A financial institution joint credit risk control method and system

Abstract

This invention relates to the field of financial risk control technology, specifically disclosing a joint credit risk control method and system for financial institutions. The method includes: classifying standardized feature data by sensitivity to obtain high and low sensitivity feature sets; obtaining first and second encrypted data using fully homomorphic and partially homomorphic encryption, respectively; removing unselected feature dimensions to obtain a pruned encrypted feature set; jointly applying a homomorphic perturbation factor to the encrypted feature set with financial institutions and risk control nodes to generate encrypted verification feature values; performing consistency verification to obtain verification results; iteratively correcting the results based on risk weights and credit ratios, and synchronizing the updated values ​​to financial institutions. This invention can achieve secure, efficient, and adaptively optimized joint credit risk control.

Description

Technical Field

[0001] This invention relates to the field of financial risk control technology, and in particular to a joint credit risk control method and system for financial institutions. Background Technology

[0002] In collaborative credit risk control scenarios among financial institutions, multiple institutions need to share customer credit data to assess the risk of multiple borrowing and joint credit limits. Traditional solutions typically aggregate the raw customer feature data from each institution to a risk control platform, and then perform unified credit scoring and credit decisions. However, customer credit data contains highly sensitive fields such as identity information and financial statements, and directly aggregating plaintext data poses serious privacy and compliance risks. To address this, existing technologies attempt to introduce cryptographic techniques such as homomorphic encryption and secure multi-party computation to allow institutions to complete joint computations in encrypted form. However, existing methods often apply the same encryption strength to all features, ignoring the sensitivity differences of different feature dimensions. This results in highly sensitive features being treated the same as low-sensitivity features, leading to high encryption computation overhead and insufficient flexibility.

[0003] In existing technologies, joint credit risk control methods lack mechanisms for differentiated encryption of customer characteristics. This makes it difficult to reduce the computational burden of encryption for low-sensitivity data while ensuring the highest security level for highly sensitive data, resulting in low overall ciphertext processing efficiency and limited data availability. Furthermore, existing methods lack verification mechanisms for data integrity and source authenticity during ciphertext transmission, making them vulnerable to attacks that maliciously tamper with or forge encrypted feature values, thus compromising the credibility of joint risk control results. In addition, existing credit parameter adjustment strategies largely rely on static rules or offline model updates, making it difficult to dynamically adjust risk weights and credit ratios based on the actual repayment performance of each institution. This leads to poor adaptability of the joint risk control system, causing credit decisions to gradually deviate from the actual risk level. Therefore, there is an urgent need to develop a joint credit risk control method for financial institutions that can achieve feature-level encryption, ciphertext perturbation verification, and closed-loop iterative correction to address the aforementioned deficiencies in security, efficiency, and adaptability, thereby improving the privacy protection level, computational efficiency, and decision accuracy of joint credit risk control. Summary of the Invention

[0004] This invention provides a joint credit risk control method and system for financial institutions to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a joint credit risk control method for financial institutions, comprising: S1: Divide the standardized feature data of customer credit in financial institutions into sensitivity levels to obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data; S2: The highly sensitive feature set is encrypted using a fully homomorphic encryption scheme to obtain the first ciphertext data, and the low-sensitivity feature set is encrypted using a partially homomorphic encryption scheme to obtain the second ciphertext data; S3: Remove feature dimensions from the first ciphertext data and the second ciphertext data to obtain the pruned encrypted feature set of the standardized feature data; S4: Based on financial institutions and joint risk control nodes, a homomorphic perturbation factor is added to the cropped encrypted feature set to obtain the encrypted verification feature value of the cropped encrypted feature set; S5: Perform a consistency check on the encrypted verification feature value to obtain the consistency check result; S6: Iteratively correct the risk weights and credit limit allocation ratios in the consistency verification results to obtain updated risk weights and credit limit allocation ratios, and synchronize them to the financial institution.

[0006] In a preferred embodiment, the step of classifying the standardized feature data of customer credit in financial institutions into sensitivity levels to obtain a high-sensitivity feature set and a low-sensitivity feature set of the standardized feature data includes: The business semantic labels of the feature fields in the standardized feature data are parsed and compared with the sensitivity dictionary of the financial institution to obtain the basic sensitivity level of the feature fields; Based on the historical access frequency of the feature field in the financial institution, the basic sensitivity level is adaptively adjusted to obtain the comprehensive sensitivity level of the feature field; The comprehensive sensitivity level is compared horizontally with the level threshold of the sensitivity dictionary to obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data.

[0007] In a preferred embodiment, the step of applying a fully homomorphic encryption scheme to the high-sensitivity feature set to obtain the first ciphertext data, and applying a partially homomorphic encryption scheme to the low-sensitivity feature set to obtain the second ciphertext data, includes: The feature values ​​in the highly sensitive feature set are encoded and converted to obtain the plaintext polynomial of the feature values; Based on the fully homomorphic encryption public key of the financial institution, the plaintext polynomial is encrypted and weighted to obtain the first ciphertext component of the feature value; The feature values ​​in the highly sensitive feature set are traversed, and all first ciphertext components are aggregated to obtain the first ciphertext data; Based on the partial homomorphic encryption public key of the financial institution, ciphertext deduction is performed on the feature values ​​in the low-sensitivity feature set to obtain the second ciphertext component of the feature values. The feature values ​​in the low-sensitivity feature set are traversed, and all second ciphertext components are aggregated to obtain the second ciphertext data.

[0008] In a preferred embodiment, the step of removing feature dimensions from the first ciphertext data and the second ciphertext data to obtain the pruned encrypted feature set of the standardized feature data includes: A feature mask vector for the feature requirement list of the joint risk control task in the financial institution is obtained by compiling and setting the feature requirement list. Based on the feature mask vector, the first ciphertext data and the second ciphertext data are subjected to identifier removal to obtain the retained identifier of the feature mask vector; Based on the reserved identifier, the first ciphertext component and the feature mask vector are logically synchronized bit by bit to obtain the first trimmed ciphertext component of the first ciphertext data. Based on the reserved identifier, the second ciphertext component and the feature mask vector are logically synchronized bit by bit to obtain the second trimmed ciphertext component of the second ciphertext data. The first and second cropped ciphertext components are sequentially merged to obtain the cropped encrypted feature set of the standardized feature data.

