logic, that is, whether the prediction result of the feature on the target variable of the service processing model conforms to the service logic.

For example, the service processing device may determine, according to an expert global interpretation in the expert global interpretation, whether a global service interpretation corresponding to a prediction result of the feature on the target variable conforms to a service logic.

For example, if the global business interpretation corresponding to the prediction result of the feature on the target variable conflicts with the expert global interpretation in the expert interpretation, the business processing device may determine that the global business interpretation corresponding to the prediction result of the feature on the target variable does not conform to the business logic, that is, the prediction result of the feature on the target variable of the business processing model does not conform to the business logic. If the global business interpretation corresponding to the prediction result of the target variable by the feature is not in conflict with the expert global interpretation in the expert interpretation, the business processing equipment can determine that the global business interpretation corresponding to the prediction result of the target variable by the feature conforms to business logic, namely that the prediction result of the target variable by the feature conforms to business logic.

Further, when it is determined that the feature conforms to the business logic for the prediction result, step S504 is executed; when it is determined that the feature does not conform to the business logic for the predicted result, step S505 is performed.

Step S504, using the data corresponding to the feature (or referred to as feature data) as the to-be-acquired service data related to the service processing model.

Therefore, the service processing device can determine the service data to be acquired related to the service processing model according to the global interpretation result, which is not only beneficial for the service processing device to perform targeted training on the preconfigured initialization service processing model or further perform update training on the service processing model according to the service data corresponding to the service data to be acquired in the preprocessed sample data, thereby improving the training efficiency; the service processing equipment can acquire service data more related to the service processing model according to the service request to be processed, and accuracy of a prediction result of the target variable of the service processing model is improved.

And step S505, carrying out training processing on the business processing model again.

In this step, the service processing device may further train the existing service processing model to obtain the trained service processing model, or may train the pre-configured initialized service processing model to obtain the retrained service processing model, so that guidance of training the service processing model according to the global interpretation result is achieved, and the accuracy of the model is improved.

And step S506, judging whether the prediction result of the feature on the target variable of the business processing model conforms to business logic by adopting a SHAP graph or an accumulative local effect graph (ALE).

In this step, the service processing device may determine, according to the SHAP diagram or the accumulated local effect diagram, a global service interpretation corresponding to the prediction result of the feature on the target variable, and then may determine, in combination with an expert interpretation, whether the global service interpretation corresponding to the prediction result of the feature on the target variable conforms to a service logic, that is, whether the prediction result of the feature on the target variable of the service processing model conforms to the service logic.

For example, if the global business interpretation corresponding to the prediction result of the feature on the target variable conflicts with the expert global interpretation in the expert interpretation, the business processing device may determine that the global business interpretation corresponding to the prediction result of the feature on the target variable does not conform to the business logic, that is, the prediction result of the feature on the target variable of the business processing model does not conform to the business logic. If the global business interpretation corresponding to the prediction result of the target variable by the feature is not in conflict with the expert global interpretation in the expert interpretation, the business processing device may determine that the global business interpretation corresponding to the prediction result of the target variable by the feature conforms to business logic, that is, the prediction result of the target variable by the feature conforms to business logic.

Further, when it is determined that the prediction result of the feature pair conforms to the business logic, step S507 is executed; when it is determined that the feature does not conform to the business logic for the predicted result, step S505 is performed.

Step S507, using the data corresponding to the feature (or referred to as feature data) as the to-be-acquired service data related to the service processing model.

Optionally, after the service processing device executes the step S507, the service processing device may further perform derivation processing on the feature to obtain a derived new feature, and when it is determined that the prediction result of the new feature conforms to the service logic, data (or referred to as feature data) corresponding to the new feature is used as service data to be obtained related to the service processing model, so as to reduce interference of the model.

