Parameter model training method, application method, training device and application device

Abstract

The invention provides a training method, an application method, a training device and an application device of a parameter model, relates to the technical field of data processing, and can solve the problem that the accuracy and efficiency of manual risk slow release capability evaluation are too low. According to the specific technical scheme, the method comprises the steps of obtaining object attribute features of a plurality of sample objects; calculating a first value parameter of the sample object according to a data operation rule related to the financial supervision rule; and taking the object attribute feature and the first value parameter as training samples, training the first parameter model to obtain a trained second parameter model, and obtaining a second value parameter of the target object based on the object attribute feature of the target object so as to obtain the risk slow release value of the target object under the financial supervision rule. By implementing the scheme, the accuracy and the automation degree of risk slow release value calculation can be improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, and in particular, to a parameter model training method and application method, a parameter model training device and application device. BACKGROUND

[0002] Currently, financial institutions release risk exposure through specific objects (such as collateral) to release risk. However, the risk release value of the specific object is usually manually evaluated by business personnel, and the accuracy and calculation efficiency of the obtained risk release value are low. SUMMARY

[0003] Embodiments of the present application provide a parameter model training method, an application method, a training device and an application device, which can improve the accuracy and automation of risk release value calculation.

[0004] To achieve the above-mentioned purpose, the embodiments of the present application adopt the following technical solutions: In a first aspect, a parameter model training method is provided. The method first obtains object attribute features of a plurality of sample objects, the sample objects being collateral or guarantee objects for releasing risk exposure to release risk, and the object attribute features being used to indicate whether the sample objects meet corresponding financial regulatory rules. Then, a first value parameter of the sample objects is calculated according to a data operation rule related to the financial regulatory rules, the first value parameter being used to indicate a risk release value of the sample objects and serving as a model label of the object features of the sample objects. Then, the object attribute features and the first value parameter of each sample object in the plurality of sample objects are respectively taken as training samples to train a first parameter model, and a trained second parameter model is obtained. The first parameter model is constructed and fused based on at least two decision tree algorithms, and the decision tree algorithms include random forest and gradient boosting tree. The trained second parameter model has the ability to obtain a second value parameter of a target object based on object attribute features of the target object, and the second value parameter is used to indicate a risk release value of the target object under the financial regulatory rules.

[0005] The embodiments of the present application convert the financial regulatory rules into computable data operation rules to generate model labels, which not only solves the problems of insufficient sample data and poor data quality under new financial regulatory rules, but also ensures the consistency of label quality of model training and regulatory requirements, thereby improving the accuracy and compliance of the model prediction results. Moreover, the fusion modeling method based on multiple decision tree algorithms fully utilizes the robustness of random forest in processing high-dimensional features and the accuracy of gradient boosting tree in capturing complex relationships between features, so that the model can more comprehensively learn the mapping relationship between object attribute features and risk mitigation value, improve the generalization ability and prediction accuracy of the model, and thus improve the accuracy of risk mitigation value calculation. In addition, the method realizes the automation and intelligentization of risk mitigation value calculation, reduces the workload and error risk of manual calculation, and improves the automation degree of risk mitigation value calculation.

[0006] In a possible implementation form of the first aspect, the sample object includes a collateral, and the object attribute features include one or more of the following: at least one of an identifier indicating whether the collateral is prudent, an identifier indicating whether the collateral ownership is legal and non-controversial, an identifier indicating whether the collateral is subject to inquiry, freezing and deduction, an identifier indicating whether the collateral is registered, an identifier indicating whether the collateral is enforceable, a priority order, an identifier indicating whether the collateral valuation is prudent, and a positive correlation identifier.

[0007] These object attribute features comprehensively cover the key regulatory elements of the financial regulatory rules for the mortgage or guarantee, including compliance review (such as ownership legality, whether to be subject to inquiry, freezing and deduction), risk control elements (such as valuation prudence, priority order) and enforceability judgment. By taking these key regulatory elements as input features of the model, the model can fully learn the internal logic of the financial regulatory rules, thereby accurately identifying the key factors affecting the risk mitigation value in actual application, and improving the adaptability of the model to the financial regulatory rules and the explainability of the prediction results.

[0008] In another possible implementation form of the first aspect, the data operation rule includes a conversion coefficient. Calculating the first value parameter of the sample object according to the data operation rule related to the financial regulatory rule includes: obtaining the original value data of the sample object and the conversion coefficient; and determining the first value parameter of the sample object according to the original value data of the sample object and the conversion coefficient.

[0009] The discount factor is a crucial parameter in financial regulatory rules used to adjust the value of collateral or security. It reflects the regulatory authorities' calculation standards for the risk mitigation capabilities of collateral or security under different types and conditions. By introducing the discount factor into risk mitigation value calculation, the regulatory requirements in financial regulatory rules can be transformed into specific calculation logic, ensuring that the primary value parameter, when used as a model label, accurately reflects the risk mitigation value of collateral or security under financial regulatory rules. This label generation method based on financial regulatory rules improves the model's applicability and reliability.

[0010] In another possible implementation of the first aspect, the conversion factor includes at least one of a first conversion factor corresponding to the type of sample object and a second conversion factor calculated based on the validity period of the sample object and the termination period of the risk exposure. If the value unit of the sample object differs from the value unit of the risk exposure, the conversion factor also includes a preset third conversion factor. The first conversion factor reflects the basic discount requirements of regulatory rules for different types of collateral or guarantees, reflecting the inherent risk differences of different risk mitigation tools. The second conversion factor considers the impact of the time dimension on the risk mitigation effect; when there is a mismatch between the validity period of the collateral or guarantee and the risk exposure period, a discount adjustment is used to reasonably reflect the risk mitigation capability. The third conversion factor solves the value conversion problem when value units are not uniform, ensuring the accuracy of value calculation in cross-currency or cross-unit-of-measure scenarios.

[0011] Electronic devices can determine the conversion ratio based on the various items included in the conversion factor, and multiply the original value data of the sample object by the conversion ratio to obtain the first value parameter of the sample object.

[0012] By integrating conversion factors from different dimensions, the calculation of the first value parameter becomes more comprehensive and accurate, thereby providing high-quality training labels for the model and improving the effectiveness of model training and the accuracy of prediction results.

[0013] In another possible implementation of the first aspect, before calculating the first value parameter of the sample object according to the data processing rules related to financial regulatory rules, the method further includes: determining the eligibility of the sample object. The eligibility of the sample object includes at least one of the following: the sample object is a first object type; the characteristic value of the first object attribute feature of the sample object is a first value, and the first object attribute feature is one of the object attribute features of the sample object; the effective period of the sample object is greater than the termination period of the risk exposure.

[0014] By conducting eligibility assessments before value calculation, sample objects that do not meet the necessary requirements of financial regulatory rules are screened out in advance, avoiding the negative impact of unqualified samples on model training. Specifically, the restriction on object types ensures that only collateral or guarantee types recognized by financial regulatory rules are included in model training, improving the quality of training data. Requirements for specific object attribute features guarantee that sample objects meet the key conditions in financial regulatory rules, ensuring that the sample data used for model training is valid data that complies with financial regulatory rules. The determination of the effective period ensures, from a time perspective, that the collateral or guarantee can cover the time of risk exposure, avoiding the risk mitigation failure caused by maturity mismatch.