[0009] In a preferred embodiment, the step of adding a homomorphic perturbation factor to the pruned encrypted feature set based on financial institutions and joint risk control nodes to obtain the encrypted verification feature value of the pruned encrypted feature set includes: Trusted nodes of the aforementioned financial institutions are jointly elected and deployed to obtain a joint risk control node; Distributed key generation is performed on the financial institution and the joint risk control node to obtain a shared random seed for the financial institution and the joint risk control node; Based on the shared random seed, the encrypted feature values ​​of the cropped encrypted feature set are mapped to obtain the homomorphic perturbation factor corresponding to the encrypted feature value; Homomorphic addition is performed on the encrypted feature values, and the homomorphic perturbation factor is superimposed on the corresponding encrypted feature values ​​to obtain the perturbation encrypted feature components of the encrypted feature set; The hash checksum of the shared random seed is appended to the perturbation encryption feature component to obtain the encryption checksum value of the encryption feature set, and then uploaded to the joint risk control node.

[0010] In a preferred embodiment, based on the shared random seed, the financial institution maps the encrypted feature values ​​of the pruned encrypted feature set to obtain a homomorphic perturbation factor corresponding to the encrypted feature values, including: The encrypted feature values ​​of the cropped encrypted feature set are allocated according to the storage order to obtain the sequential number of the encrypted feature values; The binary value of the shared random seed is concatenated with the sequence number to obtain a concatenated byte string, and the concatenated byte string is weighted and integrated to obtain the hash output value of the concatenated byte string; Convert the low-order byte segment of the hash output value into an integer to obtain the initial integer offset of the encryption feature value; The initial integer offset is moduloed by the upper limit of the perturbation value to obtain the homomorphic perturbation factor corresponding to the encrypted feature value.

[0011] In a preferred embodiment, performing a homomorphic addition operation on the encrypted feature value and superimposing the homomorphic perturbation factor onto the corresponding encrypted feature value to obtain the perturbed encrypted feature component of the encrypted feature set includes: The calculation formula for the perturbation encryption feature component is as follows: ; in, For the first encrypted feature set after cropping One encrypted feature value; For the first The first auxiliary factor for each feature dimension; To and The corresponding homomorphic perturbation factor; For the first The second auxiliary factor for each feature dimension; The first weighted disturbance is the result of superimposed weighted disturbance. One encrypted feature component; The encrypted feature values ​​in the cropped encrypted feature set are traversed, and the calculation of the perturbation encrypted feature components is repeatedly performed to obtain the set of perturbation encrypted feature components corresponding to the encrypted feature values.

[0012] In a preferred embodiment, the step of performing a consistency check on the encryption verification feature value to obtain a consistency check result includes: The same version of distributed key generation is performed on the joint risk control node and the financial institution to obtain an independent shared random seed for the financial institution. Based on the financial institution's independent shared random seed, the components in the perturbation encryption feature component set are inversely perturbed to obtain the original encryption feature components of the encryption feature set, and then re-weighted and integrated to obtain the local hash check code of the shared random seed. The local hash check code is compared one by one with the hash check code of the shared random seed to obtain the consistency check result.

[0013] In a preferred embodiment, the iterative correction of the risk weights and credit limit allocation ratios in the consistency verification results to obtain updated risk weights and credit limit allocation ratios, and the synchronization with the financial institution, includes: The joint risk scoring ciphertext in the financial institution is compared with the scoring threshold to obtain the adjustment coefficient corresponding to the joint risk scoring ciphertext, which is used as the correction coefficient of the financial institution. The correction coefficient is superimposed and bound to the original risk weight of the financial institution to obtain the updated risk weight; The correction coefficient is superimposed and bound to the original credit limit allocation ratio of the financial institution to obtain the updated credit limit allocation ratio. Based on the public key of the financial institution, the updated risk weight and the updated credit limit allocation ratio are encrypted to obtain the encrypted synchronization data packet of the financial institution and synchronized to the financial institution for the next joint risk control decision.

[0014] To address the aforementioned problems, the present invention also provides a joint credit risk control system for financial institutions, the system comprising: The data classification module is used to classify the standardized feature data of customer credit in financial institutions into sensitivity levels, and obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data. An encryption module is used to obtain first ciphertext data by applying a fully homomorphic encryption scheme to the high-sensitivity feature set, and to obtain second ciphertext data by applying a partially homomorphic encryption scheme to the low-sensitivity feature set; The feature trimming module is used to remove feature dimensions from the first ciphertext data and the second ciphertext data to obtain the trimmed encrypted feature set of the standardized feature data; The perturbation addition module is used to add a homomorphic perturbation factor to the cropped encrypted feature set based on financial institutions and joint risk control nodes, so as to obtain the encrypted verification feature value of the cropped encrypted feature set. The consistency verification module is used to perform consistency verification on the encrypted verification feature value and obtain the consistency verification result. The iterative correction module is used to iteratively correct the risk weights and credit limit allocation ratios in the consistency verification results, obtain updated risk weights and credit limit allocation ratios, and synchronize them to the financial institution.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention categorizes standardized customer credit feature data into sensitivity levels, obtaining high-sensitivity and low-sensitivity feature sets. These sets are then encrypted using fully homomorphic and partially homomorphic encryption respectively, achieving differentiated hierarchical encryption: fully homomorphic encryption of high-sensitivity features ensures privacy and security, while partially homomorphic encryption of low-sensitivity features reduces computational overhead. Simultaneously, redundant feature dimensions are eliminated based on feature mask vectors, reducing ciphertext transmission and storage, avoiding interference from irrelevant features, and improving data utilization accuracy.

