Strategy classification model construction method, strategy processing method, apparatus and device, computer program product and readable storage medium

Abstract

The invention provides a strategy classification model construction method, a strategy processing method and device, equipment, a computer program product and a readable storage medium. The method comprises the steps that a plurality of sample strategies are acquired, a strategy mark of each sample strategy is acquired, the strategy marks are used for representing whether the sample strategies are high-quality strategies or poor-quality strategies, and the sample strategies are used for carrying out risk control on services; a plurality of test indexes of each sample strategy are obtained, and the test indexes are used for quantifying the test sample strategies; obtaining the index weight of the sample strategy corresponding to each test index; constructing a strategy index set according to a plurality of index weights corresponding to each sample strategy; a strategy classification model is trained through the strategy index set, the strategy mark and a preset loss function, and the strategy classification model is used for predicting the probability that the to-be-tested strategy is a high-quality strategy. According to the method and the device, the strategy classification model can be constructed so as to be used for objectively and comprehensively testing the business strategy for risk control.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method for constructing a policy classification model, a policy processing method, an apparatus, a device, a computer program product, and a readable storage medium. Background Art

[0002] There are already various strategies in related technologies for risk control. For example, in electronic payment scenarios, relevant strategies are used to intercept illegal behaviors (such as false transactions, etc.) to implement operations such as illegal behavior interception, thereby performing risk control. Therefore, the quality of the strategy will directly affect the effectiveness of risk control in various business scenarios.

[0003] The strategic testing of related technologies mostly comes from test indicators directly specified by the business. The way to assign weights to test indicators is mostly through the entropy weight method or the Analytic Hierarchy Process (AHP). Among them, the entropy weight method relies on the volatility of the test indicator itself, ignores the importance of the test indicator itself, and cannot match the business cognition. The Analytic Hierarchy Process is subjective and qualitative, and the test results are not objective and accurate enough. Summary of the Invention

[0004] The embodiments of the present application provide a method for constructing a policy classification model, a policy processing method, an apparatus, a device, a computer program product, and a computer-readable storage medium, which can construct a policy classification model for objectively and comprehensively testing the risk control strategies of a business.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present invention provides a method for constructing a policy classification model, the method comprising:

[0007] Acquire multiple sample strategies and acquire a strategy tag for each of the sample strategies, wherein the strategy tag is used to characterize whether the sample strategy is a high-quality strategy or a low-quality strategy, and the sample strategy is used to perform risk control on the business;

[0008] Acquire a plurality of test indicators for each of the sample strategies, wherein the test indicators are used to quantitatively test the sample strategies;

[0009] Obtaining an indicator weight of each of the test indicators corresponding to the sample strategy, wherein the indicator weight is used to quantitatively represent the importance of the test indicator when testing the sample strategy;

[0010] Constructing a strategy indicator set according to the plurality of indicator weights corresponding to each of the sample strategies;

[0011] A strategy classification model is trained using the strategy indicator set, the strategy label, and a preset loss function, wherein the strategy classification model is used to predict the probability that the strategy to be tested is a high-quality strategy.

[0012] The present invention provides a policy processing method for a policy classification model, the method comprising:

[0013] Obtain the risk control strategy to be tested;

[0014] Obtaining, by means of the strategy classification model, a predicted probability that the risk control strategy to be tested is a high-quality strategy;

[0015] The predicted probability is subjected to score conversion processing to obtain a strategy score of the risk control strategy to be tested.

[0016] An embodiment of the present application provides a device for constructing a policy classification model, including:

[0017] an acquisition module, configured to acquire a plurality of sample strategies and acquire a strategy tag of each of the sample strategies, wherein the strategy tag is used to characterize whether the sample strategy is a high-quality strategy or a low-quality strategy, and the sample strategy is used to perform risk control on a business;

[0018] A construction module is used to obtain a plurality of test indicators for each of the sample strategies, wherein the test indicators are used to quantitatively test the sample strategies;

[0019] The construction module is further configured to construct a sample strategy indicator set using the multiple test indicators of each sample strategy;

[0020] The construction module is further configured to obtain an indicator weight of each test indicator corresponding to the sample strategy, wherein the indicator weight is used to quantitatively represent the importance of the test indicator when testing the sample strategy;

[0021] The construction module is further configured to reconstruct the sample strategy indicator set according to each indicator weight to obtain a strategy indicator set;

[0022] A training module is used to train a strategy classification model using the strategy indicator set, the strategy label and a preset loss function, wherein the strategy classification model is used to predict the probability that the strategy to be tested is a high-quality strategy.

[0023] The present invention provides a policy processing device for a policy classification model, including:

[0024] Acquisition module, used to obtain the risk control strategy to be tested;

[0025] A prediction module, configured to obtain, through the strategy classification model, a predicted probability that the risk control strategy to be tested is a high-quality strategy;

[0026] The scoring module is used to perform scoring conversion processing on the predicted probability to obtain a strategy score of the risk control strategy to be tested.

[0027] An embodiment of the present application provides an electronic device, including:

[0028] a memory for storing computer-executable instructions;

[0029] The processor is used to implement the method for constructing the policy classification model provided in the embodiment of the present application or the policy processing method of the policy classification model provided in the embodiment of the present application when executing the computer executable instructions stored in the memory.

[0030] An embodiment of the present application provides a computer-readable storage medium storing a computer program or computer-executable instructions, which, when executed by a processor, implements the method for constructing a policy classification model provided in an embodiment of the present application or the policy processing method for the policy classification model provided in an embodiment of the present application.

[0031] An embodiment of the present application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, the method for constructing a policy classification model provided in the embodiment of the present application or the policy processing method of the policy classification model provided in the embodiment of the present application is implemented.

[0032] The embodiments of the present application have the following beneficial effects:

[0033] By systematically building test indicators through the strategy analysis model, we obtain the indicator data of multiple test indicators of the sample strategy, and then obtain the corresponding indicator weight of each test indicator. This allows us to objectively assign weights to each test indicator based on the importance of the test indicator itself. Finally, we conduct a comprehensive and objective test of the strategy through the trained strategy classification model, thereby providing valuable reference information for discovering high-quality strategies in the business, as well as redundant or degraded low-quality strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a schematic diagram of the structure of the risk control system provided by the embodiment of the present application;

[0035] Figure 2A 1 is a schematic structural diagram of a policy management server 100-1 for constructing a policy classification model provided in an embodiment of the present application;

[0036] Figure 2B1 is a schematic structural diagram of a policy management server 100-2 for policy scoring of a policy classification model provided in an embodiment of the present application;

[0037] Figure 3A This is a first flow chart of a method for constructing a policy classification model provided in an embodiment of the present application;

[0038] Figure 3B This is a second flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0039] Figure 3C This is a third flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0040] Figure 3D This is a fourth flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0041] Figure 3E This is a fifth flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0042] Figure 3F This is a sixth flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0043] Figure 3G This is a seventh flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0044] Figure 3H This is an eighth flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0045] Figure 3I This is a ninth flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0046] Figure 3J This is a tenth flow chart of a method for constructing a policy classification model provided in an embodiment of the present application;

[0047] Figure 3K This is the eleventh flow chart of the method for constructing a policy classification model provided in an embodiment of the present application;

[0048] Figure 4A This is a schematic diagram of the first interface of the strategy analysis model provided in the embodiment of the present application;

[0049] Figure 4B This is a schematic diagram of the second interface of the strategy analysis model provided in the embodiment of the present application;

[0050] Figure 4C This is a schematic diagram of the third interface of the strategy analysis model provided in the embodiment of the present application;

[0051] Figure 5A This is a first flow chart of the policy processing method of the policy classification model provided in an embodiment of the present application;

[0052] Figure 5B This is a second flow chart of the policy processing method of the policy classification model provided in an embodiment of the present application;

[0053] Figure 5C This is a third flow chart of the policy processing method of the policy classification model provided in an embodiment of the present application;

[0054] Figure 6A This is a flow chart of a strategy analysis model for building an electronic payment scenario provided by an embodiment of the present application;

[0055] Figure 6B This is a first flow chart of building a strategy scorecard in an electronic payment scenario provided by an embodiment of the present application;

[0056] Figure 6C This is a second flow diagram of building a strategy scoring card in an electronic payment scenario provided by an embodiment of the present application;

[0057] Figure 7A This is a flow chart of building a strategy analysis model in a credit scenario provided by an embodiment of the present application;

[0058] Figure 7B This is a flow chart of building a strategy scorecard in a credit scenario provided by an embodiment of the present application;

[0059] Figure 8A This is a schematic diagram of the business process impact link in the electronic payment scenario provided by an embodiment of the present application;

[0060] Figure 8B This is a schematic diagram of the business process impact link in the credit scenario provided by the embodiment of the present application;

[0061] Figure 9A This is a first schematic diagram of a binning trend graph provided in an embodiment of the present application;

[0062] Figure 9B This is a second schematic diagram of a binning trend graph provided in an embodiment of the present application;

[0063] Figure 10 It is a schematic diagram of the network structure of the policy classification model provided in the embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0065] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0066] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0067] The relevant data collection and processing in the embodiments of this application should be strictly in accordance with the requirements of relevant national laws and regulations when applied in examples, and the informed consent or separate consent of the personal information subject should be obtained. Subsequent data use and processing should be carried out within the scope of authorization of laws and regulations and the personal information subject.

[0068] Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application have the same meanings as those commonly understood by those skilled in the art. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0069] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0070] 1) Business: A general term for services provided based on the Internet, such as Internet-based electronic payment and online credit.

[0071] 2) The strategy analysis model, also known as the Objective-Strategy-Funnel-Measurement (OSFM) model, is different from the algorithmic model in that it is a business analysis framework designed based on human experience. It is suitable for situations where the goals and strategies are clear. It breaks down business goals and implements them into specific, feasible, and measurable test indicators to ensure that the strategy does not deviate from the business goals.

[0072] 3) Strategy, also known as business strategy, includes a series of risk control rules designed by the business operation and maintenance party, which are used to control business risks, such as controlling the risks of electronic payment-based violations (such as false transactions) and controlling the default risks of online credit.

[0073] 4) Objectives, short for business objectives, are the goals achieved by implementing a business strategy. For example, in the case of electronic payment, business objectives could be risk margin and risk control intervention. Risk margin quantifies the effectiveness of the sample strategy after implementation, such as the total amount of fraudulent transactions. Risk control intervention indicates the extent to which the sample strategy interferes with the business during implementation, such as the number and amount of fraudulent transactions affected by the strategy.

[0074] 5) Funnel Analysis, also known as Funnel Analysis, is a data analysis method used to evaluate the conversion rate or loss of each link in the user conversion process of multiple links in the business chain, from initial interaction to final goal completion.

[0075] 6) Indicators, i.e. test indicators, are parameters used to test strategies derived by calculating the conversion rate and conversion volume of each link in the strategy's impact chain through a strategy analysis model.