[0015] This regulatory-rule-based sample selection mechanism not only improves the compliance and representativeness of training data, but also enables the trained model to more accurately identify and evaluate risk mitigation tools that meet regulatory requirements, thereby increasing the model's application value in actual business.

[0016] Secondly, a method for applying a parametric model is provided. This method includes: obtaining the object attribute characteristics of a target object, where the target object is collateral or security used to mitigate risk exposure, and the object attribute characteristics of the target object are used to indicate whether the target object meets the corresponding financial regulatory rules; processing the object characteristics of the target object through a trained second parametric model to obtain a second value parameter of the target object, which is used to indicate the risk mitigation value of the target object.

[0017] By implementing the embodiments of this application, the risk mitigation value of a target object is automatically evaluated through a trained second-parameter model, thereby improving the efficiency of risk assessment. Since the model is trained based on sample data that complies with financial regulatory rules, the second-value parameter output by the model accurately reflects the risk mitigation value under the regulatory rules, improving the accuracy and automation of risk mitigation value calculation.

[0018] Thirdly, a training device for a parametric model is provided, comprising: The feature acquisition module is used to acquire the object attribute features of multiple sample objects; wherein, the multiple sample objects are collateral or guarantees used to mitigate risk exposure, and the object attribute features of the sample objects are features used to indicate whether the sample objects meet the corresponding financial regulatory rules. The value calculation module is used to calculate the first value parameter of the sample object according to the data operation rules related to financial regulatory rules; wherein, the first value parameter is used to indicate the risk mitigation value of the sample object, and the first value parameter serves as the model label of the object feature of the sample object; The model training module is used to train the first parameter model by taking the object attribute features and first value parameters of each sample object from multiple sample objects as training samples, and then obtaining the trained second parameter model. The first parameter model is constructed and fused based on at least two decision tree algorithms, including random forest and gradient boosting tree. Among them, the second parameter model has the ability to obtain the second value parameter of the target object based on the object attribute characteristics of the target object. The second value parameter is used to indicate the risk mitigation value of the target object under financial regulatory rules.

[0019] Fourthly, an application device for a parametric model is provided, comprising: The feature acquisition module is used to acquire the object attribute features of the target object; wherein, the target object is collateral or guarantee used to mitigate risk exposure, and the object attribute features of the target object are features used to indicate whether the target object meets the corresponding financial regulatory rules; The model application module is used to process the object features of the target object through the trained second parameter model in the above embodiments to obtain the second value parameter of the target object. The second value parameter is used to indicate the risk mitigation value of the target object.

[0020] Fifthly, an electronic device is provided, the method comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, the computer program code including computer instructions. When the processor executes the computer instructions, it causes the electronic device to perform the method as described in the first aspect, the second aspect, and any possible implementation thereof.

[0021] Sixthly, embodiments of this application provide a computer-readable storage medium storing computer instructions. When executed by a processor, these computer instructions are used to implement the methods described in the first aspect, the second aspect, and any possible implementation thereof.

[0022] In a seventh aspect, embodiments of this application provide a computer program product that, when run on a computer or executed by a computer's processor, implements the method described in the first aspect and any possible design thereof. The computer may be the electronic device described in the fifth aspect and any possible implementation thereof.

[0023] Understandably, the beneficial effects that can be achieved by the training device for the parametric model in the third aspect, the application device for the parametric model in the fourth aspect, the electronic device in the fifth aspect, the computer-readable storage medium in the sixth aspect, and the computer program product in the seventh aspect can be referred to as the beneficial effects in the first aspect, the second aspect, and any possible implementation thereof, and will not be repeated here. Attached Figure Description

[0024] Figure 1 A flowchart illustrating a training method for a parametric model provided in an embodiment of this application; Figure 2 A schematic diagram of a parametric model training process provided in an embodiment of this application; Figure 3 A flowchart illustrating an application method of a parameter model provided in an embodiment of this application; Figure 4 A schematic diagram of the structure of a training device for a parametric model provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an application device for a parameter model provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0027] The collection, storage, use, processing, transmission, provision, and disclosure of information such as object attribute characteristics involved in the technical solutions provided in this application comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0028] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0029] In the operations of financial institutions, there are risk exposures. These exposures refer to assets that expose the financial institution to credit risk, such as a loan. Financial institutions mitigate these exposures through collateral, also known as security or collateral. Risk mitigation specifically refers to the business activity of reducing the credit risk of risk exposures through collateral; it can also be called credit risk mitigation.

[0030] To comply with relevant financial regulatory rules, such as Basel III (referred to as Basel III in this application), financial institutions are also required to conduct capital measurement on collateral. Capital measurement refers to the calculation by financial institutions, in accordance with financial regulatory rules, of the amount of capital they need to hold to cope with risks. The accuracy of the calculation of the mitigation capacity of the collateral used for risk mitigation also affects the accuracy of capital measurement.

[0031] In related technologies, financial institutions typically employ manual verification when assessing or calculating the risk mitigation capacity of collateral. Specifically, staff manually examine various attributes of the collateral to determine if it meets financial regulatory requirements and calculate its mitigation value. However, this manual verification method has the following drawbacks: The quality of collateral data is often poor, with potential issues such as missing data or inaccurate values, which can lead to biases in the manually calculated mitigation capacity, thus affecting the accuracy of capital measurement. Furthermore, manual verification is inefficient and struggles to meet the stringent requirements of financial regulations, potentially resulting in regulatory risks and economic losses due to non-compliance.

[0032] Based on this, embodiments of this application provide a method for training and applying a parametric model. This method acquires the object attribute features of multiple sample objects, calculates the first value parameter of the sample objects as model labels according to data operation rules related to financial regulatory rules, and trains the parametric model based on a decision tree algorithm. This enables the trained parametric model to automatically predict the risk mitigation value of collateral. By converting financial regulatory rules into computable data operation rules to generate model labels, this method not only solves the problems of insufficient sample data and poor data quality under new financial regulatory rules, but also ensures the consistency between the label quality of the model training and regulatory requirements. The model can learn the mapping relationship between object attribute features and risk mitigation value during model training, improving the model's generalization ability and prediction accuracy, thereby improving the accuracy and automation of risk mitigation value calculation.

[0033] The parameter model training and application methods provided in this application can be applied to electronic devices with data processing capabilities, such as servers. Alternatively, the electronic device may include a personal computer (PC), tablet computer, laptop computer, portable computer (such as a mobile phone), wearable electronic device (such as a smartwatch), augmented reality (AR) / virtual reality (VR) device, in-vehicle computer, etc. The following embodiments do not impose special limitations on the specific form of the electronic device. The execution subject of the parameter model training and application methods provided in this application can be the aforementioned electronic device, or it can be the corresponding parameter model training device and application device, which can be integrated into the electronic device or the processor of the electronic device. For ease of explanation, the following content of this application embodiment will be described using an electronic device as the execution subject.

[0034] Please see Figure 1 , Figure 1 This is a flowchart illustrating a training method for a parametric model provided in an embodiment of this application.

[0035] S101, the electronic device acquires the object attribute features of multiple sample objects.