[0016] 2. This invention introduces a homomorphic perturbation factor addition and consistency verification mechanism between financial institutions and joint risk control nodes. A shared random seed is generated using a distributed key to produce perturbation factors that correspond one-to-one with the encrypted feature value. These perturbation factors are then superimposed to form the encrypted verification feature value. Any tampering will cause the hash verification to fail, ensuring the reliability of the result. Furthermore, risk weights and credit limit allocation ratios are iteratively adjusted based on the consistency verification results. Parameters are dynamically adjusted in conjunction with the encrypted scoring of each institution and actual repayment performance, achieving closed-loop feedback and automatic optimization. This improves the system's dynamic adaptability and decision-making accuracy, ultimately achieving secure, efficient, and sustainably optimized joint credit management. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a joint credit risk control method for financial institutions, provided as an embodiment of the present invention. Figure 2 A functional module diagram of a joint credit risk control system for financial institutions provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a joint credit risk control method for financial institutions. The executing entity of this joint credit risk control method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the joint credit risk control method for financial institutions can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a joint credit risk control method for financial institutions according to an embodiment of the present invention. In this embodiment, the joint credit risk control method for financial institutions includes: S1: The standardized feature data of customer credit in financial institutions is classified into sensitivity levels to obtain a high-sensitivity feature set and a low-sensitivity feature set, including: In this embodiment of the invention, the business semantic labels of the feature fields in the standardized feature data are parsed and compared with the sensitivity dictionary of the financial institution to obtain the basic sensitivity level of the feature fields; Based on the historical access frequency of the feature field in the financial institution, the basic sensitivity level is adaptively adjusted to obtain the comprehensive sensitivity level of the feature field; The comprehensive sensitivity level is compared horizontally with the level threshold of the sensitivity dictionary to obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data.

[0021] The business semantic labels carried by each feature field in the standardized feature data are extracted. Financial institutions pre-store a sensitivity dictionary, which records the basic sensitivity levels corresponding to different business semantic labels. The financial institution parses and compares the business semantic labels of the feature fields with the entries in the sensitivity dictionary, matching each entry one by one. For each matched feature field, its corresponding basic sensitivity level is assigned to that field, ultimately obtaining the basic sensitivity levels for all feature fields.

[0022] Based on the historical access frequency of each feature field within the financial institution, which records the number of times the feature field has been queried or used over a past period, the financial institution adaptively adjusts the already obtained basic sensitivity level according to this historical access frequency. Specifically, for feature fields with high historical access frequency, their basic sensitivity level is adjusted upward by one level; for feature fields with low historical access frequency, their basic sensitivity level is adjusted downward by one level, thereby obtaining the comprehensive sensitivity level for each feature field.

[0023] The overall sensitivity level of each feature field is compared horizontally with a preset sensitivity threshold in the sensitivity dictionary. The sensitivity threshold is a dividing line; feature fields with an overall sensitivity level higher than the threshold are classified into the high-sensitivity feature set, and feature fields with an overall sensitivity level lower than or equal to the threshold are classified into the low-sensitivity feature set, thus obtaining the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data.

[0024] The beneficial effects are as follows: This invention achieves accurate differentiation between highly sensitive and low-sensitive features by classifying standardized customer credit feature data into sensitivity levels and adaptively adjusting the comprehensive sensitivity level based on historical access frequency. This hierarchical mechanism allows subsequent encryption operations to adopt differentiated encryption strategies for different sensitivity levels, ensuring the security of highly sensitive data while avoiding resource waste caused by over-encryption of low-sensitive data. Simultaneously, the feature set classification results obtained by horizontally comparing level thresholds provide a clear basis for the selective removal of subsequent feature dimensions, effectively improving the targeting and overall efficiency of encrypted processing in the joint risk control process. Furthermore, this classification process relies entirely on the financial institution's existing dictionary and access logs, without depending on external models or algorithms, making it simple and reliable to implement and helping to reduce system complexity and operational costs.

[0025] S2: Apply a fully homomorphic encryption scheme to the high-sensitivity feature set to obtain the first ciphertext data, and apply a partially homomorphic encryption scheme to the low-sensitivity feature set to obtain the second ciphertext data, including: In this embodiment of the invention, the feature values ​​in the high-sensitivity feature set are encoded and converted to obtain the plaintext polynomial of the feature values; Based on the fully homomorphic encryption public key of the financial institution, the plaintext polynomial is encrypted and weighted to obtain the first ciphertext component of the feature value; The feature values ​​in the highly sensitive feature set are traversed, and all first ciphertext components are aggregated to obtain the first ciphertext data; Based on the partial homomorphic encryption public key of the financial institution, ciphertext deduction is performed on the feature values ​​in the low-sensitivity feature set to obtain the second ciphertext component of the feature values. The feature values ​​in the low-sensitivity feature set are traversed, and all second ciphertext components are aggregated to obtain the second ciphertext data.

[0026] The financial institution extracts each feature value from the highly sensitive feature set and converts the value into a plaintext polynomial according to a preset encoding rule. The coefficients of this plaintext polynomial are filled with the binary representation of the feature value. The financial institution loads a pre-stored fully homomorphic encryption public key and uses it to perform an encrypted weighted integration operation on the plaintext polynomial. Specifically, it performs a polynomial multiplication between the plaintext polynomial and the polynomial in the public key and adds noise to generate the first ciphertext component corresponding to the feature value. The financial institution iterates through all feature values ​​in the highly sensitive feature set, repeating the encoding conversion and encrypted weighted integration operation for each feature value to obtain the first ciphertext component for each feature value. These first ciphertext components are then aggregated in the original order of the feature values ​​to form the first ciphertext data.