[0076] 7) The weight value of the interval, also known as the indicator weight or weight of evidence (WoE), is used to characterize the degree and direction of the influence of the test indicator on whether the strategy is an inferior strategy within the current interval. For example, when the WOE is positive, the current value of the test indicator has a positive influence on whether the strategy is an inferior strategy. When the WOE is negative, the current value of the test indicator has a negative influence on whether the strategy is an inferior strategy. The size of the WOE value reflects the degree of this influence.

[0077] 8) Information value (IV) is also called information value. IV is calculated based on word of edge (WOE) and can be used as a weighted sum of WOE. IV is used to reflect the predictive ability of the test indicator.

[0078] First, the strategic testing indicators of related technologies are mostly directly specified by the business, and there is a problem that the indicators are not comprehensive and cannot cover the entire business process. Secondly, the method of weighting test indicators in related technologies is mostly through entropy weight method or hierarchical analysis method (Analytic Hierarchy Process, AHP) to weight various test indicators. However, the entropy weight method relies on the volatility of the test indicator itself, ignores the importance of the test indicator itself, and cannot match the business cognition. The hierarchical analysis method is subjective weighting, biased towards qualitative, and the test results are not objective enough. Therefore, the current indicator weighting method cannot ensure that the business can be explained while objectively weighting various indicators.

[0079] In order to solve the above problems, the embodiments of the present application provide a method for constructing a policy classification model, a policy processing method, an apparatus, a device, a computer program product and a computer-readable storage medium, which can systematically obtain multiple test indicators of sample strategies in combination with business cognition, thereby constructing a sample strategy indicator set and obtaining the indicator weight corresponding to each test indicator, thereby realizing the objective empowerment of each test indicator based on the importance of the test indicator itself, and finally achieving the beneficial effect of comprehensive and objective testing of the strategy through the trained policy classification model, thereby assisting in discovering redundant or degraded low-quality strategies in the business, ensuring the high-quality and streamlined strategy set, and avoiding additional system risks caused by the increase in strategies during business development.

[0080] See also Figure 1 , Figure 1 This is a schematic diagram of the structure of the risk control system architecture provided by the embodiment of the present application, for example, Figure 1 The present invention involves a policy management server 100, a business server 200, a terminal device 300, and a network 400. The terminal device 300 is connected to the policy management server 100 and the business server 200 via the network 400. The network 400 can be a wide area network or a local area network, or a combination of the two. The risk control system provided in the embodiments of the present application can be applied to scenarios requiring risk control, such as electronic payment and credit.

[0081] Taking the application of the embodiment of the present application to the electronic payment scenario as an example, the risk control system provided by the embodiment of the present application can be implemented by the server and the terminal in collaboration. For example, the terminal device 300 runs a client, such as an instant messaging client (including an electronic payment function) or an electronic payment client, which supports users to make electronic payments, such as transfers, online shopping, etc. When the client receives an electronic payment operation from a target object (for example, a user account currently logged in in the instant messaging client), the terminal device 300 sends an electronic payment request (corresponding to the target object) to the business server 200. Figure 1), after receiving the electronic payment request, the business server 200 sends a request to the policy management server 100 to obtain the risk control policy in the electronic payment scenario. After receiving the request to obtain the risk policy, the policy management server 100 performs policy scoring through the policy classification model construction method and the policy processing method of the policy classification model provided in the embodiment of the present application, and sends the risk control policy (such as multi-factor identity verification policy, transaction limit policy, transaction frequency control policy, abnormal transaction pattern recognition policy and data encryption policy, etc.) with a quality score that reaches the score threshold to the business management server 200. The business manager 200 performs risk control processing on the electronic payment request of the target object through the received risk control policy (such as confirming the identity of the target object through the multi-factor identity verification policy, intercepting or releasing the target object's current payment operation through the transaction limit policy, etc.), and sends the risk control processing result to the terminal device 300.

[0082] Taking the application of the embodiment of the present application in a credit scenario as an example, in some embodiments, the risk control system provided by the embodiment of the present application can be implemented by the server and the terminal in collaboration. For example, the terminal device 300 runs a client, such as an instant messaging client (including an electronic payment function) or an electronic payment client, which supports users to make electronic payments, such as transfers, online shopping, etc. When the client receives a loan application operation from a target object (for example, a user account currently logged in in the instant messaging client), the terminal device 300 sends a loan application request (corresponding to the target object) to the business server 200. Figure 1 ), after receiving the loan application request, the business server 200 sends a request to the policy management server 100 to obtain the risk control strategy in the credit scenario. After receiving the risk strategy request, the policy management server 100 performs strategy scoring through the strategy classification model construction method and the strategy processing method of the strategy classification model provided in the embodiment of the present application. The strategy scoring is used to quantitatively characterize the quality of the risk control strategy to be tested, and sends the risk control strategy whose quality score reaches the score threshold (such as the credit testing strategy in the pre-loan stage, etc.; the repayment supervision strategy, loan condition adjustment strategy, etc. in the loan stage; the credit repair strategy, recovery strategy, credit management strategy, etc. in the post-loan stage) to the business management server 200. The business manager 200 processes the target object's loan application request through the received risk control strategy (for example, intercepting or releasing the target object's current loan application operation through the credit testing strategy, etc.), and sends the risk control processing result to the terminal device 300.

[0083] Here, the policy management server 100 can be a single server. In this case, the method for constructing a policy classification model provided by the embodiment of the present application and the policy processing method for the policy classification model provided by the embodiment of the present application can be implemented by the same server. The policy management server 100 can also be a cluster of servers. In the case where the policy management server 100 is a server cluster, the method for constructing a policy classification model provided by the embodiment of the present application and the policy processing method for the policy classification model provided by the embodiment of the present application can be implemented by different servers. For example, one server is used to construct a policy classification model, and another server is used to deploy the constructed policy classification model to provide the risk control service in the electronic payment scenario or credit scenario described above to the terminal. The embodiment of the present application does not limit this.

[0084] In some embodiments, the terminal or server can implement the construction method of the policy classification model provided by the embodiment of the present application and the policy processing method of the policy classification model provided by the embodiment of the present application by running various computer executable instructions or computer programs. For example, computer executable instructions can be commands, machine instructions or software instructions at the microprogram level. The computer program can be a native program or software module in the operating system; it can be a local (Native) application (APP, Application), that is, a program that needs to be installed in the operating system to run, such as an instant messaging application or an electronic payment application; it can also be a small program embedded in any APP, that is, a program that only needs to be downloaded to a browser environment and can be run. In short, the above-mentioned computer executable instructions can be instructions in any form, and the above-mentioned computer program can be an application, module or plug-in in any form.

[0085] In some embodiments, the policy management server 100 and the business server 200 can be independent physical servers, or a server cluster or distributed system composed of multiple physical servers. They can also be cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. Among them, cloud services can be interactive processing services for terminals to call.

[0086] In some embodiments, multiple servers may form a blockchain network, with the policy management server 100 or the business server 200 being a node on the blockchain. Information connections may exist between each node in the blockchain, and information may be transmitted between nodes via such connections. Data related to the method for constructing a policy classification model and the policy processing method for the policy classification model provided in the embodiments of the present application may be stored on the blockchain.

[0087] See also Figure 2A , Figure 2A is a structural diagram of a policy management server 100-1 for constructing a policy classification model provided in an embodiment of the present application. Figure 2A The policy management server 100-1 shown includes: at least one processor 110-1, a memory 130-1, and at least one network interface 120-1. The various components in the policy management server 100-1 are coupled together via a bus system 140-1. It is understood that the bus system 140-1 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 140-1 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 140-1 is not described in detail. Figure 2A In FIG. 1 , various buses are labeled as bus system 140 - 1 .

[0088] Processor 110-1 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0089] Memory 130-1 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. Memory 130-1 may optionally include one or more storage devices physically located remotely from processor 110-1.

[0090] The memory 130-1 includes volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 130-1 described in the embodiments of the present application is intended to include any suitable type of memory.

[0091] In some embodiments, the memory 130 - 1 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplified below.

[0092] Operating system 131-1, including system programs for processing various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, which are used to implement various basic services and process hardware-based tasks;

[0093] A network communication module 132-1, configured to communicate with other electronic devices via one or more (wired or wireless) network interfaces 120-1. Exemplary network interfaces 120-1 include Bluetooth, Wi-Fi, and USB.

[0094] In some embodiments, the apparatus provided in the embodiments of the present application may be implemented in software. Figure 2A The device 133 for constructing a policy classification model stored in memory 130-1 is shown. This device 133 can be software in the form of a program or plug-in, and includes the following software modules: an acquisition module 1331, a construction module 1332, and a training module 1333. These modules are logical and can be arbitrarily combined or further separated based on the functions they implement. The functions of each module will be described below.

[0095] See also Figure 2B , Figure 2B 1 is a schematic diagram of the structure of a policy management server 100-2 for policy scoring of a policy classification model provided in an embodiment of the present application. Figure 2B The policy management server 100-2 shown includes: at least one processor 110-2, a memory 130-2, and at least one network interface 120-2. The various components in the policy management server 100-2 are coupled together via a bus system 140-2. It is understood that the bus system 140-2 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 140-2 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 140-2 is not described in detail. Figure 2B In the figure, various buses are labeled as bus system 140-2. The detailed description of processor 110-2 and memory 130-2 is as above and will not be repeated here.

[0096] In some embodiments, the apparatus provided in the embodiments of the present application may be implemented in software. Figure 2BThe policy processing device 134 of the policy classification model stored in the memory 130-2 is shown. This device can be software in the form of a program or plug-in, and includes the following software modules: an acquisition module 1341, a prediction module 1342, and a scoring module 1343. These modules are logical and can be arbitrarily combined or further divided according to the functions they implement. The functions of each module will be described below.

[0097] In other embodiments, the device provided in the embodiments of the present application can be implemented in hardware. As an example, the device provided in the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the policy classification model construction method and the policy processing method of the policy classification model provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can adopt one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs) or other electronic components.

[0098] The following will describe the construction method of the policy classification model provided by the embodiment of the present application in conjunction with the exemplary application and implementation of the policy management server provided by the embodiment of the present application. Figure 3A , Figure 3A This is a first flow chart of the method for constructing a policy classification model provided in the embodiment of the present application, which will be combined with Figure 3A The steps shown are explained.

[0099] In step 101, a plurality of sample strategies are obtained, and a strategy tag of each sample strategy is obtained, wherein the strategy tag is used to characterize whether the sample strategy is a high-quality strategy or a low-quality strategy, and the sample strategy is used to perform risk control on the business.

[0100] In some embodiments, multiple sample policies may be policies stored in the policy management server described above, and sample policies for different scenarios may be selected based on different business needs, such as interception policies for violations (e.g., false transactions) in electronic payment scenarios, credit overdue policies in credit scenarios, etc. Sample policies are obtained through processes such as risk assessment, data analysis, and rule-making, and can be adjusted and continuously optimized and updated according to different business scenarios. For example, in electronic payment scenarios, policies can be adjusted according to different electronic payment platforms.