[0036] In this embodiment, the sample object is collateral or guarantee used to mitigate the risk of a risk exposure. It should be understood that, in order to reduce the credit risk of a risk exposure, financial institutions may require the customer corresponding to that risk exposure to provide collateral or guarantee. For ease of description, the sample object may also be referred to as collateral in this embodiment.

[0037] The sample objects can include various types of collateral. For example, collateral can be real estate collateral, including residential real estate, commercial real estate, and construction in progress. Collateral can also be financial collateral, including cash, bonds, stocks, and other financial assets. Furthermore, collateral can be accounts receivable pledges, including transaction-related accounts receivable and rent receivable.

[0038] Different types of collateral possess different attribute characteristics. That is, different sample objects can have different sample attribute characteristics. Object attribute characteristics are used to indicate whether a sample object meets the corresponding financial regulatory rules. These characteristics reflect the compliance status and risk characteristics of the collateral under the regulatory rules. Electronic devices read the object attribute characteristics of sample objects from a database storing collateral data. This database can be a database within a financial institution's credit system, collateral management system, or capital measurement system. Electronic devices can obtain collateral-related data from these systems through data interfaces, achieving automatic acquisition of collateral-related data.

[0039] In one specific implementation, the object attribute features include one or more of the following: an indicator of whether the collateral is prudent, an indicator of whether the ownership of the collateral is legal and undisputed, an indicator of whether the collateral is subject to investigation, freezing or seizure, an indicator of whether the collateral has been registered, an indicator of whether the collateral is executable, priority, an indicator of whether the valuation of the collateral is prudent, and a positive correlation indicator.

[0040] The following sections will provide a detailed explanation of these object attributes.

[0041] The first type of object attribute feature: an indicator that shows whether the collateral has been prudently handled.

[0042] The label indicating whether collateral is prudent can also be called a prudential label. This label is used to indicate whether the collateral meets prudential requirements. Prudence requirements are basic requirements for collateral stipulated by financial regulatory rules to ensure that the collateral has risk mitigation capabilities.

[0043] In one example, for real estate collateral that is not under construction, the prudent label is marked as yes; for collateral that is under construction, the prudent label is marked as no due to the high uncertainty of its value.

[0044] In another embodiment, if a construction-in-progress collateral reaches completion during the period of risk exposure, i.e., it is converted from a construction-in-progress collateral to a non-construction-in-progress collateral, the electronic device responds to the collateral type change operation in the business process system, redetermines the collateral type, re-associates the risk exposure with the collateral, and determines that the value of the prudential identification of the collateral is yes.

[0045] It should be noted that "construction in progress" collateral refers to projects that are not yet completed, and the value of such collateral is uncertain. "Completion level" refers to the degree to which the construction of the project is finished and ready for use. Once the project reaches this completion level, the uncertainty of the value of the construction in progress collateral decreases, and the prudence requirement can be met.

[0046] The second type of object attribute characteristic: an identifier indicating that the ownership of the collateral is legal and undisputed.

[0047] An indication that the ownership of collateral is legal and undisputed is used to indicate whether the ownership of the collateral is legal and undisputed. The legality and undisputed ownership of collateral are prerequisites for its legal disposal.

[0048] In one specific implementation, the initial value of this identifier is "No," indicating that the collateral ownership is legal and undisputed. If the collateral ownership is defective or disputed, the value of this identifier is "Yes." For example, if there are multiple disputes over the ownership of a property, or if the property title certificate is defective, the value of this identifier is "Yes."

[0049] It should be noted that ownership refers to the attribution of property. Defective ownership refers to a legal flaw in the ownership of property.

[0050] The third type of object attribute feature: an indicator that shows whether the collateral is subject to investigation, freezing, or seizure.

[0051] The markings indicating whether collateral has been seized, frozen, or detained are used to indicate whether the collateral has been seized, frozen, or detained by relevant authorities. It is understandable that seizure, freezing, or detention would prevent the collateral from being disposed of normally, affecting its risk mitigation capabilities.

[0052] In one specific implementation, the initial value of this identifier is "No," indicating that the collateral is not subject to seizure, freezing, or detention. If the collateral is seized, frozen, or detained, the value of this identifier will be "Yes." For example, if a property is seized by relevant authorities due to a dispute, the value of this identifier will be "Yes."

[0053] It should be noted that "seizure" refers to a coercive measure by which relevant departments prohibit the transfer or disposal of property. "Freezing" refers to a coercive measure by which relevant departments restrict the use or disposal of property. "Detention" refers to a coercive measure by which relevant departments temporarily confiscate property.

[0054] The fourth type of object attribute feature: an identifier indicating whether the collateral has been registered.

[0055] An indicator indicating whether collateral has been registered serves to demonstrate whether a valid mortgage registration has been completed. Mortgage registration is a necessary condition for collateral to have risk mitigation capabilities.

[0056] In one specific implementation, the electronic device determines whether the collateral has been registered based on whether the current time is within the validity period of the collateral. If the current time is within the validity period of the collateral, the value of the identifier is "yes," indicating that the collateral has been legally registered. If the current time is outside the validity period of the collateral, the value of the identifier is "no," indicating that the collateral has not been legally registered or that the registration has expired.

[0057] It should be noted that mortgage registration refers to the act of registering and filing the mortgage right on collateral with the relevant institution. The effective period of the collateral refers to the time from when the mortgage registration begins to have legal effect to when the mortgage registration ceases to have legal effect.

[0058] The fifth type of object attribute feature: an identifier indicating whether the collateral is executable.

[0059] The indicator indicating whether collateral is enforceable can also be called the indicator of whether collateral is enforceable under the relevant contract. This indicator is used to indicate whether the collateral is enforceable under the relevant contract, that is, whether the financial institution can dispose of the collateral if the debtor corresponding to the collateral defaults.

[0060] In one specific implementation, the electronic device can determine whether the collateral is executable based on the values ​​of the prudence indicator, the indicator indicating that the collateral is legally and undisputed, the indicator indicating whether the collateral is subject to seizure or freezing, and the indicator indicating whether the collateral has been registered. When the prudence indicator is positive, the indicator indicating that the collateral is legally and undisputed is negative, the indicator indicating whether the collateral is subject to seizure or freezing is negative, and the indicator indicating whether the collateral has been registered is positive, it indicates that the collateral meets the prudence requirements, has legal and undisputed ownership, is not subject to seizure or freezing, and has been legally registered. In this case, the value of the indicator indicating whether the collateral is executable is positive. When any of the above four indicators does not meet the above conditions, the value of the indicator indicating whether the collateral is executable is negative.

[0061] It should be noted that enforceability refers to a financial institution's ability to legally dispose of collateral in the event of a debtor's default. This indicator reflects whether the collateral simultaneously meets multiple regulatory requirements and represents a comprehensive assessment of its compliance.

[0062] The sixth type of object attribute characteristic: priority.

[0063] Priority ranking is used to indicate the order of collateralization. When multiple exposures correspond to the same collateral, there is a priority order for collateralization. The first priority collateralization order indicates that the exposure has priority in obtaining the collateral value in the event of default.

[0064] In this embodiment of the application, according to financial regulatory rules, if the collateral is in the first priority category, then the collateral meets the regulatory requirements. If the collateral is not in the first priority category, then the regulatory requirements are not met.

