Multi-party secure computing combined enterprise debtor cross-domain portrait feature extraction method and system

Abstract

The present application relates to the technical field of data processing, and more particularly to a method and system for extracting cross-domain portrait features of enterprise debtors in combination with multi-party secure computation, which comprises the following steps: obtaining enterprise debtor original data stored locally by participants; connecting each participant to a multi-party secure computation system through a secure communication channel; generating an encryption key pair based on threshold cryptography; distributing the public key to each participant and keeping the private key by the authorized result user; encrypting the original data of each participant using the public key to obtain encrypted enterprise debtor data; scoring the encrypted data according to importance; preferentially selecting data with a score higher than a preset threshold to perform multi-party secure computation; outputting the multi-dimensional ciphertext portrait feature vector formed by integrating the ciphertext calculation results output by the multi-party secure computation protocol to the authorized result user; and decrypting the plaintext portrait features through the private key.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a multi-party secure computing combined enterprise debtor cross-domain portrait feature extraction method and system. BACKGROUND

[0002] In the field of financial risk control and enterprise credit assessment, cross-institutional data collaborative analysis is the core means of accurately depicting enterprise debtor portraits. The fusion of multi-source heterogeneous data such as tax authorities, banks, power grids, and business registration authorities can effectively reveal the operating status and debt repayment ability of enterprises. However, due to data privacy protection regulations and data silos between institutions, traditional data collection modes are difficult to implement in compliance. Privacy computing technologies represented by multi-party secure computing (MPC) enable cross-domain feature joint extraction through data availability and invisibility mechanisms, becoming a key path to resolving the contradiction between data sharing and privacy protection.

[0003] However, as the dimensions of enterprise portraits continue to expand and the scale of participating institutions expands, enterprise portraits involve a large number of nonlinear operations, such as division in tax sales ratio calculation and logistic regression in risk scoring models. Homomorphic encryption relies on polynomial approximation or interactive protocol conversion for operations such as division and square root, and the computational complexity increases exponentially with the data size. For example, a pilot project by a certain bank showed that when processing tax sales ratio features for 100,000 enterprises, the single-feature calculation time delay increased from seconds to tens of minutes as the number of participants increased, far exceeding the real-time response threshold required by financial risk control scenarios.

[0004] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a multi-party secure computing combined enterprise debtor cross-domain portrait feature extraction method and system, aiming to solve the technical problem that privacy computing technologies such as homomorphic encryption have an exponential increase in computational complexity with data size when processing nonlinear complex operations such as division and logistic regression, resulting in excessive processing time for large-scale data and insufficient effectiveness in cross-institutional data collaborative analysis in the field of financial risk control and other fields.

[0006] To achieve the above purpose, the present application provides a multi-party secure computing combined enterprise debtor cross-domain portrait feature extraction method, which comprises:

[0007] Obtaining enterprise debtor original data stored in the local database of each participant, wherein the participants include commercial banks, tax authorities, and power grid companies, and the enterprise debtor original data includes enterprise basic information, financial account data, tax data, and energy consumption data;

[0008] The parties are accessed to the multi-party secure computing system through a secure communication channel, and when receiving the feature generation computing task instruction issued by the central dispatcher to the parties, the encryption key pair required by the multi-party secure computing protocol is generated based on threshold cryptography or a trusted third party, the public key is distributed to the parties, and the private key is kept by the authorized result user;

[0009] The enterprise debtor original data of the parties is encrypted using the public key to obtain encrypted enterprise debtor data, and the encrypted enterprise debtor data is scored for importance, and the encrypted enterprise debtor data with a score greater than a preset score threshold is preferentially selected to perform secure computing on the corresponding multi-party secure computing technology combination;

[0010] Based on the ciphertext calculation result output by the multi-party secure computing protocol, an enterprise debtor ciphertext portrait feature vector containing multi-dimensional features is formed, and the enterprise debtor ciphertext portrait feature vector is output to the authorized result user, so that the authorized result user decrypts the multi-dimensional feature enterprise debtor plaintext portrait feature through the private key.

[0011] Optionally, the encrypted enterprise debtor data with a score greater than a preset score threshold is preferentially selected to perform secure computing on the corresponding multi-party secure computing technology combination, comprising:

[0012] The encrypted enterprise debtor data with a score greater than a preset score threshold is read through a blocking mode, and the encrypted enterprise debtor data read by the blocking mode is preferentially selected to perform secure computing on the corresponding multi-party secure computing technology combination.