[0101] In some embodiments, the sample strategy is marked with a high-quality (for example, marked as 1) or low-quality (for example, marked as 0) sample mark. Here, the sample mark can be obtained manually, or the strategy can be marked with a high-quality or low-quality sample mark by combining experience with data rules. The embodiment of the present application does not limit the method of obtaining the sample mark of the sample strategy. Here, the judgment of the quality of the sample strategy is not an individual's subjective judgment, but an objective and comprehensive judgment based on the risk control effect generated by the strategy in the business. For example, after the implementation of a certain sample strategy, the risk margin or risk control intervention value in the business objective is compared with the historical data (for example, the preset benchmark data). The difference in the value before and after the change accounts for a percentage of the historical data that exceeds the preset quality threshold. In this case, this sample strategy is a high-quality strategy.

[0102] In step 102, a plurality of test indicators of each sample strategy are obtained, wherein the test indicators are used to quantify the test sample strategy.

[0103] In some embodiments, see Figure 3B , Figure 3A The illustrated step 102 can be implemented by following steps 1021 to 1022, which are described in detail below.

[0104] In step 1021, a strategy analysis model is constructed, wherein the strategy analysis model includes modules constructed step by step in the following order: business objectives of the business, sample strategies for achieving the business objectives, processing links of the sample strategies for achieving the business objectives, and test indicators of multiple links in the processing links.

[0105] In some embodiments, see Figure 3C , Figure 3B Step 1021 shown can be implemented by sending a policy analysis interface of a policy classification model to the terminal device so that the terminal device executes the following steps 10211 to 10215, which are described in detail below.

[0106] In step 10211, the policy analysis interface of the policy analysis model is displayed, and the analysis template of the policy analysis model is loaded into the policy analysis interface.

[0107] In some embodiments, see Figure 4A , Figure 4A This is a schematic diagram of the first interface of the policy analysis model provided in an embodiment of the present application. The analysis template of the policy analysis model may include business goal determination, policy selection, business link generation and test indicator generation. Here, the policy analysis interface is only used as an example. The embodiment of the present application does not limit the specific controls and corresponding functions of the policy analysis interface of the policy analysis model, and can be adjusted and updated according to different development platforms (for example, batch policy analysis, etc.).

[0108] In step 10212, in response to the goal setting operation in the analysis template, the business goals set for the business are displayed in the business goal section of the analysis template.

[0109] In some embodiments, business objectives include: risk margin and risk control intervention amount. The risk margin is used to quantitatively characterize the effect of the sample strategy after implementation. For example, the risk margin in the electronic payment scenario can be the total amount of illegal transactions (such as unapproved or unauthorized electronic transactions, etc.), the total amount of false transactions (such as electronic transactions conducted using false information, etc.), etc.; the risk margin in the credit scenario can be the total amount of overdue credit, etc. The risk control intervention amount is used to characterize the degree to which the sample strategy intervenes in the business during implementation. For example, the risk control intervention amount in the electronic payment scenario can be the number / amount of illegal transaction strategy interventions, the number / amount of false transaction strategy interventions, etc.; the risk control intervention amount in the credit scenario can be the number / amount of credit overdue strategy interventions, etc., wherein the values of the risk margin and the risk control intervention amount are negatively correlated with the quality of the sample strategy.

[0110] In some embodiments, the response is to Figure 4A The triggering operation of the middle control 401 can display a drop-down menu of business goal 1 (multiple preset business goals are displayed in the drop-down menu). Select (single selection) the current business goal 1 in the drop-down menu. Here, business goal 1 can be regarded as the risk margin of the business after the strategy is executed, such as the total amount of false transactions, etc. Similarly, select business goal 2. Here, business goal 2 can be regarded as the risk control intervention amount of the business during the execution of the strategy, such as the number / amount of false transaction strategy intervention, the number / amount of illegal transaction strategy intervention, etc. Next, in response to the Figure 4A The trigger operation of the middle control 403 is to fill in the specific value of target 1, and similarly fill in the specific value of target 2, thereby completing the target setting operation.

[0111] In step 10213, in response to the policy setting operation in the analysis template, the set sample policy for achieving the business goal is displayed in the policy section of the analysis template.

[0112] In some embodiments, the categories of sample policies may include irrevocable policies or revocable policies. Revocable policies are policies whose implemented risk control processing cannot be revoked, and revocable policies are policies whose implemented risk control processing can be revoked.

[0113] In some embodiments, in response to Figure 4A The trigger operation of the control 405 in the middle displays multiple policies corresponding to the business goal in the drop-down menu (for example, the corresponding policies are obtained through the policy management server mentioned above). Then, the policy to be analyzed is selected (single selection) from the drop-down menu (corresponding to the sample policy used to achieve the business goal, such as Figure 4B The illegal behavior interception strategy 001 shown in the figure, where the illegal behavior interception strategy 001 can be an anti-false trading strategy), is used to perform the following strategy analysis.

[0114] In step 10214, in response to the link analysis operation in the analysis template, the processing links generated for the sample policy for achieving the business goal are displayed in the link section of the analysis template.

[0115] In some embodiments, the processing link includes an in-process link and post-process feedback, and the in-process link and post-process feedback respectively include multiple links for processing the business. Among them, the in-process link is used to represent the implementation link (also called the execution link) in the business process. At this stage, business activities are in progress, and various operations and decisions occur immediately. The purpose is to ensure that the business can proceed smoothly as expected. Post-process feedback refers to the evaluation and summary stage after the business process is completed. At this stage, the business process has ended, and the entire business process can be reviewed and analyzed to obtain feedback information about the business execution process.

[0116] In some embodiments, in response to Figure 4A The trigger operation of the control 406 in the analysis template displays the processing link generated for the sample strategy to achieve the business goal (corresponding to Figure 4A (Business Impact Link Diagram in

[15] ).

[0117] In step 10215, in response to the indicator generation operation in the analysis template, the test indicators generated for the processing link are displayed in the indicator section of the analysis template.

[0118] In some embodiments, each responsive link is listed in the order of processing links, a funnel model is built according to each stage (corresponding to the in-process links and post-process feedback mentioned above), the conversion rate and conversion volume of each link in the funnel model are calculated, and multiple test indicators of the sample strategy are obtained.

[0119] In some embodiments, in response to Figure 4AThe triggering operation of the middle control 407 displays the test indicators generated for the processing link and the indicator values of the test indicators in the indicator section of the analysis template.

[0120] For example, the test indicators for the non-releasable strategy in the electronic payment scenario may include at least one of the following:

[0121] Policy interception volume refers to the number of electronic payment requests intercepted by the electronic payment policy within a set window of time (e.g., one month);

[0122] The number of users who filed complaints is the total number of users who filed complaints due to incorrect blocking of electronic payment requests by the electronic payment policy within a set window.

[0123] Complaint rate: the ratio of the number of users complaining to the number of users blocked by the policy within a set window;

[0124] The number of users released refers to the number of users who applied for the release of their blockade within a set window.

[0125] Release rate: the ratio of the number of released users to the number of complaining users within a set window of time;

[0126] The test indicators of the detachable strategy may include at least one of the following:

[0127] Policy intervention refers to the number of times the electronic payment policy intervenes in the user's electronic payment within a set window time. For example, risk warnings are issued during the user's electronic payment process, and the user confirms that the risk is known before making subsequent payment operations.

[0128] Continued payment volume refers to the number of users who choose to continue payment after the electronic payment policy intervenes in the user's electronic payment within a set window time. For example, the number of users who choose to continue payment after being prompted with a risk warning during the electronic payment process;

[0129] Continuation payment rate, the ratio of the amount of continued payment to the amount of strategic intervention within a set window of time;

[0130] Transaction failure reports: This refers to the number of users who, within a set window, reported problems with their transactions after the payment was completed, despite the electronic payment policy intervening in their electronic payment. For example, a risk warning appears during the electronic payment process, and the user confirms the risk and chooses to continue payment, but reports problems with the transaction after payment is complete (e.g., a user purchases goods through the platform and reports to the platform's customer service that the seller did not ship the goods after the user paid, or the seller's store is closed, making it impossible to contact the seller).

[0131] The transaction failure reporting rate is the ratio of the continued payment amount to the transaction failure reporting amount within a set window time.

[0132] For another example, in a credit scenario, the test indicators for non-releasable policies may include at least one of the following: policy interception volume, number of users complaining, complaint rate, number of users being released, and release rate; the test indicators for releasable policies may include at least one of the following:

[0133] Strategy intervention refers to the number of times a credit policy intervenes in a user's loan business within a set window. For example, risk warnings may be issued during the loan application process (e.g., prompting users to repay on time and to pay fees if they exceed the repayment period). Users can then proceed with subsequent loan applications only after confirming that the risks are known.

[0134] Continued loan applications: This refers to the number of users who choose to continue applying for loans after the credit policy intervenes in their loan business within a set window. For example, this refers to the number of users who confirm the risk and continue applying for loans after receiving a risk warning during the loan application process.

[0135] Continued loan application rate, the ratio of continued payment amount to strategic intervention amount within a set window time;

[0136] Loan application failure reports: The number of users who, within a set window, chose to continue applying for a loan but abandoned the application midway, or who successfully applied but failed to repay the loan on time after the credit policy intervened in the user's loan business;

[0137] The loan application failure rate is the ratio of the number of loan application failure reports to the number of continued loan applications within a set window period.

[0138] In step 10216, the service objectives, sample strategies, processing links and test indicators sent by the terminal device are received.

[0139] In some embodiments, the policy management server receives the service objectives, sample policies, processing links, and test indicators sent by the terminal device.

[0140] Continue to see Figure 3B , in step 1022, multiple test indicators of each sample strategy are obtained from the strategy analysis model.

[0141] Continuing with the above example, we calculate the conversion rate and conversion volume of each link to obtain multiple test indicators for the sample strategy.

[0142] Continue to see Figure 3A In step 103, the indicator weight of each test indicator corresponding to the sample strategy is obtained, wherein the indicator weight is used to quantitatively represent the importance of the test indicator when testing the sample strategy.

[0143] In some embodiments, see Figure 3D , Figure 3A The step 103 shown can be implemented by following steps 1031 to 1033, which are described in detail below.

[0144] In step 1031, different indicator data corresponding to multiple sample strategies of the test indicator are obtained.

[0145] In some embodiments, each sample strategy corresponds to multiple test indicators, and thus corresponds to multiple indicator data, for example Figure 4A The indicators 1 to 7 shown in FIG respectively correspond to the indicator values (ie, indicator data).

[0146] Taking the test indicator of complaint volume as an example, multiple sample strategies include anti-illegal transaction strategies and anti-false transaction strategies applied in electronic payment scenarios. The complaint volumes correspond to different indicator data of anti-illegal transaction strategies and anti-false transaction strategies, namely the anti-illegal transaction strategy complaint volume and the anti-false transaction strategy complaint volume. The anti-illegal transaction strategy complaint volume refers to the number of complaints initiated by users among the electronic payment requests intercepted by the anti-illegal transaction strategy within a set window time (for example, one month); the anti-false transaction strategy complaint volume refers to the number of complaints initiated by users among the electronic payment requests intercepted by the anti-false transaction strategy within a set window time (for example, one month).

[0147] In step 1032, different indicator data corresponding to multiple sample strategies of the test indicator are divided into multiple intervals.

[0148] In some embodiments, see Figure 3E , Figure 3D Step 1032 shown can be implemented by following steps 10321 to 10324, which are described in detail below.