[0027] The financial institution extracts each feature value from the low-sensitivity feature set and directly converts the value into a plaintext integer. It then loads a pre-stored partial homomorphic encryption public key and uses it to perform a ciphertext derivation operation on the plaintext integer. Specifically, it performs a modular exponentiation operation between the plaintext integer and the modulus in the public key to generate the second ciphertext component corresponding to that feature value. The financial institution iterates through all feature values ​​in the low-sensitivity feature set, repeating the encoding conversion and ciphertext derivation operation for each feature value to obtain the second ciphertext component for each feature value. These second ciphertext components are then aggregated according to the original order of the feature values ​​to form the second ciphertext data.

[0028] The beneficial effects are as follows: This invention achieves differentiated encryption protection by using fully homomorphic encryption to obtain the first ciphertext data for the highly sensitive feature set and partially homomorphic encryption to obtain the second ciphertext data for the low-sensitivity feature set. Fully homomorphic encryption allows highly sensitive feature values ​​to support arbitrary addition, subtraction, multiplication, and division operations in the ciphertext state, fully protecting the privacy and security of core data such as identity information and financial records. Partially homomorphic encryption only supports addition operations, but its encryption and decryption speeds are much faster than fully homomorphic encryption. Applying it to the low-sensitivity feature set significantly reduces the overall encryption computation time and storage overhead. The combined use of the two encryption schemes neither sacrifices the security of highly sensitive data nor wastes resources by over-encrypting low-sensitivity data, improving the efficiency of ciphertext processing in the joint risk control process. Simultaneously, the traversal aggregation operation ensures the integrity and sequential consistency of the first and second ciphertext data, providing a reliable data foundation for subsequent feature dimension elimination and joint scoring.

[0029] S3: Perform feature dimension removal on the first ciphertext data and the second ciphertext data to obtain the pruned encrypted feature set of the standardized feature data, including: In this embodiment of the invention, a feature requirement list for joint risk control tasks in the financial institution is compiled and set to obtain the feature mask vector of the feature requirement list. Based on the feature mask vector, the first ciphertext data and the second ciphertext data are subjected to identifier removal to obtain the retained identifier of the feature mask vector; Based on the reserved identifier, the first ciphertext component and the feature mask vector are logically synchronized bit by bit to obtain the first trimmed ciphertext component of the first ciphertext data. Based on the reserved identifier, the second ciphertext component and the feature mask vector are logically synchronized bit by bit to obtain the second trimmed ciphertext component of the second ciphertext data. The first and second cropped ciphertext components are sequentially merged to obtain the cropped encrypted feature set of the standardized feature data.

[0030] Financial institutions sorted out the feature dimensions required for this joint risk control task and compiled them into a feature requirement list. The feature requirement list was then compiled and set by assigning a binary bit identifier to each feature dimension in the list. All binary bits were arranged in the order of the feature dimensions to form a vector, which is called the feature mask vector. The bit with the first binary value indicates that the feature dimension should be retained, and the bit with the second binary value indicates that the feature dimension should be removed.

[0031] Based on the feature mask vector, financial institutions perform an identifier removal operation on the first and second ciphertext data. Specifically, they iterate through each binary bit in the feature mask vector. When the binary bit is the first value, the index of the feature dimension corresponding to that binary bit is recorded as a reserved identifier. When the binary bit is the second value, no record is made. The set of all reserved identifiers obtained after the iteration is the reserved identifier of the feature mask vector.

[0032] Financial institutions perform bit-by-bit logical synchronous verification on each first ciphertext component in the first ciphertext data based on the retention identifier. Specifically, for each first ciphertext component, the original index of its associated feature dimension is extracted, and the existence of the index is checked in the retention identifier set. If it exists, the first ciphertext component is marked as retained; if it does not exist, it is marked as removed. All first ciphertext components marked as retained constitute the first pruned ciphertext components with retained feature dimensions in the first ciphertext data.

[0033] Financial institutions perform bit-by-bit logical synchronous verification on each second ciphertext component in the second ciphertext data based on the retention identifier. Specifically, for each second ciphertext component, the original index of its associated feature dimension is extracted, and the existence of the index is checked in the retention identifier set. If it exists, the second ciphertext component is marked as retained; if it does not exist, it is marked as removed. All second ciphertext components marked as retained constitute the second pruned ciphertext components with retained feature dimensions in the second ciphertext data.

[0034] Financial institutions arrange and merge the first and second cropped ciphertext components according to their original feature dimensions in the standardized feature data, and then concatenate the first and second cropped ciphertext components into a whole encrypted feature set, which is the cropped encrypted feature set of the standardized feature data.

[0035] The beneficial effects are as follows: This invention obtains a feature mask vector by compiling and setting the feature requirement list for joint risk control tasks, thus achieving clear identification of the required feature dimensions. Based on the feature mask vector, the first and second ciphertext data are identified and removed, retaining only the feature dimensions actually needed for the joint risk control task. This effectively eliminates redundant ciphertext data and reduces the amount of data transmitted and calculated subsequently. Through bit-by-bit logical synchronous verification, it is ensured that each ciphertext component strictly matches its feature dimension index, avoiding errors in risk control results caused by dimension misalignment. The first and second trimmed ciphertext components are sequentially merged to form a unified trimmed encrypted feature set, enabling seamless integration of data from different encryption schemes. This provides a clear and concise data foundation for adding homomorphic perturbation factors and consistency checks, improving the overall efficiency of the joint risk control process and the accuracy of data processing.

[0036] S4: Based on financial institutions and joint risk control nodes, a homomorphic perturbation factor is added to the pruned encrypted feature set to obtain the encrypted verification feature value of the pruned encrypted feature set, including: In this embodiment of the invention, trusted nodes of the financial institution are jointly elected and deployed to obtain a joint risk control node; Distributed key generation is performed on the financial institution and the joint risk control node to obtain a shared random seed for the financial institution and the joint risk control node; Based on the shared random seed, the encrypted feature values ​​of the cropped encrypted feature set are mapped to obtain the homomorphic perturbation factor corresponding to the encrypted feature value; Homomorphic addition is performed on the encrypted feature values, and the homomorphic perturbation factor is superimposed on the corresponding encrypted feature values ​​to obtain the perturbation encrypted feature components of the encrypted feature set; The hash checksum of the shared random seed is appended to the perturbation encryption feature component to obtain the encryption checksum value of the encryption feature set, and then uploaded to the joint risk control node.