Fig. 6 is a schematic flow chart of a service processing method according to another embodiment of the present application, and based on the above embodiment, an implementation manner of a training process of the service processing model is introduced in the embodiment of the present application by combining the preset local interpretation manner and the global interpretation manner. As shown in fig. 6, the method of the embodiment of the present application may include:

step S601, sample data is obtained.

Step S602, preprocessing the sample data.

The implementation manners of steps S601 and S602 may refer to the related contents of step S401, and are not described herein again.

Step S603, determining service data (or referred to as feature data or variable data) related to the service processing model from the sample data according to the service data to be acquired related to the service processing model.

And step S604, training a business processing model.

The implementation manners of steps S603 and S604 may refer to the related contents of step S402, which is not described herein again.

And step S605A, performing interpretation processing by adopting a preset global interpretation mode to obtain global service interpretations of different feature data.

The implementation manner of this step may refer to the related contents in steps S501 to S507, which are not described herein again.

Step S605B, performing interpretation processing by using a preset local interpretation mode to obtain local service interpretations of different feature data.

The implementation manner of this step can refer to the relevant contents in step S203 and steps S301 to S304, and will not be described herein again.

Step S606, judging whether the global service explanation and the local service explanation of the different characteristic data accord with the service logic.

Illustratively, the business processing device may determine whether the global business interpretations of the different features conform to the business logic according to the expert global interpretations in the obtained expert interpretations, and determine whether the local business interpretations of the different feature data conform to the business logic according to the expert local interpretations in the obtained expert interpretations.

Further, if the global service interpretation and the local service interpretation of the different feature data both conform to the service logic, step S607 is executed; if the global service interpretation and/or the local service interpretation of any feature data do not conform to the service logic, the service processing device may return to execute the processes of the above step S603 to step S606 again according to each feature data conforming to the service logic and the feature data not selected in the above step S603 until the global service interpretation and the local service interpretation of all the selected feature data conform to the service interpretation.

And step S607, determining a service processing strategy according to the global service explanation and the local service explanation.

In this step, the service processing device may obtain, according to the global service interpretation conforming to the service logic, contribution ranking of different features to prediction results of target variables of all sample data, and determine service data to be obtained related to the service processing model, so that on one hand, the service processing device may obtain, according to a service request to be processed, service data more related to the service processing model, which is beneficial to improving accuracy of the prediction results of the target variables of the service processing model, and meanwhile, may also display, according to the basis, handling results of service handling to service staff or clients, thereby improving rationality and accuracy of service handling; on the other hand, the method is convenient for screening and screening the client group suitable for the service to be processed in the client obtaining stage, is beneficial to realizing accurate marketing and improving the client obtaining quality and the conversion rate.

In addition, the service processing equipment can obtain the contribution degree sequence of different characteristics to the prediction result of the target variable of single sample data according to the local service explanation which accords with the service logic, so that the characteristics with higher contribution degree of the prediction result can be rechecked and checked in the subsequent service auditing process, and a client group suitable for the service to be processed can be determined in a client obtaining stage by combining the local service explanation, and the handling basis of the service handling result can be displayed to service personnel or clients according to the basis, thereby improving the model reliability.

For example, for the application scenario of credit consumption, assuming that the local business interpretation indicates that the "number of overdue" feature has a large influence on the target variable (such as risk probability of customer default), and that customers with more overdue are more susceptible to default, the business processing equipment can enhance the differentiation and audit of the number of overdue customers in the process of business audit (or referred to as risk control).

Fig. 7 is a schematic structural diagram of a service processing apparatus according to an embodiment of the present application. As shown in fig. 7, a device for processing a service provided in an embodiment of the present application may include: a first obtaining module 701, a second obtaining module 702, and a third obtaining module 703.

The first obtaining module 701 is configured to obtain service data related to a service processing model according to a service request to be processed;

a second obtaining module 702, configured to process the service data by using the service processing model to obtain a prediction result of a target variable of the service processing model;

a third obtaining module 703, configured to interpret the prediction result of the target variable in a preset local interpretation manner, to obtain a local service interpretation corresponding to the prediction result of the target variable; the preset local interpretation mode is a calculation mode used for determining a local service interpretation corresponding to the prediction result.