[0065] It should be noted that the priority of a mortgage refers to the order in which multiple creditors have mortgage rights over the same collateral. First priority means that the creditor has priority over the mortgage right on the collateral, that is, they can receive compensation first when the collateral is disposed of.

[0066] The seventh type of object attribute feature: an indicator of whether the valuation of collateral is prudent.

[0067] Indicators indicating whether collateral valuation is prudent are used to indicate whether the valuation of collateral meets the prudence requirement. The prudence requirement for collateral valuation means that the assessed value of the collateral is objective, that is, neither overestimating nor underestimating the actual value of the collateral.

[0068] In one specific implementation, the value of this identifier can be set to "yes" by default. If it is determined that the valuation of the collateral is imprudent, the electronic device then modifies the value of this identifier to "no".

[0069] It should be noted that collateral valuation refers to the process of assessing the value of collateral. Prudent valuation means assessing the value of collateral according to objective and conservative principles, avoiding overestimation of collateral value.

[0070] The eighth type of object attribute feature: positive correlation identifier.

[0071] The positive correlation indicator is used to indicate whether there is a positive correlation between the value of collateral and the risk level of the debtor corresponding to the collateral.

[0072] It's important to note that positive correlation refers to a relationship where the value of collateral and the debtor's risk level move in the same direction. When the debtor's risk level increases, a positive correlation exists if the value of the collateral also decreases. For example, when a real estate development company uses its developed properties as collateral, if the company's business deteriorates, the value of its developed properties may also decrease; in this case, the collateral value and the debtor's risk level are positively correlated. If there is a positive correlation between collateral and the debtor's risk level, the risk mitigation capability of the collateral decreases.

[0073] In one example, when the collateral type includes transaction receivables with an original term of less than one year, rent receivables with an original term of less than one year, or financial collateral, the electronic device can determine that the collateral has a positive correlation.

[0074] In another example, when the collateral type is commercial or residential real estate, the electronic device determines the positive correlation indicator for that collateral's credit line. For domestic collateral, if the credit line of the collateral's contract is real estate development, the electronic device determines the value of the positive correlation indicator to be no. For overseas collateral, if the product associated with the collateral's contract includes real estate development, the electronic device determines the value of the positive correlation indicator to be yes. For other businesses with credit lines not associated with real estate development, the electronic device determines the value of the positive correlation indicator to be no.

[0075] By defining and acquiring the above-mentioned object attribute features, electronic devices can comprehensively obtain the feature information of sample objects, providing a data foundation for subsequent model training.

[0076] S102, the electronic device calculates the first value parameter of the sample object in accordance with the data operation rules related to financial regulatory rules.

[0077] In this embodiment, the first value parameter is used to indicate the risk mitigation value of the sample object. The first value parameter serves as a model label for the object's characteristics. Risk mitigation value refers to the value of collateral that can be used to reduce credit risk exposure under financial regulatory rules. The model label refers to the true value used to train the model in machine learning.

[0078] Data processing rules are computable rules derived from financial regulatory rules. These rules stipulate relevant conditions for assessing the risk mitigation capability of collateral. Electronic devices convert these conditions into data processing rules, thereby automatically calculating the primary value parameters of sample objects.

[0079] In one specific implementation, the data processing rules include a discount factor. This discount factor is used to adjust the original value of the collateral at a discount to obtain a risk mitigation value that meets regulatory requirements.

[0080] Specifically, the electronic device acquires the original value data and conversion factor of the sample object. Original value data refers to the appraised value or distributable value of the collateral. Distributable value refers to the portion of the collateral that can be used to repay the risk exposure upon disposal. Based on the original value data and conversion factor of the sample object, the electronic device determines the first value parameter of the sample object.

[0081] For example, a discount factor can be used to indicate the portion of the original value of collateral that should be retained. Electronic devices can multiply the original value data of a sample object by the discount factor to obtain a first value parameter. As another example, a discount factor can be used to indicate the portion of the original value of collateral that should be removed. The discount factor can be less than 1. Electronic devices can calculate 1 minus the discount factor to obtain a first residual value, and then multiply the original value data of the sample object by this first residual value to obtain a first value parameter.

[0082] In some specific embodiments, the conversion factor includes at least one of a first conversion factor corresponding to the type of the sample object and a third conversion factor indicating that the value unit of the sample object differs from the value unit of the risk exposure. When the value unit of the sample object differs from the value unit of the risk exposure, the conversion factor also includes a preset third conversion factor. That is, the conversion factor may include the first conversion factor and / or the third conversion factor depending on the actual situation; however, when the value unit of the sample object differs from the value unit of the risk exposure, the electronic device needs to use the third conversion factor as one of the conversion factors used in this calculation of the first value parameter.

[0083] These conversion factors will be explained in detail below.

[0084] The first type of conversion factor: the first conversion factor.

[0085] The first conversion factor is a preset standard conversion factor. This conversion factor can correspond to the type of collateral. For example, financial collateral and real estate collateral can correspond to different first conversion factors.

[0086] In a specific example, for financial collateral, the first discount factor can be calculated using the following formula: ; in, The first conversion factor is used. The standard conversion factor is for a minimum holding period of 10 days. This refers to the actual number of trading days between adjustments to capital market trading margins or revaluations of mortgage loans. This is the minimum holding period for the transaction.

[0087] It should be noted that the minimum holding period refers to the shortest time required to dispose of the collateral. The actual interval trading days refers to the time interval for revaluing the collateral.

[0088] The second type of conversion factor: the third type of conversion factor.

[0089] The third conversion factor can also be a preset factor, but it can only be used to convert the original value data of the sample object when the value unit of the sample object is different from the value unit of the risk exposure. Since the third conversion factor and the first conversion factor convert the original value data in the same way, the third conversion factor will be explained as the second type of conversion factor first, and the second conversion factor will be explained as the third type of conversion factor later.

[0090] Understandably, exchange rate risk exists when the currency of collateral differs from the currency of the risk exposure. A third conversion factor is used to adjust for this exchange rate risk. In one example, if the value unit of the sample object differs from the value unit of the risk exposure, i.e., a currency mismatch exists, the third conversion factor would be 8%.

[0091] When the conversion factor includes a third conversion factor, the electronic device can calculate 1 minus the third conversion factor to obtain the second residual value, and multiply the original value data of the sample object by the second residual value to obtain the first value parameter. Alternatively, when the conversion factor includes a first conversion factor and a second conversion factor, the electronic device can calculate 1 minus the third residual value of the first and third conversion factors, and multiply the original value data of the sample object by the third residual value to obtain the first value parameter.

[0092] It's important to note that currency mismatch refers to a situation where the currency used to denominate the collateral is different from the currency used to denominate the risk exposure. For example, if the risk exposure is denominated in RMB, but the collateral is denominated in USD, a currency mismatch exists. Due to exchange rate fluctuations, the actual mitigating value of the collateral may change, thus requiring a discount adjustment.

[0093] The third type of conversion factor: the second conversion factor.

[0094] The second conversion factor is calculated based on the effective period of the sample object and the termination period of the risk exposure. When the effective period of the collateral is less than the termination period of the risk exposure, a maturity mismatch exists. The second conversion factor is used to adjust for this maturity mismatch.