[0149] In step 10321, the indicator data corresponding to the test indicators of multiple sample strategies are arranged in sequence to obtain the indicator data to be divided.

[0150] Taking the number of complaints as an example, we obtain the numerical values of the number of complaining users for different sample strategies and arrange them in order from small to large to obtain the band-divided indicator data for the number of complaining users. Similarly, we perform the above processing on other test indicators.

[0151] In step 10322, a separation point is inserted between every two adjacent indicator data in the indicator data to be divided.

[0152] Continuing with the above example, for example, the indicator data to be divided for the number of users is represented as: [indicator data 1, indicator data 2, indicator data 3, indicator data 4]. Here, the data in a pair of square brackets corresponds to an indicator data to be divided. At this time, a separation point or separation mark is inserted between every two adjacent indicator data in the divided data. The divided indicator data after inserting the separation point can be expressed as: [1, (separation point 1), 2, (separation point 2), 3, (separation point 3), 4].

[0153] In step 10323, the information entropy corresponding to each separation point is obtained, and the separation point corresponding to the minimum information entropy is used as the division point.

[0154] Continuing with the above example, obtaining the information entropy corresponding to separation point 1 can be expressed as follows: splitting at separation point 1 to obtain two subsets (i.e., [1] and [2,3,4] subsets), calculating the information entropy of these two subsets respectively, and finally taking the weighted average information entropy (for example, setting the weights of the two subsets to 0.5 and 0.5 respectively) as the information entropy corresponding to separation point 1.

[0155] In some embodiments, the calculation of the information entropy of each subset can be achieved by:

[0156] First, calculate the frequency or probability of each indicator data in the subset. This can be obtained by counting the number of times each indicator data appears in the subset and dividing it by the total size of the subset.

[0157] Next, use these frequencies or probabilities to calculate the information content of each indicator data. The information content can be calculated using the formula in information theory: information content = -p*log2(p), where p is the probability of the indicator data occurring.

[0158] Then, multiply the information content of each indicator data by its corresponding probability and add up the information content of all indicator data to obtain the information entropy corresponding to the subset. Information entropy can be expressed as follows: Information entropy = Σ(-p*log2(p)), where Σ represents the sum of all data.

[0159] In step 10324, the indicator data to be divided is divided into two intervals based on the division point, and the two intervals are used as new indicator data to be divided, and the process of determining new division points and new intervals is iteratively performed until a preset number of intervals are obtained.

[0160] As an example, when performing interval partitioning processing through a decision tree, the preset number of intervals is equal to the preset maximum depth of the decision tree. As an example, the indicator data can be divided into multiple intervals by methods such as equal-distance binning and equal-frequency binning. The embodiments of this application do not limit the specific test indicator interval partitioning method.

[0161] Continue to see Figure 3D In step 1033, the weight values of the test indicators corresponding to the multiple intervals are obtained as the indicator weights of the test indicators corresponding to the sample strategy.

[0162] In some embodiments, see Figure 3F , Figure 3D Step 1033 shown can be implemented by following steps 10331 to 10333, which are described in detail below.

[0163] In step 10331, based on the strategy tag of the sample strategy, the number of indicator data of the test indicator within the interval that hits the inferior strategy is determined as the number of inferior samples, and the number of indicator data of the test indicator within the interval that hits the superior strategy is determined as the number of superior samples.

[0164] In some embodiments, see Figure 9A , obtain the weight values of multiple intervals corresponding to the complaint volume (corresponding to the test indicator), and take the first interval [c0, c1) as an example. The number of sample strategies corresponding to the statistical indicator data is a high-quality strategy or a low-quality strategy. For example, [c0, c1) has a total of 10 indicator data, among which the number of sample strategies corresponding to the indicator data is a high-quality strategy is 9 (corresponding to the number of high-quality samples), and the number of low-quality strategies is 1 (the number of low-quality samples).

[0165] In step 10332, the number of poor-quality strategies among the multiple sample strategies is obtained as the total number of poor-quality samples, and the number of high-quality strategies among the multiple sample strategies is obtained as the total number of high-quality samples.

[0166] In some embodiments, for example, there are 20 sample strategies, the number of strategies that are low-quality strategies is 5 (corresponding to the total number of low-quality samples), and the number of strategies that are high-quality strategies is 15 (corresponding to the total number of high-quality samples).

[0167] In step 10333, the ratio of the number of poor quality samples to the total number of poor quality samples is determined as the proportion of poor quality samples, the ratio of the number of high quality samples to the total number of high quality samples is determined as the proportion of high quality samples, and the weight value of the interval is determined based on the ratio of the proportion of poor quality samples to the proportion of high quality samples.

[0168] Continuing with the above example, the ratio of the number of poor-quality samples to the total number of poor-quality samples is taken as the proportion of poor-quality samples (1 / 5), and the ratio of the number of high-quality samples to the total number of high-quality samples is taken as the proportion of high-quality samples (9 / 15). The weight value of the interval is determined based on the ratio of the proportion of poor-quality samples to the proportion of high-quality samples. For example, the ratio of the proportion of poor-quality samples to the proportion of high-quality samples is logarithmically transformed, and the logarithmic transformation result is used as the weight value.

[0169] Continue to see Figure 3A,In step 104, a strategy indicator set is constructed according to the multiple indicator weights corresponding to each sample strategy.

[0170] In some embodiments, see Figure 3G , Figure 3A Step 104 shown can be implemented by following steps 1041 to 1042, which are described in detail below.

[0171] In step 1041, the indicator weights corresponding to each test indicator of each sample strategy are arranged into rows.

[0172] In some embodiments, the indicator weights corresponding to each test indicator of each sample strategy are arranged in rows. For example, the five test indicators of the violation strategy, namely, the policy interception amount, the number of complaining users, the complaint rate, the number of released users and the release rate, correspond to indicator weights of 0.8, 0.7, 0.6, 0.5 and 0.4 respectively. These five weight values are arranged in rows.

[0173] In step 1042, the rows corresponding to the plurality of sample strategies are organized into a matrix, and the matrix is used as a strategy indicator set.

[0174] In some embodiments, the rows corresponding to each sample strategy are organized into a matrix, and the matrix is used as a strategy indicator set. Here, when the number of test indicators of different sample strategies is different, the end of the row of the corresponding sample strategy can be represented by an empty value or treated as zero, so that the strategy indicator set is represented as an N×K matrix, where N represents the number of sample strategies and K represents the number of test indicators.

[0175] Continue to see Figure 3A In step 105, a strategy classification model is trained using a strategy indicator set, a strategy tag, and a preset loss function, wherein the strategy classification model is used to predict the probability that the strategy to be tested is a high-quality strategy.

[0176] In some embodiments, see Figure 3H , Figure 3A The step 105 shown can be implemented by following the steps 1051 to 1054, which are described in detail below.

[0177] In step 1051, the policy features of the policy indicator set are obtained through the initialized policy classification model.

[0178] In some embodiments, the policy features of the policy indicator set are obtained through a pre-trained policy classification model (for example, feature scaling is performed through a normalization method, feature dimensionality reduction is performed through principal component analysis (PCA), etc., thereby obtaining policy features).

[0179] In step 1052, the strategy features are mapped using the initialized strategy classification model to obtain the predicted probability that each sample strategy in the strategy indicator set is a high-quality strategy.

[0180] In some embodiments, the strategy features are mapped using a pre-trained strategy classification model to obtain a predicted probability that each sample strategy in the strategy indicator set is a high-quality strategy (for example, the strategy features are mapped to a probability value between 0 and 1 using a sigmoid function).

[0181] In step 1053, a loss value corresponding to a preset loss function is obtained through the predicted probability and the strategy tag.

[0182] In some embodiments, a loss value corresponding to a preset loss function (eg, a cross entropy loss function) is obtained through the prediction probability and the strategy tag.

[0183] In step 1054, the model parameters of the initialized policy classification model are updated using the loss value, and a trained policy classification model is constructed based on the updated model parameters.

[0184] In some embodiments, the gradient of the loss function with respect to the model parameters is obtained through the loss value (for example, through a gradient line descent algorithm); the difference between the model parameters and a preset learning rate is obtained, and the difference is multiplied by the gradient to obtain the updated model parameters, wherein the strategy classification model is used to predict the probability that each sample strategy is a high-quality strategy.

[0185] In some embodiments, see Figure 3I ,exist Figure 3E After step 10324 shown, the following steps 10325 to 10326 may also be performed, as described in detail below.

[0186] In step 10325, the quality rate of the test indicator in each interval is obtained according to the monotonically increasing order of the indicator data of multiple intervals, wherein the quality rate is the proportion of the quality samples in the interval used to represent the quality strategy that hits the indicator data of the interval in all the sample strategies hit by the interval.

[0187] Continuing with the example in step 10331, for example, there are 10 indicator data in [c0, c1), among which the number of sample strategies corresponding to the indicator data is high-quality strategies is 9 (corresponding to the number of high-quality samples), and the number of low-quality strategies is 1 (the number of low-quality samples), then the high-quality rate of the interval [c0, c1) is 9 / 10.

[0188] In some embodiments, see Figure 9A , Figure 9AThis is the first schematic diagram of the binning trend diagram provided by the embodiment of the present application. When the test index interval is too small or too large, it will cause the quality rate to fluctuate too much or the monotonicity is not obvious. At this time, the size of the interval can be adjusted according to the manual adjustment instruction. The goal is to make the interval division of the final test index meet the monotonicity principle, that is, the larger the interval value, the higher or lower the quality rate should be. Special values (such as null values, negative numbers, and fill values) can be divided into intervals separately and do not participate in the monotonicity test. For example Figure 9A The partitioning of the complaint volume indicator shows a monotonic trend that the larger the complaint volume, the lower the strategy quality rate, which is completely consistent with the business logic. Therefore, the interval division of the complaint volume as a single indicator is reasonable. For example, Figure 9B In the example, the number of complaints is divided into 7 intervals, and the quality rate does not conform to monotonicity. Therefore, the interval division for the single indicator of the number of complaints is unreasonable. The number of intervals can be adjusted (for example, when partitioning through the decision tree, the preset maximum depth of the decision tree can be reduced). Here, for each test indicator, the above steps of obtaining the monotonic relationship between the test indicator and the quality rate and judging whether the interval division conforms to the business logic need to be performed.

[0189] In step 10326, in response to the fact that the change in the quality rate of multiple intervals complies with monotonicity, each division point of the test indicator is determined.

[0190] In some embodiments, in response to the fact that the change in the quality rate of multiple intervals conforms to monotonicity, the division point of each test indicator is solidified, for example Figure 9A The partition points in can be expressed as: [c0,c1,c2,c3,c4,c5,c6].

[0191] In some embodiments, manual adjustments are made to the test indicators whose policy quality ratios do not obey monotonicity (eg, the intervals are increased or decreased), thereby solidifying the division points of each test indicator.

[0192] In some embodiments, see Figure 3J ,exist Figure 3A After step 105 shown, the following steps 106 to 107 may be performed, which are described in detail below.