[0037] The encrypted feature values ​​of the cropped encrypted feature set are allocated according to the storage order to obtain the sequential number of the encrypted feature values; The binary value of the shared random seed is concatenated with the sequence number to obtain a concatenated byte string, and the concatenated byte string is weighted and integrated to obtain the hash output value of the concatenated byte string; Convert the low-order byte segment of the hash output value into an integer to obtain the initial integer offset of the encryption feature value; The initial integer offset is moduloed by the upper limit of the perturbation value to obtain the homomorphic perturbation factor corresponding to the encrypted feature value.

[0038] The calculation formula for the perturbation encryption feature component is as follows: ; in, For the first encrypted feature set after cropping One encrypted feature value; For the first The first auxiliary factor for each feature dimension; To and The corresponding homomorphic perturbation factor; For the first The second auxiliary factor for each feature dimension; The first weighted disturbance is the result of superimposed weighted disturbance. One encrypted feature component; The encrypted feature values ​​in the cropped encrypted feature set are traversed, and the calculation of the perturbation encrypted feature components is repeatedly performed to obtain the set of perturbation encrypted feature components corresponding to the encrypted feature values.

[0039] In the implementation environment of joint credit risk control by financial institutions, the trusted nodes of each financial institution participating in the joint credit risk control business are jointly voted and configured according to the preset trust judgment rules, forming a joint risk control node for unified execution of joint risk control tasks.

[0040] The financial institution and the joint risk control node that has completed the election deployment carry out key collaborative generation operations in accordance with the distributed key generation execution method, so that the financial institution and the joint risk control node jointly generate and hold a shared random seed exclusive to both parties.

[0041] Based on the data correspondence, the shared random seed is used as the mapping basis. Numerical mapping transformation is performed on each encrypted feature value in the cropped encrypted feature set to generate a homomorphic perturbation factor corresponding to each encrypted feature value.

[0042] According to the execution rules of homomorphic addition, homomorphic addition is performed on each encrypted feature value in the clipped encrypted feature set. The generated homomorphic perturbation factor is superimposed on each encrypted feature value after homomorphic addition according to the numerical correspondence, forming the perturbation encrypted feature component of the encrypted feature set.

[0043] The hash check code obtained by hashing the shared random seed is appended to the corresponding position of each perturbation encryption feature component to form the encryption check feature value of the encryption feature set, and all encryption check feature values ​​are transmitted to the joint risk control node for storage.

[0044] According to the predetermined storage order of the encrypted feature values ​​within the clipped encrypted feature set, a sequential allocation operation is performed on each encrypted feature value to generate a corresponding sequential number for each encrypted feature value.

[0045] The binary value obtained by converting the shared random seed is concatenated with the sequential number corresponding to each encrypted feature value according to their positional relationship to form a concatenated byte string. The data of each byte in the concatenated byte string is then weighted and integrated to generate the hash output value of the concatenated byte string.

[0046] The low-order byte segment inside the hash output value is converted into integer form according to the numerical conversion rules, forming the initial integer offset corresponding to each encryption feature value.

[0047] The initial integer offset and the preset perturbation upper limit are subjected to modulo operation according to the modulo rule to generate a homomorphic perturbation factor corresponding to each encryption feature value.

[0048] Financial institutions obtain a preset perturbation upper limit value, divide the initial integer offset by the perturbation upper limit value, and take the remainder. This remainder is the homomorphic perturbation factor that uniquely corresponds to the current encrypted feature value.

[0049] For the first parameter in the formula , representing the first in the set of encrypted features after cropping. The encrypted feature value is derived from the ciphertext data obtained after sensitivity grading, differential encryption, and feature dimension removal in the previous steps. It is the first encrypted feature value. The numerical representation of each feature dimension in encrypted form, without any added perturbation information.

[0050] parameter Indicates the first The first auxiliary factor of the first feature dimension, which is determined by the financial institution based on the first feature dimension. Each feature dimension has a preset weight coefficient for its data type, which is used to adjust the influence of the homomorphic perturbation factor on the encrypted feature value. Different feature dimensions can be set with different first auxiliary factors.

[0051] parameter Indicates and The corresponding homomorphic perturbation factor, which is generated by hash mapping and modulo operation from a shared random seed and sequential number, is an integer offset used to apply randomization perturbation to the encrypted feature value.

[0052] parameter Indicates the first The second auxiliary factor of the first feature dimension, which is determined by the financial institution based on the first feature dimension. The data distribution range of each feature dimension is preset with a constant offset, which is used to add a fixed baseline displacement during the disturbance process.

[0053] parameter Indicates the th after superimposed weighted perturbation Each encrypted feature component is a component of the original encrypted feature value. With homomorphic perturbation factor Multiply by the first auxiliary factor The product plus the second auxiliary factor The final result is the encrypted data unit that is uploaded to the joint risk control node.

[0054] For the calculation steps, the financial institution first extracts the first [feature] from the pruned encrypted feature set. Each encryption feature value Financial institutions obtain the first auxiliary factor corresponding to this feature dimension. and will With homomorphic perturbation factor Performing a multiplication operation yields a weighted disturbance term. Financial institutions obtain the second auxiliary factor. encrypt the feature value Weighted disturbance term and second auxiliary factor The three are added together, that is... Add the weighted disturbance term and A new value is generated, which is the value of the weighted perturbation after superposition. Each encrypted feature component Financial institutions will The data is stored in the perturbation-encrypted feature component set, and the above operation is repeated for the next encrypted feature value. The financial institution iterates through all encrypted feature values ​​in the trimmed encrypted feature set, performing the same multiplication and addition operations on each encrypted feature value, and collects all generated data. Arranged in their original order, the complete set of perturbation encryption feature components is finally obtained.