[0095] In a specific example, the electronic device calculates the first value parameter after maturity mismatch adjustment according to the following formula: ; in, The first value parameter after adjustment for maturity mismatch. The first value parameter is the value after adjustment by the first and / or second adjustment factors, before maturity mismatch adjustment. The smaller of the remaining time from the current time to the end of the risk exposure period and 5. The remaining time until the expiration date of the collateral is equal to the current time. The smaller of the two values, where the time period can be in years. The second conversion factor can be... .

[0096] It should be noted that maturity mismatch refers to a situation where the effective period of collateral is shorter than the termination period of the risk exposure. When the effective period of collateral is less than the termination period of the risk exposure, the collateral cannot provide protection for the entire duration of the risk exposure, and therefore the risk mitigation value needs to be discounted.

[0097] By applying the aforementioned conversion factors, electronic devices can transform the original value data of collateral into a risk mitigation value that complies with financial regulatory rules, serving as a label for model training. Specifically, the electronic device can determine a conversion ratio based on at least one of the first, third, and second conversion factors, and multiply the original value data of the sample object by the conversion ratio to obtain the first value parameter of the sample object.

[0098] According to the above, when the conversion factor includes a first conversion factor and / or a third conversion factor, the electronic device can subtract the first conversion factor and / or the third conversion factor from 1 to obtain the conversion ratio, which is one of the first residual value, second residual value, and third residual value mentioned above. When the conversion factor includes a second conversion factor, the electronic device can directly use the second conversion factor as the conversion ratio.

[0099] When the conversion factor includes at least one of the first and third conversion factors, and also includes the second conversion factor, the electronic device can multiply one of the first, second, and third residual values ​​mentioned above by the third conversion factor to obtain the conversion ratio. Specifically, when the conversion factor includes both the first and third conversion factors, the conversion ratio is the product of the first residual value and the third conversion factor; when the conversion factor includes both the second and third conversion factors, the conversion ratio is the product of the second residual value and the third conversion factor; and when the conversion factor includes all three conversion factors, the conversion ratio is the product of the third residual value and the third conversion factor.

[0100] By implementing this embodiment, the original value data of the sample object can be accurately converted, thereby improving the reliability and accuracy of the first value parameter.

[0101] In some specific embodiments, the electronic device may also determine that the sample object is qualified before calculating the first value parameter of the sample object in accordance with the data calculation rules related to financial regulatory rules.

[0102] A qualified sample object includes at least one of the following: the sample object is of the first object type; the characteristic value of the first object attribute of the sample object is the first value; the validity period of the sample object is greater than the termination period of the risk exposure.

[0103] The following provides a detailed explanation of these eligibility criteria.

[0104] The first qualification condition is that the sample object belongs to the first object type.

[0105] The first category refers to the types of collateral that can be used as pre-defined risk mitigation tools. According to financial regulatory rules, only specific types of collateral can serve as eligible risk mitigation instruments. For example, under one evaluation system, financial collateral, accounts receivable collateral, and real estate collateral can be considered eligible mitigation tools.

[0106] The electronic device determines whether the type of the sample object belongs to the first object category. If the type of the sample object belongs to the first object category, the qualification condition is met, and the electronic device determines that the sample object is qualified. If the type of the sample object does not belong to the first object category, the qualification condition is not met, and the electronic device determines that the sample object is unqualified.

[0107] The second qualification condition is that the feature value of the first object attribute feature of the sample object is the first value.

[0108] The first object attribute feature is one of the object attribute features of a sample object. The first value is the feature value that meets financial regulatory rules. For example, the first object attribute feature can be a positive correlation indicator, and the first value can be negative, indicating that there is no positive correlation between the value of the collateral and the debtor's risk level.

[0109] The electronic device determines whether the feature value of the first object attribute of the sample object is the first value. If the feature value of the first object attribute of the sample object is the first value, then the qualification condition is met, and the electronic device determines that the sample object is qualified. If the feature value of the first object attribute of the sample object is not the first value, then the qualification condition is not met, and the electronic device determines that the sample object is unqualified.

[0110] In a specific example, the electronic device determines whether the positive correlation indicator for a sample object is negative. If the positive correlation indicator is positive, it means that the value of the collateral is positively correlated with the debtor's risk level, and the electronic device determines that the sample object is ineligible. If the positive correlation indicator is negative, it means that the value of the collateral is not positively correlated with the debtor's risk level, and the ineligibility condition is met.

[0111] The third eligibility criterion is that the validity period of the sample subject is longer than the termination period of the risk exposure. In the above embodiments, the value of collateral with maturity mismatch can be adjusted using a second conversion factor. In this embodiment, the electronic device determines whether the validity period of a sample object is greater than the termination date of the risk exposure, thereby excluding data of collateral with maturity mismatch from the model training sample data. Specifically, if the validity period of a sample object is greater than the termination date of the risk exposure, the qualification condition is met, there is no maturity mismatch, and the electronic device determines that the sample object is qualified. If the validity period of a sample object is not greater than the termination date of the risk exposure, a maturity mismatch exists, and the electronic device can determine that the sample object is unqualified.

[0112] In one specific implementation, the electronic device comprehensively judges whether a sample object is qualified based on the aforementioned qualification criteria. Only when a sample object meets all qualification criteria is the electronic device determined that the sample object is qualified, and the first value parameter of the sample object is calculated. If the sample object does not meet any qualification criteria, the electronic device determines that the sample object is unqualified. In this case, the first value parameter can be set to 0, or the data of the sample object can not be used in the subsequent model training process.

[0113] By assessing the eligibility of sample objects, it can be ensured that the sample data used for model training complies with the requirements of financial regulatory rules, thereby improving the quality of model training.

[0114] In some embodiments, when multiple sample objects jointly guarantee the same risk exposure, the electronic device needs to split the risk exposure and allocate the mitigation value in order of collateral liquidity from strong to weak.

[0115] Specifically, the electronic device breaks down risk exposures in the order of financial collateral, accounts receivable, commercial and residential real estate, and other collateral. For each collateral, the electronic device calculates the portion of the risk exposure that collateral can mitigate and calculates the corresponding first value parameter.

[0116] For example, a risk exposure of 10 million yuan has three collaterals: financial collateral A (distributable value of 3 million yuan), accounts receivable collateral B (distributable value of 4 million yuan), and real estate collateral C (distributable value of 5 million yuan). The server first uses financial collateral A to mitigate the 3 million yuan risk exposure and calculates the first value parameter for this portion. Then, it uses accounts receivable collateral B to mitigate 4 million yuan of the remaining 7 million yuan and calculates the first value parameter for this portion. Finally, it uses real estate collateral C to mitigate the remaining 3 million yuan and calculates the first value parameter for this portion.

[0117] It should be noted that collateral liquidity refers to the ease and speed at which collateral can be converted into funds upon disposal. Financial collateral typically has the highest liquidity, followed by accounts receivable, while real estate has relatively lower liquidity. Allocating mitigating value according to liquidity order complies with financial regulatory requirements.