[0193] In step 106, the indicator information value is obtained by corresponding the test indicator to multiple intervals and multiple indicator weights, wherein the indicator information value is used to quantitatively represent the importance of the test indicator in the test sample strategy when the indicator data of the test indicator is distributed in multiple intervals.

[0194] In some embodiments, see Figure 3K , Figure 3J The illustrated step 106 can be implemented by following the steps 1061 to 1062, which are described in detail below.

[0195] In step 1061, the difference between the proportion of low-quality samples and the proportion of high-quality samples in the interval is obtained, and the difference is multiplied by the weight value of the interval to obtain the information value of the interval.

[0196] In some embodiments, the information value can be expressed as Represents the information value of the lth interval, where woe l Indicates the indicator weight of the lth interval, Bad l Indicates the number of poor quality samples in the lth interval, Bad T Indicates the total number of poor quality samples, Good l Indicates the number of high-quality samples in the lth interval, Good T Represents the total number of high-quality samples.

[0197] In step 1062, the information values corresponding to the multiple intervals are added together to obtain the index information value.

[0198] In some embodiments, the indicator information value can be expressed as Where L represents the total number of intervals.

[0199] Continue to see Figure 3J In step 107, the test indicators in the strategy indicator set are screened out according to the indicator information value, wherein the strategy indicator set after the screening process is used to replace the strategy indicator set before the screening process for training the strategy classification model.

[0200] For example, the test indicators with IV ≥ 0.02 can be retained, and the remaining test indicators can be eliminated to obtain a screened set of strategy indicators for training the strategy classification model.

[0201] Through steps 101 to 107, the test indicators are systematically constructed through the strategy analysis model, the indicator data of multiple test indicators of the sample strategy are obtained, and then the corresponding indicator weight of each test indicator is obtained, so as to realize the objective empowerment of each test indicator based on the importance of the test indicator itself, and finally achieve the beneficial effect of comprehensive and objective testing of the strategy through the strategy classification model obtained through training.

[0202] Next, the policy processing method of the policy classification model provided by the embodiment of the present application will be described in conjunction with the exemplary application and implementation of the policy management server provided by the embodiment of the present application. Figure 5A , Figure 5A This is a first flow chart of the policy processing method of the policy classification model provided in the embodiment of the present application, which will be combined with Figure 5A The steps shown are explained.

[0203] In step 201, a risk control strategy to be tested is obtained.

[0204] In some embodiments, a risk control strategy to be tested is obtained. Here, the risk control strategy to be tested can be any one of the multiple sample strategies mentioned above (because although the sample strategy carries a sample mark that represents the quality, it does not have a specific quality score), or it can be a strategy optimized and updated by the developer based on the sample strategy, or a new strategy formulated by the developer, etc.

[0205] In step 202, the predicted probability that the risk control strategy to be tested is a high-quality strategy is obtained through the strategy classification model.

[0206] In some embodiments, the strategy characteristics of the risk control strategy to be tested are obtained through a strategy classification model (for example, feature scaling is performed through a normalization method, feature dimensionality reduction is performed through principal component analysis (PCA) and other processes to obtain strategy characteristics); the strategy characteristics are mapped through the strategy classification model to obtain the predicted probability that the risk control strategy to be tested is a high-quality strategy (for example, the strategy characteristics are mapped to a probability value between 0 and 1 through the sig moid function).

[0207] In step 203, the predicted probability is converted into a score to obtain a strategy score of the risk control strategy to be tested.

[0208] In some embodiments, see Figure 5B , Figure 5A The step 203 shown can be implemented by the following steps 2031 to 2034, which are described in detail below.

[0209] In step 2031, the probability ratio is obtained by predicting the probability.

[0210] In some embodiments, represents the odds ratio, p i represents the predicted probability.

[0211] In step 2032, the translation parameters are obtained using the preset benchmark score, score change value, and benchmark ratio.

[0212] In some embodiments, the translation parameter may be expressed as: Among them, P0 is the preset benchmark score, PDO is the preset score change value, which is used to indicate the score reduction when odds0 doubles, and odds0 represents the preset benchmark probability ratio.

[0213] In step 2033, a scaling parameter is obtained by performing logarithmic transformation on the preset score change value.

[0214] In some embodiments, the scaling parameter may be expressed as:

[0215] In step 2034, a linear transformation is performed based on the translation parameter, the scaling parameter and the probability ratio to obtain a strategy score for the risk control strategy to be tested.

[0216] In some embodiments, the strategy score of the risk control strategy to be tested can be expressed as: Score=A-Blog(odds).

[0217] In some embodiments, see Figure 5C ,exist Figure 5A After step 203 shown, the following steps 204 to 207 may be performed, which are described in detail below.

[0218] In step 204 , the risk control strategies to be tested are graded according to preset strategy scoring thresholds, wherein the strategy scoring thresholds include a first scoring threshold and a second scoring threshold, and the first scoring threshold is higher than the second scoring threshold.

[0219] In some embodiments, the risk control strategies to be tested are graded according to preset strategy scoring thresholds. For example, if the strategy score is full of 10, the first scoring threshold is set to 8 and the second scoring threshold is set to 4.

[0220] In step 205 , in response to the strategy score being lower than the second score threshold, the risk control strategy to be tested is marked as a low-quality strategy and is taken offline.

[0221] In step 206 , in response to the strategy scoring threshold being higher than the second scoring threshold and lower than the first scoring threshold, the risk control strategy to be tested is marked as a strategy to be optimized.

[0222] In some embodiments, for the strategy to be optimized, the relevant developers can further optimize and update it and re-score the strategy.

[0223] In step 207 , in response to the strategy scoring threshold being higher than the first scoring threshold, the risk control strategy to be tested is marked as a high-quality strategy.

[0224] Through steps 201 to 207, it is achieved to assist in discovering redundant or degraded low-quality strategies in the business, thereby achieving the beneficial effect of ensuring high-quality streamlining of the strategy set and avoiding the increase and non-decrease of strategies during business development, thereby avoiding the additional system risks.

[0225] In some embodiments, the policy scoring processing of the policy management server 100 described above is constructed step by step in the following order: building a policy analysis model, building a policy quality score card, and calculating a policy quality score. Below, combined with the above three construction steps and the exemplary application and implementation of the combination of the server and terminal device provided in the embodiment of the present application, the construction method of the policy classification model provided in the embodiment of the present application in the electronic payment scenario and the policy processing method of the policy classification model are explained.

[0226] In some embodiments, for building a strategic analysis model, see Figure 6A , Figure 6A This is a flow chart of the strategy analysis model for building an electronic payment scenario provided by the embodiment of the present application, which will be based on the server (such as Figure 1 The policy management server 100 shown is the execution entity, combined with Figure 6A The steps shown are explained.

[0227] In step 301, the overall goal of policy management for risk control in an electronic payment scenario is obtained.

[0228] For example, the following two types of goals are taken as overall goals: 1) Risk margin: such as the total amount of illegal transactions, the total amount of false transactions, etc.; 2) Transaction risk control intervention amount, such as the number / amount of illegal trading strategy interventions, the number / amount of false trading strategy interventions, etc. Among them, the smaller the risk margin and transaction risk control intervention amount, the better the strategy management effect.

[0229] In step 302, the strategies to be managed are sorted out according to the overall goal and the strategies are classified.

[0230] In some embodiments, after formulating the overall goal of policy management, the policies that need to be managed are sorted out according to the overall goal. Online policies are often generated through data models or rules and deployed in risk control systems (e.g. Figure 1 In the risk control system shown in the figure), daily automated implementation is carried out. There are two types of strategies: non-releasable strategies and releasable strategies. When a non-releasable strategy intercepts a certain link in the business process, the user cannot release it by themselves at this link, and manual intervention can be performed. When the manual review is passed, the user is manually released to continue the next link in the business process.

[0231] In step 303, the business process is sorted out according to the policy type, and the policy impact link of the policy is determined through the business process, wherein the policy impact link includes multiple links.

[0232] In some embodiments, see Figure 8A , Figure 8AThis is a schematic diagram of the business process impact link in the electronic payment scenario provided by an embodiment of the present application. Each policy in the risk control system has a unique identifier (ID). The risk control system makes a real-time judgment on whether to intercept the current link (for example, the payment operation of the current transaction) based on the risk judgment result returned by the policy, and intervenes in the current link according to the interception type specified by the policy (corresponding to the releasable policy type and the non-releasable policy type mentioned above). The business process impact link may include in-process links and post-event feedback. For example, in the electronic payment scenario, the in-process link may include: initiating an order, retrieving a violation interception policy, hitting a policy or missing a policy, intercepting a transaction, reminding the user to self-release, etc. Post-event feedback may include: user complaints of being mistakenly intercepted, transaction protection, and user release, etc. Here, the business process impact link is only used as an example. The embodiment of the present application does not limit the various links of the business process impact link.

[0233] In step 304, the conversion rate and conversion volume of each link are obtained as multiple testing indicators of the strategy.

[0234] In some embodiments, continuing with the above example, the conversion rate and conversion volume of each link are obtained as multiple test indicators of the strategy. For example, the test indicators of the non-releasable strategy in the electronic payment scenario may include: the policy interception volume, the number of users who complain (corresponding to the conversion volume), the complaint rate (corresponding to the conversion rate), the number of users who were released, the release rate (here it means that there will be manual intervention after the appeal. If the appeal is successful, the electronic payment user will be manually released), etc. The test indicators of the releasable strategy may include: the policy intervention volume, the continued payment volume (corresponding to the conversion volume), the continued payment rate (corresponding to the conversion rate), the transaction problem reporting volume, the transaction problem reporting rate, etc.

[0235] In some embodiments, see Figure 4B , Figure 4B This is a schematic diagram of the second interface of the strategy analysis model provided in the embodiment of the present application. Figure 4B It can be seen that the corresponding test indicators of the strategy in the electronic payment scenario can be obtained through the strategy analysis interface of the strategy analysis model. For the specific description of the strategy analysis interface, please refer to the description of step 1021 above.

[0236] In some embodiments, for building a strategy scorecard, see Figure 6B , Figure 6B This is a first flow chart of building a strategy scorecard in an electronic payment scenario provided by an embodiment of the present application, which uses a server (such as Figure 1 The policy management server 100 shown is the execution entity, combined with Figure 6B The steps shown are explained.

[0237] In step 305, a sample strategy indicator set of multiple strategies for risk control in an electronic payment scenario is constructed.

[0238] In some embodiments, the indicator derivation is completed through the strategy analysis model described above, thereby establishing a sample strategy indicator set. The sample strategy indicator set can be expressed by the following formula:

[0239]

[0240] Among them, N represents the number of sample strategies, K represents the number of test indicators, and s ij Indicates the index value of the jth indicator of the i-th strategy.

[0241] In step 306 , a sample tag of each sample strategy in the sample strategy indicator set is determined.

[0242] In some embodiments, the sample strategy is marked with a high-quality (e.g., marked as 1) or a low-quality (e.g., marked as 0) sample label. Here, the labeling can be performed manually, or the strategy can be marked with a good or bad label based on experience combined with data rules. The embodiment of the present application does not limit the method for obtaining the sample label of the sample strategy. The sample label of each sample strategy can be expressed by the following formula:

[0243]

[0244] Where N represents the number of sample strategies, Y i Indicates the quality of the strategy, Y i ∈{0,1}.