[0055] The beneficial effects are as follows: This invention ensures the trustworthiness and decentralized nature of joint risk control nodes through joint election and deployment. Distributed key generation uses a shared random seed, ensuring that the generation of perturbation factors involves both parties and cannot be predicted unilaterally, thus improving the security of the perturbation. Generating unique homomorphic perturbation factors based on the shared random seed and sequential numbering for feature mapping provides differentiated noise across different feature dimensions, effectively resisting statistical attacks. Homomorphic addition of the perturbation factor forms a perturbation encryption feature component, enhancing the randomness of the ciphertext without destroying the homomorphic property. Adding a hash checksum generates an encrypted checksum feature value and uploads it, providing integrity evidence for subsequent consistency checks and detecting tampering during transmission, thereby improving the reliability and tamper-proof capability of joint risk control data transmission.

[0056] S5: Perform a consistency check on the encrypted verification feature value to obtain the consistency check result, including: In this embodiment of the invention, the joint risk control node and the financial institution are generated with the same version of distributed key to obtain an independent shared random seed for the financial institution. Based on the financial institution's independent shared random seed, the components in the perturbation encryption feature component set are inversely perturbed to obtain the original encryption feature components of the encryption feature set, and then re-weighted and integrated to obtain the local hash check code of the shared random seed. The local hash check code is compared one by one with the hash check code of the shared random seed to obtain the consistency check result.

[0057] The joint risk control node and the financial institution implement the same version of the distributed key generation protocol. Both parties generate random number fragments in the same way as before, and exchange the hash commitments of the fragments through a secure channel. The two parties combine and reconstruct the exchanged fragments to generate an independent shared random seed with the same value as the shared random seed stored locally by the financial institution. This seed is the financial institution's independent shared random seed.

[0058] The joint risk control node obtains the set of perturbation-encrypted feature components uploaded by financial institutions and extracts each perturbation-encrypted feature component. Based on the financial institution's independent shared random seed, the joint risk control node generates a homomorphic perturbation factor corresponding to each encrypted feature value according to the same mapping rules as the financial institution. The joint risk control node performs an inverse perturbation operation on each perturbation-encrypted feature component, specifically subtracting the product of the first auxiliary factor and the homomorphic perturbation factor from the perturbation-encrypted feature component, and then subtracting the second auxiliary factor to obtain the original encrypted feature component. The joint risk control node collects all recovered original encrypted feature components and arranges them in order to obtain the original encrypted feature components of the encrypted feature set. The joint risk control node inputs the financial institution's independent shared random seed into the same one-way hash mapping function as the financial institution, and re-weights and integrates them to obtain the local hash checksum of the seed.

[0059] The joint risk control node compares the local hash checksum with the hash checksum of the shared random seed parsed from the encrypted checksum value uploaded by the financial institution, bit by bit, to see if every binary bit of the two hash checksums is completely identical. If all binary bits are identical, a consistency check result is generated that passes; if any binary bit is different, a consistency check result is generated that fails.

[0060] The beneficial effects are as follows: This invention obtains an independent shared random seed by having the joint risk control node and financial institution perform the same version of distributed key generation. This ensures that both parties use the same seed parameters during the verification process, avoiding misjudgments caused by inconsistent seeds. Based on the independent shared random seed, the perturbed encrypted feature components are inversely perturbed, successfully recovering the original encrypted feature components and verifying the reversibility and correctness of the perturbation process. The reweighted integration yields a local hash checksum, which is then compared one by one with the received hash checksum, accurately detecting whether the ciphertext has been tampered with or corrupted during transmission. This consistency verification mechanism does not rely on third-party trust and fully utilizes cryptographic means to verify data integrity and source authenticity, effectively resisting the risks of man-in-the-middle attacks and malicious node data forgery, thus improving the security and reliability of the joint risk control system in untrusted network environments.

[0061] S6: Iteratively correct the risk weights and credit limit allocation ratios in the consistency verification results to obtain updated risk weights and credit limit allocation ratios, and synchronize them to the financial institution, including: In this embodiment of the invention, the joint risk scoring ciphertext in the financial institution is compared with the scoring threshold to obtain the adjustment coefficient corresponding to the joint risk scoring ciphertext, which is used as the correction coefficient of the financial institution. The correction coefficient is superimposed and bound to the original risk weight of the financial institution to obtain the updated risk weight; The correction coefficient is superimposed and bound to the original credit limit allocation ratio of the financial institution to obtain the updated credit limit allocation ratio. Based on the public key of the financial institution, the updated risk weight and the updated credit limit allocation ratio are encrypted to obtain the encrypted synchronization data packet of the financial institution and synchronized to the financial institution for the next joint risk control decision.

[0062] The joint risk control node obtains the encrypted joint risk score uploaded by financial institutions, and compares the encrypted joint risk score with multiple preset scoring threshold ranges one by one. Each scoring threshold range corresponds to an adjustment coefficient. When the encrypted joint risk score falls into a certain threshold range after decryption, the adjustment coefficient corresponding to that range is extracted and used as the correction coefficient for the financial institution.

[0063] The joint risk control node obtains the original risk weight currently used by the financial institution, and then performs an overlay and binding operation on the original risk weight and the correction coefficient. Specifically, the value of the original risk weight is added to the value of the correction coefficient to obtain a new value, which is the updated risk weight.

[0064] The joint risk control node obtains the original credit limit allocation ratio currently used by the financial institution, and then performs an overlay and binding operation on the original credit limit allocation ratio and the correction coefficient. Specifically, the value of the original credit limit allocation ratio is added to the value of the correction coefficient to obtain a new value, which is the updated credit limit allocation ratio.