In one possible implementation, the apparatus further includes:

wherein the fifth obtaining module is configured to:

wherein the sixth obtaining module is configured to:

wherein the seventh obtaining module is configured to:

acquiring sample data and preprocessing the sample data;

wherein the eighth obtaining module is configured to:

In a possible implementation manner, the eighth obtaining module is specifically configured to:

In a possible implementation manner, the eighth obtaining module is further configured to:

The service processing apparatus provided in this embodiment of the present application may be configured to execute the technical solution in the foregoing service processing method embodiment of the present application, and the implementation principle and the technical effect are similar, which are not described herein again.

Fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application. For example, the electronic device provided in the embodiment of the present application may be a service processing device in the above embodiment of the present application. As shown in fig. 8, an electronic device provided in an embodiment of the present application may include: a memory 801, a processor 802, and a computer program stored on the memory 801 and executable on the processor 802. Illustratively, the electronic device may further comprise a communication interface 803 for communicating with other devices, wherein the memory 801, the processor 802 and the communication interface 803 may be connected by a system bus.

Meanwhile, when the processor 802 executes the computer program, the technical solution in the embodiment of the service processing method of the present application is implemented, and the implementation principle and the technical effect are similar, and are not described herein again.

Optionally, the Processor may be a Central Processing Unit (CPU), or may be another general-purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

In an exemplary embodiment, the electronic device may also be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

Optionally, the memory may include a high-speed RAM memory, and may further include a non-volatile memory NVM, such as at least one disk memory.

Alternatively, the system bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended ISA (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, the buses in the figures of the present application are not limited to only one bus or one type of bus.

An embodiment of the present application further provides a computer-readable storage medium, where a computer execution instruction is stored in the computer-readable storage medium, and the computer execution instruction is executed by a processor to implement the technical solution in the embodiment of the service processing method in the present application, and the implementation principle and the technical effect are similar, and are not described herein again.

Illustratively, the memory (storage medium) described above may be implemented by any type of volatile or non-volatile storage device or combination thereof, such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disk. Readable storage media can be any available media that can be accessed by a general purpose or special purpose computer.

It should be understood by those of ordinary skill in the art that, in the various embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of the processes should be determined by their functions and inherent logic, and should not limit the implementation process of the embodiments of the present application.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for processing a service, comprising:

2. The method of claim 1, further comprising:

3. The method of claim 2, further comprising:

4. The method of claim 2, further comprising:

5. The method according to any one of claims 1 to 4, wherein the obtaining manner of the business process model comprises:

acquiring sample data and preprocessing the sample data;

6. The method of claim 5, further comprising:

7. The method according to claim 6, wherein the interpreting the preprocessed sample data in a preset global interpretation manner, and determining the service data to be acquired related to the service processing model according to the interpretation result, comprises:

8. The method of claim 7, further comprising:

9. The method of claim 8, further comprising:

10. The method of claim 7 or 8, further comprising:

11. An apparatus for processing traffic, comprising:

12. The apparatus of claim 11, further comprising:

13. The apparatus of claim 12, further comprising: a fifth obtaining module;

wherein the fifth obtaining module is configured to:

14. The apparatus of claim 12, further comprising: a sixth obtaining module;

wherein the sixth obtaining module is configured to:

15. The apparatus of any one of claims 11 to 14, further comprising: a seventh obtaining module;

wherein the seventh obtaining module is configured to:

acquiring sample data and preprocessing the sample data;

16. The apparatus of claim 15, further comprising: an eighth obtaining module;

wherein the eighth obtaining module is configured to:

17. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of processing a service according to any of claims 1 to 10 when executing the computer program.

18. A computer-readable storage medium, having stored thereon computer-executable instructions for implementing a method of processing a service according to any one of claims 1 to 10 when executed by a processor.