[0245] In step 307, each indicator is discretized and binned based on the sample label, and the monotonicity and business applicability of each test indicator are tested.

[0246] In some embodiments, see Figure 6C , Figure 6B The illustrated step 307 can be implemented by executing the following steps 3071 to 3072 for each test indicator, which will be described in detail below.

[0247] In step 3071, the test indicators are discretized and binned, and the quality rate of each bin is calculated.

[0248] In some embodiments, indicator data of the same test indicator from multiple sample strategies are obtained, and the indicator values are arranged in ascending order to obtain the indicator data to be binned (corresponding to the indicator data to be divided above); the midpoint between each group of adjacent data in the indicator data to be binned is used as a separation point to obtain multiple separation points; the information entropy corresponding to each of the separation points is obtained, and the separation point with the smallest information entropy is used as a bin node (corresponding to the division point above); the data to be binned is divided at the bin node to obtain two subsets of the indicator data to be binned, and the bin nodes corresponding to each subset are repeatedly obtained, and the subsets are divided until the bin nodes reach a preset number (for example, the maximum depth of the decision tree preset when binning is performed by a decision tree), and multiple bins are obtained (corresponding to the multiple intervals above), wherein each bin is used to characterize the value interval of the test indicator and the number of sample strategies corresponding to the value interval, and the quality rate under each bin is the ratio of the number of quality strategies of the current bin to the number of sample strategies of the current bin.

[0249] Here, refer to the description of steps 10321 to 10324 above.

[0250] In step 3072, a binning trend chart is drawn to obtain the monotonic relationship between the test index and the quality rate, and combined with the business, it is judged whether the binning process conforms to the business logic.

[0251] In some embodiments, for example Figure 9A The binning of the complaint volume indicator clearly shows a monotonic trend that the larger the complaint volume, the lower the strategy quality rate, which is completely consistent with the business logic. Therefore, the binning processing of the single indicator of complaint volume is reasonable. Here, for each test indicator, it is necessary to perform the above steps to obtain the monotonic relationship between the test indicator and the quality rate, and judge whether the binning processing conforms to the business logic. For the test indicators where the strategy quality rate does not obey the monotonicity, manual adjustments are made to solidify the binning nodes of each test indicator. For example, Figure 9A The binning nodes in can be represented as: [c0,c1,c2,c3,c4,c5,c6].

[0252] Continue to see Figure 6B , in step 308, the bin weight value and indicator information value of each test indicator are obtained.

[0253] In some embodiments, the bin weight value of each test indicator can be expressed by the following formula:

[0254]

[0255] Among them, woe l Indicates the sharing weight (i.e., evidence weight) of the test indicator in the lth bin, Bad lIndicates the number of strategy samples that are inferior strategies in the lth bin, Bad T Indicates the total number of inferior strategies in the sample strategy indicator set, Good l Indicates the number of strategy samples that are good strategies in the lth bin, Good T Represents the total number of high-quality strategies in the sample strategy indicator set.

[0256] In some embodiments, the indicator information value can be expressed by the following formula:

[0257]

[0258] Among them, L represents the total number of bins, Represents the information value of the lth bin.

[0259] In step 309, the sample strategy indicator set is reconstructed by the bin weight value and indicator information value of each test indicator to obtain a strategy weight indicator set.

[0260] In some embodiments, the policy weight indicator set may be represented by the following formula:

[0261]

[0262] in, Is the indicator function, when the test indicator s of the strategy belongs to the bin Bin l 1 if yes, 0 otherwise.

[0263] In step 310 , a strategy classification model is trained using sample labels and a strategy weight indicator set, wherein the strategy classification model is used to predict the probability that each sample strategy is a high-quality strategy.

[0264] In some embodiments, the policy features of the policy indicator set are obtained through a pre-trained policy classification model (for example, feature scaling is performed through a normalization method, feature dimensionality reduction is performed through a principal component analysis method, etc., thereby obtaining the policy features); the policy features are mapped through the pre-trained policy classification model to obtain the predicted probability that each sample strategy in the policy indicator set is a high-quality strategy (for example, the policy features are mapped to a probability value between 0 and 1 through a sigmoid function); the loss value corresponding to a preset loss function (for example, a cross-entropy loss function) is obtained through the predicted probability and the policy label; the model parameters of the pre-trained policy classification model are updated through the loss value to obtain a trained policy classification model, wherein the policy classification model is used to predict the probability that each sample strategy is a high-quality strategy.

[0265] In some embodiments, the gradient of the loss function for each model parameter is obtained through the loss value (for example, the vector value of the gradient is obtained through the gradient descent method); the difference between each model parameter and the preset learning rate is obtained, and the difference is multiplied by the gradient to obtain each updated model parameter.

[0266] In some embodiments, the trained policy classification model can be expressed as follows:

[0267]

[0268] in,

[0269] In other embodiments, any machine learning algorithm (such as decision tree, random forest, K-nearest neighbor algorithm and neural network algorithm, etc.) can be used to fit the quality of the strategy. This application does not limit the specific algorithm for fitting the strategy classification model.

[0270] In some embodiments, see Figure 10 , Figure 10 This is a schematic diagram of the network structure of the strategy classification model provided in the embodiment of the present application. When the strategy classification model is fitted by a neural network, as shown in FIG. Figure 10 As shown in the figure: a1 represents the input layer (for example, one neuron corresponds to a sample strategy in the strategy indicator set), a2 and a3 represent the hidden layers, w1, w2, and w3 represent the weight parameters connecting each layer, b1, b2, and b3 represent the bias parameters of each layer, and z represents the output layer (for example, one neuron corresponds to the predicted probability of a sample strategy). The output derivation of each layer can be expressed by the following formula:

[0271] g(w1*a1+b1)=a1

[0272] g(w2*a2+b2)=a3 (7)

[0273] g(w3*a3+b3)=z

[0274] Here, g(·) represents a nonlinear activation function.

[0275] Continue to see Figure 6B In step 311, the trained strategy classification model is used to obtain the predicted probability that each sample strategy in the strategy weight index set is a high-quality strategy.

[0276] In some embodiments, refer to the above formula (6) to obtain the prediction result of each sample strategy: {p i} i＝1,…N .

[0277] In some embodiments, to calculate the strategy quality score, a quality score corresponding to each sample strategy may be obtained by performing a score conversion process on each predicted probability.

[0278] In some embodiments, the score conversion process can be expressed as follows:

[0279] Score=AB log(odds) (8)

[0280] in, Expressed as ratios, A and B are calculated as follows:

[0281]

[0282] Among them, P0 is the preset benchmark score, PDO is the preset score change value, which is used to indicate the score reduction when odds0 doubles, and odds0 represents the preset benchmark ratio.

[0283] Continuing with the above example, the final quality score of each sample strategy can be expressed as:

[0284]

[0285] In some embodiments, after obtaining the quality score corresponding to each sample strategy, the sample strategies can also be graded according to a preset strategy scoring threshold, thereby effectively improving the efficiency of strategy management and clarifying the management direction. The strategy scoring threshold may, for example, include a first scoring threshold and a second scoring threshold, and the first scoring threshold is higher than the second scoring threshold; in response to the quality score being lower than the second scoring threshold, the corresponding sample strategy is taken offline; in response to the quality score being higher than the second scoring threshold and lower than the first scoring threshold, the corresponding sample strategy is marked as a strategy to be optimized; in response to the quality score being higher than the first scoring threshold, the corresponding sample strategy is marked as a stable strategy.

[0286] Through steps 301 to 311, by systematically building test indicators for risk control strategies in electronic payment scenarios, it is achieved that each test indicator is objectively weighted based on the importance of the test indicator itself, and finally the strategy classification model obtained through training is used to conduct a comprehensive and objective test of the strategy, thereby assisting in discovering redundant or degraded low-quality strategies in electronic payment scenario businesses, ensuring high-quality and streamlined strategy sets, and thus restricting illegal behaviors in electronic payment scenarios (such as false transactions, etc.), thereby achieving the beneficial effect of improving the security of users' electronic payments and user trust.

[0287] Below, in combination with the three construction steps of the above-mentioned strategy scoring processing, as well as the exemplary application and implementation of the combination of the server and terminal device provided in the embodiment of the present application, the construction method of the strategy classification model provided in the embodiment of the present application in the credit scenario and the strategy processing method of the strategy classification model are explained.

[0288] In some embodiments, for building a strategic analysis model, see Figure 7A , Figure 7A This is a flow chart of the construction strategy analysis model in the credit scenario provided by the embodiment of the present application, which will be based on the server (such as Figure 1 The policy management server 100 shown is the execution entity, combined with Figure 7A The steps shown are explained.

[0289] In step 401, the overall goal of the policy management for risk control in the credit scenario is obtained.

[0290] For example, the following two types of goals are taken as overall goals: 1) Risk market margin: such as the total amount of overdue credit, etc.; 2) Transaction risk control intervention amount, such as the number / amount of overdue credit strategy intervention, etc. Among them, the smaller the risk market margin and transaction risk control intervention amount, the better the strategy management effect.

[0291] In step 402, the strategies to be managed are sorted out according to the overall goal and the strategies are classified.

[0292] Here, please refer to the description of step 302 above.

[0293] In step 403, the business process is sorted out according to the policy type, and the policy impact link of the policy is determined through the business process, wherein the policy impact link includes multiple links.

[0294] In some embodiments, see Figure 8B , Figure 8B This is a schematic diagram of the business process impact link in the credit scenario provided by the embodiment of the present application. The in-process link may include, for example: initiating overdue warning, calling credit overdue strategy, hitting strategy or missing strategy, reminder / user self-service repayment and other links. Post-event feedback may include, for example: user complaint of being mistakenly intercepted, credit management, recovery processing and user relief and other links. Here, the business process impact link is only used as an example, and the embodiment of the present application does not limit the various links of the business process impact link.

[0295] In step 404, the conversion rate and conversion volume of each link are obtained as multiple testing indicators of the strategy.

[0296] In some embodiments, continuing with the above example, the conversion rate and conversion volume of each link are obtained as multiple test indicators of the strategy. For example, the test indicators of the non-releasable strategy in the credit scenario may include: strategy intervention volume, number of complaining users, user complaint rate, number of released users, release rate, etc. The test indicators of the releasable strategy may include: strategy intervention volume, continued repayment volume, continued repayment rate, etc.

[0297] In some embodiments, see Figure 4C , Figure 4C This is a schematic diagram of the third interface of the strategy analysis model provided in an embodiment of the present application. As can be seen from the figure, the corresponding test indicators of the strategy under the credit scenario can be obtained through the strategy analysis interface of the strategy analysis model. For the specific description of the strategy analysis interface, please refer to the description of step 1021 above.

[0298] In some embodiments, for building a strategy scorecard, see Figure 7B , Figure 7B This is a flow chart of the construction of a strategy scorecard in a credit scenario provided by an embodiment of the present application, which will be based on a server (such as Figure 1 The policy management server 100 shown is the execution entity, combined with Figure 7B The steps shown are explained.