[0065] The joint risk control node obtains the public key pre-uploaded by the financial institution. Using this public key, it performs encryption operations on the updated risk weights and updated credit limit allocation ratios, packaging the two encrypted data packets into a single data packet, which becomes the financial institution's encrypted synchronization data packet. The joint risk control node sends this encrypted synchronization data packet to the corresponding financial institution via the network. Upon receiving it, the financial institution decrypts it using its private key to obtain the updated risk weights and credit limit allocation ratios for use in the next joint risk control decision.

[0066] The beneficial effects are as follows: This invention quantifies the scoring results into specific correction coefficients by comparing the encrypted joint risk scoring data with scoring thresholds, achieving a precise mapping from risk level to parameter adjustment amount. This correction coefficient is then superimposed and bound to the original risk weight and original credit limit allocation ratio, allowing the risk weight and credit ratio to be dynamically adjusted according to actual risk control results, avoiding decision deviations caused by static parameters. The updated parameters are encrypted using the financial institution's public key, ensuring data confidentiality during the synchronization process and preventing parameters from being stolen or tampered with during transmission. This iterative correction mechanism forms a complete closed-loop feedback, enabling the joint risk control system to continuously optimize its parameters as business data changes, improving the responsiveness and adaptability of risk control strategies to real risk changes, and ensuring the long-term accuracy of joint credit decisions.

[0067] like Figure 2 The diagram shown is a functional block diagram of a joint credit risk control system for financial institutions provided in an embodiment of the present invention.

[0068] The financial institution joint credit risk control system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the financial institution joint credit risk control system 100 may include a data classification module 101, an encryption module 102, a feature trimming module 103, a disturbance addition module 104, a consistency verification module 105, and an iterative correction module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0069] In this embodiment, the functions of each module / unit are as follows: The data classification module 101 is used to classify the standardized feature data of customer credit in financial institutions into sensitivity levels, and obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data. The encryption module 102 is used to obtain first ciphertext data by applying a fully homomorphic encryption scheme to the high-sensitivity feature set and to obtain second ciphertext data by applying a partially homomorphic encryption scheme to the low-sensitivity feature set. The feature trimming module 103 is used to remove feature dimensions from the first ciphertext data and the second ciphertext data to obtain the trimmed encrypted feature set of the standardized feature data. The perturbation addition module 104 is used to add a homomorphic perturbation factor to the cropped encrypted feature set based on financial institutions and joint risk control nodes, so as to obtain the encryption verification feature value of the cropped encrypted feature set. The consistency verification module 105 is used to perform consistency verification on the encryption verification feature value and obtain the consistency verification result. The iterative correction module 106 is used to iteratively correct the risk weight and credit limit allocation ratio in the consistency verification result, obtain the updated risk weight and credit limit allocation ratio, and synchronize them to the financial institution.

[0070] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0071] 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 achieve the purpose of this embodiment according to actual needs.

[0072] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0073] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0074] The embodiments of this application can acquire and process relevant data based on an artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for joint credit risk control of financial institutions, characterized in that, The method includes: S1: Divide the standardized feature data of customer credit in financial institutions into sensitivity levels to obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data; S2: The highly sensitive feature set is encrypted using a fully homomorphic encryption scheme to obtain the first ciphertext data, and the low-sensitivity feature set is encrypted using a partially homomorphic encryption scheme to obtain the second ciphertext data; S3: Remove feature dimensions from the first ciphertext data and the second ciphertext data to obtain the pruned encrypted feature set of the standardized feature data; S4: Based on financial institutions and joint risk control nodes, a homomorphic perturbation factor is added to the cropped encrypted feature set to obtain the encrypted verification feature value of the cropped encrypted feature set; S5: Perform a consistency check on the encrypted verification feature value to obtain the consistency check result; S6: Iteratively correct the risk weights and credit limit allocation ratios in the consistency verification results to obtain updated risk weights and credit limit allocation ratios, and synchronize them to the financial institution.

2. The joint credit risk control method for financial institutions as described in claim 1, characterized in that, The process of classifying the standardized feature data of customer credit in financial institutions into sensitivity levels to obtain high-sensitivity feature sets and low-sensitivity feature sets includes: The business semantic labels of the feature fields in the standardized feature data are parsed and compared with the sensitivity dictionary of the financial institution to obtain the basic sensitivity level of the feature fields; Based on the historical access frequency of the feature field in the financial institution, the basic sensitivity level is adaptively adjusted to obtain the comprehensive sensitivity level of the feature field; The comprehensive sensitivity level is compared horizontally with the level threshold of the sensitivity dictionary to obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data.

3. The joint credit risk control method for financial institutions as described in claim 1, characterized in that, The process of obtaining first ciphertext data by applying a fully homomorphic encryption scheme to the high-sensitivity feature set and obtaining second ciphertext data by applying a partially homomorphic encryption scheme to the low-sensitivity feature set includes: The feature values ​​in the highly sensitive feature set are encoded and converted to obtain the plaintext polynomial of the feature values; Based on the fully homomorphic encryption public key of the financial institution, the plaintext polynomial is encrypted and weighted to obtain the first ciphertext component of the feature value; The feature values ​​in the highly sensitive feature set are traversed, and all first ciphertext components are aggregated to obtain the first ciphertext data; Based on the partial homomorphic encryption public key of the financial institution, ciphertext deduction is performed on the feature values ​​in the low-sensitivity feature set to obtain the second ciphertext component of the feature values. The feature values ​​in the low-sensitivity feature set are traversed, and all second ciphertext components are aggregated to obtain the second ciphertext data.