[0118] By applying the aforementioned data processing rules, electronic devices can accurately calculate the primary value parameters of sample objects, providing high-quality labeled data for model training. This method of generating labels based on financial regulatory rules solves the problem of obtaining a large amount of real-world labeled data under new financial regulatory rules, while ensuring the consistency of labeled data with regulatory requirements.

[0119] S103, the electronic device uses the object attribute features and first value parameters of each sample object in the multiple sample objects as training samples to train the first parameter model and obtain the trained second parameter model.

[0120] In this embodiment, the first parameter model is constructed and fused based on at least two decision tree algorithms. A decision tree is a machine learning algorithm. A decision tree consists of nodes and directed edges. Nodes are divided into internal nodes and leaf nodes. Internal nodes represent a feature or attribute, and leaf nodes represent the class or label value of a sample. The decision tree uses information gain to select internal nodes and split features. Information gain refers to the fact that after selecting a feature and splitting the samples based on the feature values, the error of each child node is smaller than the error of the parent node. In regression problems, the error can be measured using a loss function, such as the squared loss function.

[0121] Random forest is an ensemble learning algorithm. It consists of multiple decision trees. Each decision tree uses a subset of the training samples during training, and only considers a subset of features when splitting at each node. This randomization reduces the risk of overfitting and improves the model's generalization ability. During prediction, the random forest averages or votes on the predictions from multiple decision trees to obtain the final prediction.

[0122] Gradient boosting trees are also a type of ensemble learning algorithm. The basic idea of ​​gradient boosting trees is that each decision tree corrects the errors of the previous one. Specifically, the first decision tree is trained using training samples. The second decision tree is trained using the prediction errors of the first decision tree to correct its errors. This process continues, with each new decision tree correcting the accumulated errors of all previous decision trees. During prediction, the gradient boosting tree sums the predictions from all the decision trees to obtain the final prediction.

[0123] In this embodiment, the electronic device constructs a first parameter model based on two decision tree algorithms: random forest and gradient boosting tree, such as the XGBoost model.

[0124] The server uses the object attribute features of multiple sample objects and a first value parameter to form a training dataset, thereby training the model for that first parameter. Specifically, for each sample object, the server uses the object attribute features of that sample object as the input feature vector and the first value parameter of that sample object as the output label. The input feature vector is a vector containing multiple feature values, each corresponding to an object attribute feature. The output label is a numerical value representing the risk mitigation value of the sample object.

[0125] The server preprocesses the training dataset before training.

[0126] In some embodiments, the server can remove features that meet preset conditions. For example, if the value of a feature is the same in all sample objects, then the variance of the feature is zero, and the feature is considered to meet the preset condition and is meaningless for model prediction; the server can remove the feature. As another example, if the proportion of missing values ​​for a feature is greater than a preset threshold, it indicates that there is too little valid data for that feature and the data quality is poor; the feature is considered to meet the preset condition, and the server can also remove the feature.

[0127] In some embodiments, the server can also remove redundant features. Redundant features are those whose correlation coefficient with other features is greater than a preset threshold. The server calculates the correlation coefficient between any two features, and if the correlation coefficient between two features is greater than the preset threshold, the electronic device can determine that one of the two features is a redundant feature.

[0128] In some embodiments, the server fills in or processes features with missing values. For features with numerical values, the server can fill in missing values ​​using the mean, median, or mode of the feature, without limitation. For features with class values, the server can fill in missing values ​​using the mode of the feature.

[0129] In some embodiments, the server may also retain missing values ​​instead of filling them in. The first parameter model learns how to handle features containing missing values ​​during training, enabling the trained second parameter model to process features containing missing values.

[0130] After completing data preprocessing, the server begins training the first-parameter model until it meets the requirements. The server then uses the qualified first-parameter model as the trained second-parameter model and saves it for use in subsequent model applications.

[0131] The trained second-parameter model possesses the ability to obtain the second value parameter of the target object based on its object attribute features. The target object refers to the collateral whose risk mitigation value needs to be predicted. The second value parameter indicates the risk mitigation value of the target object under financial regulatory rules. The first and second value parameters are obtained in different ways: the first value parameter is calculated through data operation rules, while the second value parameter is predicted by the trained second-parameter model.

[0132] Through the training process described above, the trained second-parameter model can quickly and accurately predict the risk mitigation value of collateral based on its object attribute characteristics, without needing to use specific data calculation rules to calculate the risk mitigation value. Thus, for collateral requiring risk mitigation value prediction, the server can use this second-parameter model to pre-judge the risk mitigation value of the collateral and promptly identify collateral that does not comply with the aforementioned financial regulatory rules.

[0133] In a specific example, when a financial institution issues a loan, the customer provides real estate as collateral. The server obtains the object attribute characteristics of the real estate, including whether it is prudent, whether the ownership of the collateral is legal and undisputed, whether the collateral is subject to investigation, freezing, or seizure, whether the collateral is registered, whether the collateral is legally enforceable, whether it has the highest priority, whether the valuation of the collateral is prudent, and whether it has a positive correlation. The server inputs these object attribute characteristics into a trained second-parameter model, obtaining a second value parameter of 800,000 yuan for the real estate. Based on this second value parameter, the server determines that the real estate can provide a risk mitigation value of 800,000 yuan for the loan. If the loan amount is 1 million yuan, the real estate cannot fully mitigate the credit risk of the loan, and the financial institution needs to require the customer to provide other collateral and guarantees, or require the customer to reduce the loan amount.

[0134] Please refer to Figure 2 , Figure 2 This diagram illustrates a parametric model training process provided in an embodiment of this application. The process includes steps such as data preparation, label definition, data cleaning, feature selection, model training, and model evaluation.

[0135] During the data preparation phase, the server acquires basic data. This basic data includes the raw data of the sample objects, such as basic information about the collateral, its appraised value, mortgage registration information, and credit information.

[0136] During the label definition phase, the server performs mitigation capability analysis. Following data processing rules related to financial regulatory regulations, the server calculates a first value parameter for the sample object and uses this first value parameter as the model label for the sample object. This process involves steps such as eligibility assessment and discount factor calculation, as detailed in the above embodiment.

[0137] During the data cleaning phase, the server cleans the training dataset, as described in the embodiments above. For example, in the missing value handling phase of data cleaning, the server processes features with missing values. The server can use imputation methods, such as using the mean, median, or mode to fill in missing values, as described in the embodiments above.

[0138] During the feature selection phase, the server selects features for training the model. Based on feature selection principles, the server chooses features related to the collateral's risk mitigation capability. Specifically, the features selected by the server include prudential indicators, indicators of whether the collateral's ownership is legal and undisputed, indicators of whether the collateral is subject to investigation, freezing, or seizure, indicators of whether the collateral has been legally registered, indicators of whether the collateral is legally enforceable, priority indicators, indicators of whether the real estate valuation is prudent, and positive correlation indicators, as detailed in the above embodiments.

[0139] During the model training phase, the server constructs sub-models based on random forest and gradient boosting tree respectively, then merges the two sub-models to obtain the first parameter model, and trains the first parameter model, as described in the above embodiment.

[0140] During the model evaluation phase, the server can calculate the prediction error of the parametric model to determine whether the model meets the requirements. If it does, model training is complete; otherwise, the model is retrained. Ultimately, the server obtains the trained second-parameter model.