[0299] In step 405, a sample strategy indicator set of multiple strategies for risk control in a credit scenario is constructed.

[0300] Here, refer to the description of formula (1) in step 305 above.

[0301] In step 406 , a sample tag of each sample strategy in the sample strategy indicator set is determined.

[0302] Here, refer to the description of formula (2) in step 306 above.

[0303] In step 407, each indicator is discretized and binned based on the sample label, and the monotonicity and business applicability of each test indicator are tested.

[0304] Here, refer to the description of step 307 above.

[0305] In step 408, the bin weight value and indicator information value of each test indicator are obtained.

[0306] Here, refer to the description of formula (3-4) in step 308 above.

[0307] In step 409, the sample strategy indicator set is reconstructed by the bin weight value and indicator information value of each test indicator to obtain a strategy weight indicator set.

[0308] Here, refer to the description of formula (5) in step 309 above.

[0309] In step 410 , a strategy classification model is trained using sample labels and a strategy weight indicator set, wherein the strategy classification model is used to predict the probability that each sample strategy is a high-quality strategy.

[0310] Here, refer to the description in step 310 above.

[0311] In step 411, the predicted probability of each sample strategy being a high-quality strategy in the strategy weight index set is obtained through the trained strategy classification model.

[0312] Here, refer to the description in step 311 above.

[0313] Through steps 401 to 411, by systematically building test indicators for risk control strategies in credit scenarios, it is achieved that each test indicator is objectively weighted based on the importance of the test indicator itself, and ultimately the strategy classification model obtained through training is used to conduct comprehensive and objective testing of the strategy, thereby assisting in discovering redundant or degraded low-quality strategies in the business under the credit scenario, ensuring high-quality and streamlined strategy sets, thereby optimizing capital utilization (for example, credit management of users through credit overdue strategies), improving customer satisfaction, and promoting business growth.

[0314] The following continues to describe the exemplary structure of the strategy classification model construction device 133 provided in the embodiment of the present application as a software module. In some embodiments, such as Figure 2A As shown, the software modules stored in the strategy classification model construction device 133 of the memory 130-1 may include:

[0315] The acquisition module 1331 is used to acquire multiple sample strategies and acquire a strategy tag of each of the sample strategies, wherein the strategy tag is used to indicate whether the sample strategy is a high-quality strategy or a low-quality strategy, and the sample strategy is used to perform risk control on the business.

[0316] The construction module 1332 is configured to obtain a plurality of test indicators for each of the sample strategies, wherein the test indicators are used to quantitatively test the sample strategies.

[0317] The training module 1333 is used to train a strategy classification model using the strategy indicator set, the strategy label and a preset loss function, wherein the strategy classification model is used to predict whether the strategy to be tested is a high-quality strategy or a low-quality strategy.

[0318] In some embodiments, the construction module 1332 is further used to obtain an indicator weight of each of the test indicators corresponding to the sample strategy, wherein the indicator weight is used to quantitatively represent the importance of the test indicator when testing the sample strategy.

[0319] In some embodiments, the construction module 1332 is further configured to construct a strategy indicator set according to the multiple indicator weights corresponding to each of the sample strategies.

[0320] In some embodiments, the construction module 1332 is also used to construct a strategy analysis model, wherein the strategy analysis model includes modules constructed step by step in the following order: business goals that can be achieved by implementing the sample strategy for the business; the sample strategy for achieving the business goals; the processing link of the sample strategy for achieving the business goals; test indicators of multiple links in the processing link; and obtaining multiple test indicators for each sample strategy from the strategy analysis model.

[0321] In some embodiments, the construction module 1332 is also used to send the policy analysis interface of the policy classification model to the terminal device, so that the terminal device performs the following processing: displaying the policy analysis interface of the policy analysis model, and loading the analysis template of the policy analysis model in the policy analysis interface; in response to the goal setting operation in the analysis template, displaying the business goal set for the business in the business goal section of the analysis template; in response to the policy setting operation in the analysis template, displaying the sample policy set for achieving the business goal in the policy section of the analysis template; in response to the link analysis operation in the analysis template, displaying the processing link generated for achieving the business goal for the sample policy in the link section of the analysis template; in response to the indicator generation operation in the analysis template, displaying the test indicator generated for the processing link in the indicator section of the analysis template; receiving the business goal, the sample policy, the processing link and the test indicator sent by the terminal device.

[0322] In some embodiments, the construction module 1332 is further used to perform the following processing for each of the test indicators: obtaining different indicator data corresponding to the multiple sample strategies of the test indicator; dividing the different indicator data corresponding to the multiple sample strategies of the test indicator into multiple intervals; obtaining weight values of the test indicators corresponding to the multiple intervals respectively, as the indicator weights of the test indicators corresponding to the sample strategies. In some embodiments, the construction module 1332 is further used to sequentially arrange the indicator data corresponding to the multiple sample strategies of the test indicator respectively to obtain the indicator data to be divided; insert a separation point between every two adjacent indicator data in the indicator data to be divided; obtain the information entropy corresponding to each separation point, and use the separation point corresponding to the minimum information entropy as the separation point; divide the indicator data to be divided into two intervals based on the separation point, and use the two intervals as the new indicator data to be divided, so as to iteratively perform the process of determining the new separation point and the new interval until a preset number of intervals are obtained.

[0323] In some embodiments, the construction module 1332 is also used to obtain the quality rate of the test indicator in each of the intervals in a monotonically increasing order of the indicator data of the multiple intervals, wherein the quality rate is the proportion of the high-quality samples in the interval used to represent the high-quality strategy hit by the indicator data of the interval in all the sample strategies hit in the interval; in response to the fact that the changes in the quality rates of the multiple intervals conform to monotonicity, each of the division points of the test indicator is determined.

[0324] In some embodiments, the construction module 1332 is also used to perform the following processing for each of the intervals: according to the strategy tag of the sample strategy, determine the number of the indicator data of the test indicator in the interval that hits the inferior strategy as the number of inferior samples, determine the number of the indicator data of the test indicator in the interval that hits the high-quality strategy as the number of high-quality samples; obtain the number of the inferior strategies in the multiple sample strategies as the total number of inferior samples, obtain the number of the sample strategies that are the high-quality strategies in the multiple sample strategies as the total number of high-quality samples; determine the ratio of the number of inferior samples to the total number of inferior samples as the proportion of inferior samples, determine the ratio of the number of high-quality samples to the total number of high-quality samples as the proportion of high-quality samples, and determine the weight value of the interval based on the ratio of the proportion of inferior samples to the proportion of high-quality samples.

[0325] In some embodiments, the construction module 1332 is further used to arrange the indicator weights corresponding to each test indicator of each sample strategy into rows; form a matrix with the rows corresponding to the multiple sample strategies respectively, and use the matrix as a strategy indicator set.

[0326] In some embodiments, the construction module 1332 is also used to perform the following processing on each of the test indicators: obtain an indicator information value by respectively corresponding the test indicators to the multiple indicator weights of the multiple intervals, wherein the indicator information value is used to quantitatively represent the importance of the test indicator in testing the sample strategy when the indicator data of the test indicator is distributed in the multiple intervals; screen out the test indicators in the strategy indicator set according to the indicator information value, wherein the strategy indicator set after the screening process is used to replace the strategy indicator set before the screening process, so as to be used for training the strategy classification model.

[0327] In some embodiments, the construction module 1332 is also used to perform the following processing on each of the multiple intervals: obtain the difference between the proportion of poor quality samples and the proportion of high quality samples in the interval, and multiply the difference by the weight value of the interval to obtain the information value of the interval; add the information values corresponding to the multiple intervals to obtain the indicator information value.

[0328] In some embodiments, the training module 1333 is also used to obtain the policy features of the policy indicator set through the initialized policy classification model; map the policy features through the initialized policy classification model to obtain the predicted probability that each sample strategy in the policy indicator set is a high-quality strategy; obtain the loss value corresponding to the preset loss function through the predicted probability and the strategy label; update the model parameters of the initialized policy classification model through the loss value, and construct the trained policy classification model based on the updated model parameters.

[0329] The following continues to describe the exemplary structure of the policy processing device 134 of the policy classification model provided in the embodiment of the present application as a software module. In some embodiments, such as Figure 2B As shown, the software modules stored in the policy processing device 134 of the policy classification model in the memory 130-2 may include:

[0330] An acquisition module 1341 is used to acquire the risk control strategy to be tested;

[0331] Prediction module 1342, used to obtain the predicted probability that the risk control strategy to be tested is a high-quality strategy through the strategy classification model;

[0332] The scoring module 1343 is configured to perform score conversion processing on the predicted probability to obtain a strategy score for the risk control strategy to be tested.

[0333] In some embodiments, the scoring module 1343 is also used to obtain the probability ratio through the predicted probability; obtain the translation parameter through the preset benchmark score, score change value and benchmark ratio; obtain the scaling parameter by performing a logarithmic transformation on the preset score change value; based on the translation parameter, the scaling parameter and the probability ratio are linearly transformed to obtain the strategy score of the risk control strategy to be tested.

[0334] In some embodiments, the scoring module 1343 is also used to grade the risk control strategy to be tested according to a preset strategy scoring threshold, wherein the strategy scoring threshold includes a first scoring threshold and a second scoring threshold, and the first scoring threshold is higher than the second scoring threshold; in response to the strategy score being lower than the second scoring threshold, the risk control strategy to be tested is marked as a low-quality strategy and is taken offline; in response to the strategy scoring threshold being higher than the second scoring threshold and lower than the first scoring threshold, the risk control strategy to be tested is marked as a strategy to be optimized; in response to the strategy scoring threshold being higher than the first scoring threshold, the risk control strategy to be tested is marked as a high-quality strategy.

[0335] An embodiment of the present application provides a computer program product, which includes a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer-executable instructions from the computer-readable storage medium and executes the computer-executable instructions, causing the electronic device to perform the above-mentioned method for constructing a policy classification model or the policy processing method for a policy classification model in the embodiment of the present application.

[0336] The embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions or computer programs are stored. When the computer-executable instructions or computer programs are executed by a processor, the processor will be caused to execute the method for constructing a policy classification model or the policy processing method of the policy classification model provided in the embodiment of the present application, for example, Figure 3A The construction method of the strategy classification model shown or Figure 5A The policy processing method of the policy classification model is shown.

[0337] In some embodiments, the computer-readable storage medium may be a memory such as RAM, ROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

[0338] In some embodiments, computer-executable instructions may be in the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0339] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0340] By way of example, computer-executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed across multiple sites and interconnected by a communication network.

[0341] To sum up, through the embodiments of the present application, multiple test indicators of sample strategies are systematically obtained in combination with business cognition, thereby constructing a sample strategy indicator set, and obtaining the indicator weight corresponding to each test indicator, thereby achieving objective weighting of each test indicator based on the importance of the test indicator itself, and finally achieving the beneficial effect of comprehensive and objective testing of the strategy through the trained strategy classification model, thereby assisting in discovering redundant or degraded low-quality strategies in the business, ensuring high-quality and streamlined strategy sets, and avoiding additional system risks caused by the increase in strategies during business development.