[0013] Optionally, after the encrypted enterprise debtor data with a score greater than a preset score threshold is preferentially selected to perform secure computing on the corresponding multi-party secure computing technology combination, it further comprises:

[0014] The encrypted enterprise debtor data with a score less than a preset score threshold is read through a non-blocking mode;

[0015] After the encrypted enterprise debtor data read by the blocking mode is preferentially selected to perform secure computing on the corresponding multi-party secure computing technology combination, the encrypted enterprise debtor data read by the non-blocking mode is selected to perform secure computing on the corresponding multi-party secure computing technology combination.

[0016] Optionally, the encrypted enterprise debtor data with a score greater than a preset score threshold is preferentially selected to perform secure computing on the corresponding multi-party secure computing technology combination, comprising:

[0017] determine a computational complexity, a real-time requirement, and a current network condition of the encrypted enterprise debtor data whose importance score is greater than a preset score threshold;

[0018] determine an allocation ratio of edge device and cloud server computing resources based on the computational complexity, the real-time requirement, and the current network condition of the encrypted enterprise debtor data;

[0019] prioritize a corresponding multi-party secure computing technology combination to perform secure computing based on the allocation ratio of edge device and cloud server computing resources.

[0020] Optionally, the determining of the allocation ratio of edge device and cloud server computing resources based on the computational complexity, the real-time requirement, and the current network condition of the encrypted enterprise debtor data comprises:

[0021] splitting the encrypted enterprise debtor data into a plurality of subtask data packets based on the computational complexity, the real-time requirement, and the current network condition of the encrypted enterprise debtor data;

[0022] determining the allocation ratio of edge device and cloud server computing resources based on the subtask data packets.

[0023] Optionally, the splitting of the encrypted enterprise debtor data into a plurality of subtask data packets based on the computational complexity, the real-time requirement, and the current network condition of the encrypted enterprise debtor data comprises:

[0024] splitting the encrypted enterprise debtor data into a plurality of functional modules according to functions or logics based on the computational complexity, the real-time requirement, and the current network condition of the encrypted enterprise debtor data;

[0025] determining a dependency relationship and a data flow relationship between the functional modules;

[0026] determining a plurality of subtask data packets based on the dependency relationship and the data flow relationship between the functional modules.

[0027] Optionally, the determining of the allocation ratio of edge device and cloud server computing resources based on the subtask data packets comprises:

[0028] determining that the encrypted enterprise debtor data node and its extension module can undertake corresponding subtasks based on characteristics and resource requirements of the subtask data packets;

[0029] determining a data computation execution sequence based on the encrypted enterprise debtor data node and its extension module can undertake corresponding subtasks;

[0030] Based on the data, the execution sequence determination edge device and the cloud server computing resource allocation ratio are calculated.

[0031] Optionally, the multi-dimensional feature of the enterprise debtor ciphertext portrait feature vector includes tax and financial cross features, energy consumption and operating condition features, cross-institution financial attribute features and joint risk assessment features.

[0032] Optionally, after the authorized result user decrypts the multi-dimensional feature of the enterprise debtor plaintext portrait feature through the private key, the method further comprises:

[0033] The multi-dimensional feature of the enterprise debtor plaintext portrait feature is risk interpreted, and a risk control decision suggestion is output, wherein the risk interpretation is generated based on correlation analysis of the multi-dimensional feature, and includes a risk level, an explanation basis and a business suggestion.

[0034] In addition, to achieve the above-mentioned purpose, the application also provides an enterprise debtor cross-domain portrait feature extraction system combined with multi-party secure computing, which comprises an enterprise debtor cross-domain portrait feature extraction device combined with multi-party secure computing, and the device comprises a memory, a processor and an enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure computing stored on the memory and executable on the processor, and the enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure computing is configured to implement the steps of the enterprise debtor cross-domain portrait feature extraction method combined with multi-party secure computing as described in any one of the above.