[0141] This application also provides a method for applying a parametric model. The electronic device can also be the implementer of this method. This method is used to apply the trained second-parameter model to actual business operations to predict the risk mitigation value of a target object. See also... Figure 3 , Figure 3 This is a flowchart illustrating a method for applying a parameter model, as provided in an embodiment of this application.

[0142] S301, The electronic device acquires the object attribute characteristics of the target object.

[0143] The target object is collateral or guarantee used to mitigate risk exposure. The object attribute characteristics of the target object are features used to indicate whether the target object meets the corresponding financial regulatory rules. The method by which the electronic device obtains the object attribute characteristics of the target object can be the same as the method used to obtain the object attribute characteristics of the sample object; for details, please refer to section S101 above. S302, the electronic device processes the object features of the target object through the trained second parameter model to obtain the second value parameter of the target object.

[0144] In this embodiment, the second value parameter is used to indicate the risk mitigation value of the target object. The electronic device inputs the object attribute features of the target object into the trained second parameter model, which processes the object attribute features and outputs the second value parameter of the target object.

[0145] Specifically, the electronic device uses the object attribute features of the target object as an input feature vector. This input feature vector includes the feature values ​​of each object attribute feature of the target object. The electronic device uses this input feature vector as input to a second-parameter model. Based on the mapping relationship between object attribute features and risk mitigation value learned during model training, the second-parameter model calculates and outputs a second value parameter of the target object.

[0146] In one example, an electronic device acquires the object attribute characteristics of a real estate property, where: prudence is indicated as "yes"; ownership of the collateral is legal and undisputed as "no"; the collateral is not subject to seizure, freezing, or detention as "no"; registration of the collateral is indicated as "yes"; enforceability of the collateral is indicated as "yes"; priority is first priority; valuation of the collateral is prudent as "yes"; and positive correlation is indicated as "no". The electronic device uses these object attribute characteristics as input feature vectors and feeds them into a second-parameter model. The second-parameter model can output a second value parameter of 8.5 million yuan for the real estate.

[0147] It's important to note that the second-parameter model can handle features containing missing values ​​when processing object attribute features. During model training, the second-parameter model learns how to handle missing values. Therefore, even if some object attribute features of the target object are missing, the second-parameter model can still output the second value parameter of the target object.

[0148] In some embodiments, after obtaining a second value parameter of the target object, the electronic device compares the second value parameter with the amount of risk exposure corresponding to the target object to determine whether the target object can mitigate the risk exposure. If the risk mitigation value indicated by the second value parameter is higher than the amount of risk exposure corresponding to the target object, it is determined that the target object can mitigate the risk exposure; if the risk mitigation value indicated by the second value parameter is lower than the amount of risk exposure corresponding to the target object, it is determined that the target object cannot mitigate the risk exposure.

[0149] In some embodiments, multiple target objects may be used simultaneously to jointly guarantee the same risk exposure. In this case, the electronic device needs to obtain the object attribute characteristics of each target object separately and obtain the second value parameter of each target object through the second parameter model.

[0150] Specifically, for a given risk exposure, the electronic device can acquire all target objects associated with that risk exposure. The electronic device can acquire the object attribute characteristics of each target object, and input these characteristics into a second parameter model to obtain the second value parameter of that target object. The electronic device then aggregates the second value parameters of all target objects to obtain the total risk mitigation value corresponding to that risk exposure.

[0151] In some embodiments, the electronic device can verify the second value parameter output by the second parameter model. Specifically, the electronic device acquires the object attribute features and original value data of the target object, processes the object attribute features of the target object through the second parameter model, and obtains the second value parameter of the target object. Simultaneously, the electronic device calculates the first value parameter of the target object according to data operation rules related to financial regulatory rules. The electronic device compares the second value parameter with the first value parameter. If the difference between the second value parameter and the first value parameter is less than a preset difference threshold, the electronic device can determine that the second value parameter output by the second parameter model is accurate. If the difference between the second value parameter and the first value parameter is greater than or equal to the preset difference threshold, the electronic device can generate an anomaly alert, prompting relevant personnel to check or retrain the second parameter model.

[0152] Through the training and application methods of the parameter model provided in this application, electronic devices can automatically predict the risk mitigation value of collateral, improve the accuracy and automation of risk mitigation value calculation, reduce the workload of manual verification, avoid the estimation deviation of mitigation capacity due to poor data quality, thereby improving the accuracy of capital measurement, meeting the requirements of financial regulatory rules, and avoiding regulatory risks and economic losses caused by non-compliance.

[0153] Please refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of a training device for a parametric model provided in an embodiment of this application. For example... Figure 4 As shown, the training device for the parameter model includes: a feature acquisition module 401, a value calculation module 402, and a model training module 403.

[0154] The feature acquisition module 401 is used to acquire the object attribute features of multiple sample objects; wherein, the multiple sample objects are collateral or guarantees used to mitigate risk exposure, and the object attribute features of the sample objects are features used to indicate whether the sample objects meet the corresponding financial regulatory rules.

[0155] The value calculation module 402 is used to calculate the first value parameter of the sample object according to the data operation rules related to financial regulatory rules; wherein, the first value parameter is used to indicate the risk mitigation value of the sample object, and the first value parameter serves as a model label for the object characteristics of the sample object.

[0156] The model training module 403 is used to train a first-parameter model by using the object attribute features and first value parameters of each sample object from multiple sample objects as training samples, thereby obtaining a trained second-parameter model. The first-parameter model is constructed and fused based on at least two decision tree algorithms, including random forest and gradient boosting tree. The second-parameter model has the ability to obtain a second value parameter of the target object based on its object attribute features. This second value parameter indicates the risk mitigation value of the target object under financial regulatory rules.

[0157] In other embodiments, the sample object includes collateral, and the object attribute features include one or more of the following: an identifier indicating whether the collateral is prudent, an identifier indicating whether the ownership of the collateral is legal and undisputed, an identifier indicating whether the collateral is subject to investigation, freezing or seizure, an identifier indicating whether the collateral has been registered, an identifier indicating whether the collateral is enforceable, priority, an identifier indicating whether the valuation of the collateral is prudent, and at least one of the following positive correlation identifiers.

[0158] In other embodiments, the data calculation rules include a conversion factor; the value calculation module 402 described above is also used to obtain the original value data and conversion factor of the sample object; and to determine the first value parameter of the sample object based on the original value data and conversion factor of the sample object.

[0159] In other embodiments, the conversion factor includes at least one of a first conversion factor corresponding to the type of the sample object and a second conversion factor calculated based on the validity period of the sample object and the termination period of the risk exposure. If the value unit of the sample object differs from the value unit of the risk exposure, the conversion factor also includes a preset third conversion factor. The aforementioned value calculation module 402 is further configured to determine a conversion ratio based on the conversion factor of the sample object; and multiply the original value data of the sample object by the conversion ratio to obtain a first value parameter of the sample object.

[0160] In other embodiments, the training apparatus for the above-described parameter model may further include a qualification determination module for determining the qualification of a sample object; wherein, the qualification of a sample object includes at least one of the following: the sample object is a first object type; the feature value of the first object attribute feature of the sample object is a first value; the first object attribute feature is one of the object attribute features of the sample object; and the validity period of the sample object is greater than the termination period of the risk exposure.