4. The joint credit risk control method for financial institutions as described in claim 3, characterized in that, The step of removing feature dimensions from the first ciphertext data and the second ciphertext data to obtain the pruned encrypted feature set of the standardized feature data includes: A feature mask vector for the feature requirement list of the joint risk control task in the financial institution is obtained by compiling and setting the feature requirement list. Based on the feature mask vector, the first ciphertext data and the second ciphertext data are subjected to identifier removal to obtain the retained identifier of the feature mask vector; Based on the reserved identifier, the first ciphertext component and the feature mask vector are logically synchronized bit by bit to obtain the first trimmed ciphertext component of the first ciphertext data. Based on the reserved identifier, the second ciphertext component and the feature mask vector are logically synchronized bit by bit to obtain the second trimmed ciphertext component of the second ciphertext data. The first and second cropped ciphertext components are sequentially merged to obtain the cropped encrypted feature set of the standardized feature data.

5. The joint credit risk control method for financial institutions as described in claim 1, characterized in that, The method of adding a homomorphic perturbation factor to the pruned encrypted feature set based on financial institutions and joint risk control nodes to obtain the encrypted verification feature value of the pruned encrypted feature set includes: Trusted nodes of the aforementioned financial institutions are jointly elected and deployed to obtain a joint risk control node; Distributed key generation is performed on the financial institution and the joint risk control node to obtain a shared random seed for the financial institution and the joint risk control node; Based on the shared random seed, the encrypted feature values ​​of the cropped encrypted feature set are mapped to obtain the homomorphic perturbation factor corresponding to the encrypted feature value; Homomorphic addition is performed on the encrypted feature values, and the homomorphic perturbation factor is superimposed on the corresponding encrypted feature values ​​to obtain the perturbation encrypted feature components of the encrypted feature set; The hash checksum of the shared random seed is appended to the perturbation encryption feature component to obtain the encryption checksum value of the encryption feature set, and then uploaded to the joint risk control node.

6. The joint credit risk control method for financial institutions as described in claim 5, characterized in that, Based on the shared random seed, the financial institution maps the encrypted feature values ​​of the pruned encrypted feature set to obtain a homomorphic perturbation factor corresponding to the encrypted feature values, including: The encrypted feature values ​​of the cropped encrypted feature set are allocated according to the storage order to obtain the sequential number of the encrypted feature values; The binary value of the shared random seed is concatenated with the sequence number to obtain a concatenated byte string, and the concatenated byte string is weighted and integrated to obtain the hash output value of the concatenated byte string; Convert the low-order byte segment of the hash output value into an integer to obtain the initial integer offset of the encryption feature value; The initial integer offset is moduloed by the upper limit of the perturbation value to obtain the homomorphic perturbation factor corresponding to the encrypted feature value.

7. The joint credit risk control method for financial institutions as described in claim 5, characterized in that, The step of performing a homomorphic addition operation on the encrypted feature value and superimposing the homomorphic perturbation factor onto the corresponding encrypted feature value to obtain the perturbation encrypted feature component of the encrypted feature set includes: The calculation formula for the perturbation encryption feature component is as follows: ； in, For the first encrypted feature set after cropping One encrypted feature value; For the first The first auxiliary factor for each feature dimension; To and The corresponding homomorphic perturbation factor; For the first The second auxiliary factor for each feature dimension; The first weighted disturbance is the result of superimposed weighted disturbance. One encrypted feature component; The encrypted feature values ​​in the cropped encrypted feature set are traversed, and the calculation of the perturbation encrypted feature components is repeatedly performed to obtain the set of perturbation encrypted feature components corresponding to the encrypted feature values.

8. The joint credit risk control method for financial institutions as described in claim 5, characterized in that, The process of performing a consistency check on the encrypted verification feature value to obtain a consistency check result includes: The same version of distributed key generation is performed on the joint risk control node and the financial institution to obtain an independent shared random seed for the financial institution. Based on the financial institution's independent shared random seed, the components in the perturbation encryption feature component set are inversely perturbed to obtain the original encryption feature components of the encryption feature set, and then re-weighted and integrated to obtain the local hash check code of the shared random seed. The local hash check code is compared one by one with the hash check code of the shared random seed to obtain the consistency check result.

9. The joint credit risk control method for financial institutions as described in claim 1, characterized in that, The iterative correction of the risk weights and credit limit allocation ratios in the consistency verification results to obtain updated risk weights and credit limit allocation ratios, and the synchronization with the financial institution, includes: The joint risk scoring ciphertext in the financial institution is compared with the scoring threshold to obtain the adjustment coefficient corresponding to the joint risk scoring ciphertext, which is used as the correction coefficient of the financial institution. The correction coefficient is superimposed and bound to the original risk weight of the financial institution to obtain the updated risk weight; The correction coefficient is superimposed and bound to the original credit limit allocation ratio of the financial institution to obtain the updated credit limit allocation ratio. Based on the public key of the financial institution, the updated risk weight and the updated credit limit allocation ratio are encrypted to obtain the encrypted synchronization data packet of the financial institution and synchronized to the financial institution for the next joint risk control decision.

10. A joint credit risk control system for financial institutions, characterized in that, The system for implementing the joint credit risk control method for financial institutions as described in claim 1 includes: The data classification module is used to classify the standardized feature data of customer credit in financial institutions into sensitivity levels, and obtain the high-sensitivity feature set and low-sensitivity feature set of the standardized feature data. An encryption module is used to obtain first ciphertext data by applying a fully homomorphic encryption scheme to the high-sensitivity feature set, and to obtain second ciphertext data by applying a partially homomorphic encryption scheme to the low-sensitivity feature set; The feature trimming module is used to remove feature dimensions from the first ciphertext data and the second ciphertext data to obtain the trimmed encrypted feature set of the standardized feature data; The perturbation addition module is used to add a homomorphic perturbation factor to the cropped encrypted feature set based on financial institutions and joint risk control nodes, so as to obtain the encrypted verification feature value of the cropped encrypted feature set. The consistency verification module is used to perform consistency verification on the encrypted verification feature value and obtain the consistency verification result. The iterative correction module is used to iteratively correct the risk weights and credit limit allocation ratios in the consistency verification results, obtain updated risk weights and credit limit allocation ratios, and synchronize them to the financial institution.

Images