[0342] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A method for constructing a strategy classification model, characterized in that: The method comprises: Acquire multiple sample strategies and acquire a strategy tag for each of the sample strategies, wherein the strategy tag is used to characterize whether the sample strategy is a high-quality strategy or a low-quality strategy, and the sample strategy is used to perform risk control on the business; Acquire a plurality of test indicators for each of the sample strategies, wherein the test indicators are used to quantitatively test the sample strategies; Obtaining an indicator weight of each of the test indicators corresponding to the sample strategy, wherein the indicator weight is used to quantitatively represent the importance of the test indicator when testing the sample strategy; Constructing a strategy indicator set according to the plurality of indicator weights corresponding to each of the sample strategies; A strategy classification model is trained using the strategy indicator set, the strategy label, and a preset loss function, wherein the strategy classification model is used to predict the probability that the strategy to be tested is a high-quality strategy.

2. The method according to claim 1, characterized in that The obtaining of multiple test indicators for each of the sample strategies includes: Constructing a strategy analysis model, wherein the strategy analysis model includes modules constructed step by step in the following order: a business goal that can be achieved by implementing the sample strategy for the business; the sample strategy for achieving the business goal; a processing chain of the sample strategy for achieving the business goal; and test indicators for multiple links in the processing chain; A plurality of the test indicators of each of the sample strategies are obtained from the strategy analysis model.

3. The method according to claim 2, characterized in that The constructing of the strategy analysis model includes: Sending the policy analysis interface of the policy classification model to the terminal device so that the terminal device performs the following processing: Displaying a strategy analysis interface of the strategy analysis model, and loading an analysis template of the strategy analysis model into the strategy analysis interface; In response to a goal setting operation in the analysis template, displaying the business goal set for the business in a business goal section of the analysis template; In response to a policy setting operation in the analysis template, displaying the sample policy set for achieving the business goal in a policy section of the analysis template; In response to the link analysis operation in the analysis template, displaying the processing link generated for the sample strategy and used to achieve the business goal in the link section of the analysis template; In response to the indicator generation operation in the analysis template, displaying the test indicator generated for the processing link in the indicator section of the analysis template; Receive the service goal, the sample strategy, the processing link and the test indicator sent by the terminal device.

4. The method according to claim 2, characterized in that The business objectives include: a risk margin and a risk control intervention amount. The risk margin is used to quantitatively represent the effect of the sample strategy after implementation, and the risk control intervention amount is used to represent the degree of intervention of the sample strategy on the business during implementation. The values of the risk margin and the risk control intervention amount are negatively correlated with the quality of the sample strategy. The sample policies include irreleasable policies and releasable policies. The irreleasable policies are policies whose implemented risk control processing cannot be revoked, and the releasable policies are policies whose implemented risk control processing can be revoked. The processing link includes an in-process link and a post-process feedback link, wherein the in-process link and the post-process feedback link respectively include multiple links for processing the business; The test indicators include the conversion rate and conversion volume of each link.

5. The method according to claim 1, wherein The obtaining of the indicator weight of each test indicator corresponding to the sample strategy includes: The following processing is performed for each of the test indicators: Obtain different indicator data of the test indicators corresponding to the multiple sample strategies; Dividing the different indicator data of the test indicator corresponding to the multiple sample strategies into multiple intervals; Obtain weight values of the test indicators corresponding to the multiple intervals respectively, as indicator weights of the test indicators corresponding to the sample strategy.

6. The method according to claim 5, characterized in that Dividing the different indicator data of the test indicators corresponding to the multiple sample strategies into multiple intervals includes: Arrange the indicator data of the test indicators corresponding to the multiple sample strategies in order to obtain the indicator data to be divided; Inserting a separation point between every two adjacent index data in the index data to be divided; Obtaining the information entropy corresponding to each of the separation points, and taking the separation point corresponding to the minimum information entropy as the division point; The indicator data to be divided is divided into two intervals based on the division point, and the two intervals are used as new indicator data to be divided, so as to iteratively perform the process of determining new division points and new intervals until a preset number of intervals are obtained.

7. The method according to claim 6, characterized in that After obtaining the preset number of intervals, the method further includes: Obtaining the quality rate of the test indicator in each interval in a monotonically increasing order of the indicator data of the multiple intervals, wherein the quality rate is the proportion of the high-quality samples in the interval used to represent the high-quality strategy hit by the indicator data of the interval in all the sample strategies hit by the interval; In response to the fact that changes in the quality rates of the plurality of intervals comply with monotonicity, each of the division points of the test indicator is determined.

8. The method according to claim 5, characterized in that The obtaining of weight values of the test indicators corresponding to the plurality of intervals includes: The following processing is performed for each of the intervals: According to the strategy tag of the sample strategy, determine the number of the indicator data of the test indicator in the interval that hits the low-quality strategy, as the number of low-quality samples, and determine the number of the indicator data of the test indicator in the interval that hits the high-quality strategy, as the number of high-quality samples; Obtaining the number of the poor-quality strategies among the multiple sample strategies as the total number of poor-quality samples, and obtaining the number of the high-quality strategies among the multiple sample strategies as the total number of high-quality samples; Determine the ratio of the number of poor quality samples to the total number of poor quality samples as the proportion of poor quality samples, determine the ratio of the number of high quality samples to the total number of high quality samples as the proportion of high quality samples, and determine the weight value of the interval based on the ratio of the proportion of poor quality samples to the proportion of high quality samples.

9. The method according to claim 1, characterized in that The constructing of a strategy indicator set according to the plurality of indicator weights corresponding to each sample strategy includes: Arrange the indicator weights corresponding to each test indicator of each sample strategy into rows; The rows corresponding to the plurality of sample strategies are respectively formed into a matrix, and the matrix is used as a strategy indicator set.

10. The method according to any one of claims 1 to 9, characterized in that Before training the strategy classification model using the strategy indicator set, the strategy label, and the preset loss function, the method further includes: The following processing is performed on each of the test indicators: Obtain an indicator information value by corresponding the test indicator to the multiple indicator weights of the multiple intervals, wherein the indicator information value is used to quantitatively represent the importance of the test indicator in testing the sample strategy when the indicator data of the test indicator is distributed in the multiple intervals; The test indicators in the strategy indicator set are screened out according to the indicator information value, wherein the strategy indicator set after the screening process is used to replace the strategy indicator set before the screening process for training the strategy classification model.

11. The method according to claim 10, characterized in that The obtaining of indicator information values by respectively corresponding the test indicators to the multiple indicator weights of the multiple intervals includes: The following processing is performed on each of the plurality of intervals: Obtaining the difference between the proportion of the low-quality samples and the proportion of the high-quality samples in the interval, and multiplying the difference by the weight value of the interval to obtain the information value of the interval; The information values corresponding to the multiple intervals are added together to obtain an indicator information value.

12. The method according to any one of claims 1 to 9, characterized in that The training of the strategy classification model using the strategy indicator set, the strategy label and the preset loss function includes: Obtaining the strategy characteristics of the strategy indicator set through the initialized strategy classification model; Mapping the strategy features using the initialized strategy classification model to obtain a predicted probability that each of the sample strategies in the strategy indicator set is a high-quality strategy; Obtaining a loss value corresponding to a preset loss function through the predicted probability and the strategy mark; The model parameters of the initialized policy classification model are updated using the loss value, and the trained policy classification model is constructed based on the updated model parameters.

13. A policy processing method for a policy classification model, characterized in that: The risk control strategy test model is obtained by training the method according to any one of claims 1 to 12, the method comprising: Obtain the risk control strategy to be tested; Obtaining, by means of the strategy classification model, a predicted probability that the risk control strategy to be tested is a high-quality strategy; The predicted probability is subjected to score conversion processing to obtain a strategy score of the risk control strategy to be tested.

14. The method according to claim 13, characterized in that The step of performing score conversion processing on the predicted probability to obtain a strategy score of the risk control strategy to be tested includes: Obtaining an odds ratio using the predicted probability; Obtain translation parameters through preset benchmark scores, score change values, and benchmark ratios; Obtaining a scaling parameter by performing logarithmic transformation on the preset fractional change value; A linear transformation process is performed based on the translation parameter, the scaling parameter and the probability ratio to obtain a strategy score of the risk control strategy to be tested.

15. The method according to claim 14, characterized in that After obtaining the strategy score of the risk control strategy to be tested, the method further includes: Performing a grading process on the risk control strategy to be tested according to a preset strategy scoring threshold, wherein the strategy scoring threshold includes a first scoring threshold and a second scoring threshold, and the first scoring threshold is higher than the second scoring threshold; In response to the strategy score being lower than the second score threshold, marking the risk control strategy to be tested as a low-quality strategy and taking it offline; In response to the strategy scoring threshold being higher than the second scoring threshold and lower than the first scoring threshold, marking the risk control strategy to be tested as a strategy to be optimized; In response to the strategy scoring threshold being higher than the first scoring threshold, the risk control strategy to be tested is marked as a high-quality strategy.

16. A device for constructing a strategy classification model, characterized in that: The device comprises: an acquisition module, configured to acquire a plurality of sample strategies and acquire a strategy tag of each of the sample strategies, wherein the strategy tag is used to characterize whether the sample strategy is a high-quality strategy or a low-quality strategy, and the sample strategy is used to perform risk control on a business; A construction module is used to obtain a plurality of test indicators for each of the sample strategies, wherein the test indicators are used to quantitatively test the sample strategies; The construction module is further configured to obtain an indicator weight of each test indicator corresponding to the sample strategy, wherein the indicator weight is used to quantitatively represent the importance of the test indicator when testing the sample strategy; The construction module is further configured to construct a strategy indicator set according to the plurality of indicator weights corresponding to each sample strategy; A training module is used to train a strategy classification model using the strategy indicator set, the strategy label and a preset loss function, wherein the strategy classification model is used to predict the probability that the strategy to be tested is a high-quality strategy.

17. A policy processing device for a policy classification model, characterized in that: The device comprises: Acquisition module, used to obtain the risk control strategy to be tested; A prediction module, configured to obtain, through the strategy classification model, a predicted probability that the risk control strategy to be tested is a high-quality strategy; The scoring module is used to perform scoring conversion processing on the predicted probability to obtain a strategy score of the risk control strategy to be tested.

18. An electronic device, characterized in that: The electronic device comprises: a memory for storing computer-executable instructions; A processor is configured to implement the method for constructing a policy classification model as described in any one of claims 1 to 12 or the policy processing method for the policy classification model as described in any one of claims 13 to 15 when executing computer-executable instructions stored in the memory.

19. A computer-readable storage medium storing computer-executable instructions or a computer program, characterized in that: When the computer executable instructions or computer program are executed by a processor, the method for constructing a policy classification model according to any one of claims 1 to 12 or the policy processing method for a policy classification model according to any one of claims 13 to 15 is implemented.

20. A computer program product comprising computer executable instructions or a computer program, characterized in that When the computer executable instructions or computer program are executed by a processor, the method for constructing a policy classification model according to any one of claims 1 to 12 or the policy processing method for a policy classification model according to any one of claims 13 to 15 is implemented.

Images