[0161] Please refer toFigure 5 , Figure 5 This is a schematic diagram of the structure of an application device for a parameter model provided in an embodiment of this application. For example... Figure 5 As shown, the application device of this parameter model includes: a feature acquisition module 501 and a model application module 502.

[0162] The feature acquisition module 501 is used to acquire the object attribute features of the target object; wherein, the target object is collateral or guarantee used to mitigate risk exposure, and the object attribute features of the target object are features used to indicate whether the target object meets the corresponding financial regulatory rules.

[0163] The model application module 502 is used to process the object features of the target object through the trained second parameter model as described in the above embodiment to obtain the second value parameter of the target object, which is used to indicate the risk mitigation value of the target object.

[0164] The parameter model training device and application device provided in this application embodiment can execute the method shown in the above method embodiment. The implementation principle and beneficial effects can be referred to the relevant description in the method embodiment, and will not be repeated here.

[0165] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device includes: a memory 601, a transceiver 602, and at least one processor 603.

[0166] The transceiver 602 is used to interact with other devices to send and receive data. For example, in this embodiment, the transceiver 602 can specifically be used to obtain object attribute characteristics.

[0167] The memory 601 stores computer program code, which includes computer instructions. These computer instructions run in the described electronic device to implement the method shown in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, portable hard drive, read-only memory, disk, or optical disc, etc.

[0168] Processor 603 can be a general-purpose processor, including a Central Processing Unit (CPU), a network processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 603 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0169] The memory 601, transceiver 602, and processor 603 are communicatively connected. For example, the memory 601 and transceiver 602 can be connected to the processor 603 via a system bus and communicate with each other. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus.

[0170] Optionally, the memory 601 can be either standalone or integrated with the processor 603. When the memory 601 is set up independently, it is connected to the processor 603 via a system bus.

[0171] This application also provides a chip for executing instructions, which is used to execute the technical solutions of the parameter model training method and application method in the above embodiments.

[0172] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solutions of the parameter model training method and application method in the above embodiments. Specifically, when the computer instructions are executed by a processor, the electronic device can execute the technical solutions of the parameter model training method and application method in the above embodiments.

[0173] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solutions of the parameter model training method and application method in the above embodiments.

[0174] The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage media can be any available medium accessible to a general-purpose or special-purpose computer.

[0175] An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0176] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0177] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0178] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0179] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0180] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0181] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for training a parametric model, characterized in that, include: Obtain object attribute features of multiple sample objects; wherein, the multiple sample objects are collateral or guarantees used to mitigate risk exposure, and the object attribute features of the sample objects are features used to indicate whether the sample objects meet the corresponding financial regulatory rules; The first value parameter of the sample object is calculated according to the data operation rules related to the financial regulatory rules; wherein, the first value parameter is used to indicate the risk mitigation value of the sample object, and the first value parameter serves as a model label for the object characteristics of the sample object; The object attribute features and first value parameters of each of the plurality of sample objects are used as training samples to train a first parameter model, thereby obtaining a trained second parameter model; wherein, the first parameter model is constructed and fused based on at least two decision tree algorithms, including random forest and gradient boosting tree; The second parameter model has the ability to obtain a second value parameter of the target object based on the object attribute characteristics of the target object. The second value parameter is used to indicate the risk mitigation value of the target object under the financial regulatory rules.

2. The method according to claim 1, characterized in that, The sample objects include collateral, and the object attribute characteristics include one or more of the following: The following indicators are required: whether the collateral is prudent, whether the ownership of the collateral is legal and undisputed, whether the collateral is subject to seizure or freezing, whether the collateral is registered, whether the collateral is enforceable, priority, whether the valuation of the collateral is prudent, and at least one of the following positive correlation indicators.

3. The method according to claim 1, characterized in that, The data processing rules include conversion factors; The calculation of the first value parameter of the sample object according to the data processing rules related to the financial regulatory rules includes: Obtain the original value data and conversion factor of the sample object; Based on the original value data and conversion factor of the sample object, the first value parameter of the sample object is determined.

4. The method according to claim 3, characterized in that, The conversion factor includes at least one of a first conversion factor corresponding to the type of the sample object and a second conversion factor calculated based on the validity period of the sample object and the termination period of the risk exposure. If the value unit of the sample object is different from the value unit of the risk exposure, the conversion factor also includes a preset third conversion factor. The step of determining the first value parameter of the sample object based on its original value data and conversion factor includes: The conversion ratio is determined based on the conversion factor of the sample objects; The original value data of the sample object is multiplied by the conversion ratio to obtain the first value parameter of the sample object.

5. The method according to any one of claims 1-4, characterized in that, Before calculating the first value parameter of the sample object according to the data processing rules related to the financial regulatory rules, the method further includes: The sample object was determined to be qualified; The qualified sample objects include at least one of the following: The sample object is of the first object type; The feature value of the first object attribute feature of the sample object is a first value; the first object attribute feature is one of the object attribute features of the sample object; The validity period of the sample objects is longer than the termination period of the risk exposure.

6. A method for applying a parametric model, characterized in that, include: Obtain the object attribute characteristics of the target object; wherein, the target object is collateral or guarantee used to mitigate risk exposure, and the object attribute characteristics of the target object are characteristics used to indicate whether the target object meets the corresponding financial regulatory rules; The object features of the target object are processed by the trained second parameter model as described in any one of claims 1-5 to obtain a second value parameter of the target object, the second value parameter being used to indicate the risk mitigation value of the target object.

7. A training device for a parametric model, characterized in that, include: The feature acquisition module is used to acquire object attribute features of multiple sample objects; wherein, the multiple sample objects are collateral or guarantees used to mitigate risk exposure, and the object attribute features of the sample objects are features used to indicate whether the sample objects meet the corresponding financial regulatory rules. The value calculation module is used to calculate a first value parameter of the sample object according to the data operation rules related to the financial regulatory rules; wherein, the first value parameter is used to indicate the risk mitigation value of the sample object, and the first value parameter serves as a model label for the object characteristics of the sample object; The model training module is used to train a first parameter model by taking the object attribute features and first value parameters of each of the plurality of sample objects as training samples, and to obtain a trained second parameter model; wherein the first parameter model is constructed and fused based on at least two decision tree algorithms, and the at least two decision tree algorithms include random forest and gradient boosting tree. The second parameter model has the ability to obtain a second value parameter of the target object based on the object attribute characteristics of the target object. The second value parameter is used to indicate the risk mitigation value of the target object under the financial regulatory rules.

8. An application device for a parametric model, characterized in that, include: The feature acquisition module is used to acquire the object attribute features of the target object; wherein, the target object is collateral or guarantee used to mitigate risk exposure, and the object attribute features of the target object are features used to indicate whether the target object meets the corresponding financial regulatory rules; The model application module is used to process the object features of the target object through the trained second parameter model as described in claim 7 to obtain a second value parameter of the target object, wherein the second value parameter is used to indicate the risk mitigation value of the target object.

9. An electronic device, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the electronic device performs the method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

11. A computer program product, characterized in that, When the computer program product is run on a computer / executed by the computer's processor, it implements the method as described in any one of claims 1-6.

Images