[0035] The application provides a kind of enterprise debtor cross-domain portrait feature extraction method and system combined with multi-party secure computing, the method comprises: obtaining the enterprise debtor original data stored in local database by each participant, wherein the each participant includes commercial bank, tax bureau and power grid company, and the enterprise debtor original data includes enterprise basic information, financial account data, tax data and energy consumption data;The each participant is accessed to multi-party secure computing system through secure communication channel, when receiving the feature generation computing task instruction issued to the each participant by central dispatcher, the encryption key pair required by multi-party secure computing protocol is generated based on threshold cryptography or trusted third party, public key is distributed to the each participant, and private key is kept by authorized result user;The enterprise debtor original data of the each participant is encrypted using public key, and encrypted enterprise debtor data is obtained, and the importance score of the encrypted enterprise debtor data is carried out, and the encrypted enterprise debtor data with the importance score greater than the preset score threshold is preferentially selected to execute secure computing on the corresponding multi-party secure computing technology combination;Based on the ciphertext calculation result output by multi-party secure computing protocol, enterprise debtor ciphertext portrait feature vector containing multidimensional features is integrated and formed, and the enterprise debtor ciphertext portrait feature vector is output to the authorized result user, so that the authorized result user decrypts the enterprise debtor plaintext portrait feature of multidimensional features through private key.The application carries out screening on encrypted data through importance score mechanism, and only the key data with score greater than the preset threshold is preferentially executed secure computing, to avoid complex operation on full amount of data without distinction, wherein the key data, such as tax sales ratio in tax data and flow fluctuation characteristics in financial account, ensures the security computing priority of core sensitive features through data classification, improves efficiency without reducing privacy protection strength, realizes the dual goals of data availability and high-efficiency and expandable computing. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 It is the structure schematic diagram of the enterprise debtor cross-domain portrait feature extraction equipment combined with multi-party secure computing of the hardware running environment involved in the embodiment scheme of the application.

[0037] Figure 2 It is the flow schematic diagram of the enterprise debtor cross-domain portrait feature extraction method combined with multi-party secure computing of the application.

[0038] Figure 3 It is the framework schematic diagram of the enterprise debtor cross-domain portrait feature extraction system combined with multi-party secure computing of the application.

[0039] The implementation of the object of the application, functional characteristics and advantages will be further described with reference to the drawings. DETAILED DESCRIPTION

[0040] It should be understood that the specific embodiments described herein are merely illustrative of the present application and should not be construed as limiting the present application.

[0041] Referring to Figure 1 , Figure 1 A structural schematic diagram of an enterprise debtor cross-domain portrait feature extraction device involving a hardware running environment combined with multi-party secure computing.

[0042] As Figure 1 shown, the enterprise debtor cross-domain portrait feature extraction device combined with multi-party secure computing can include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 can include a display screen (Display), and the optional user interface 1003 can also include a standard wired interface, a wireless interface. The wired interface of the user interface 1003 can be a USB interface in the present application. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 can be a high-speed random access memory (RAM) memory, or a stable memory (Non-volatile Memory, NVM), such as a magnetic disk memory. The memory 1005 can also be a storage device independent of the aforementioned processor 1001.

[0043] Those skilled in the art can understand Figure 1 that the structure shown in the foregoing description does not constitute a limitation on the enterprise debtor cross-domain portrait feature extraction device combined with multi-party secure computing, and can include more or fewer components than the diagram, or combine certain components, or different component arrangements.

[0044] As Figure 1 shown, the memory 1005 as a computer storage medium can include an operating system, a network communication module, a user interface module, and an enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure computing.

[0045] In Figure 1The network interface 1004 is mainly used for connecting a background server and communicating data with the background server. The user interface 1003 is mainly used for connecting an external device. The enterprise debtor cross-domain portrait feature extraction device combining multi-party secure calculation calls the enterprise debtor cross-domain portrait feature extraction program stored in the memory 1005 through the processor 1001 to execute the enterprise debtor cross-domain portrait feature extraction method combining multi-party secure calculation provided by the embodiment of the application.

[0046] Based on the above hardware structure, embodiments of the enterprise debtor cross-domain portrait feature extraction method combining multi-party secure calculation are provided.

[0047] Reference Figure 2 , Figure 2 The flowchart of the embodiment of the enterprise debtor cross-domain portrait feature extraction method combining multi-party secure calculation is shown. In the embodiment, the enterprise debtor cross-domain portrait feature extraction method combining multi-party secure calculation includes the following steps.

[0048] S10: Obtain enterprise debtor original data stored in a local database by each participant, wherein the participants include a commercial bank, a tax bureau, and a power grid company, and the enterprise debtor original data includes enterprise basic information, financial account data, tax data, and energy consumption data.

[0049] It should be noted that the participant refers to each independent entity or institution participating in data sharing, calculation, or cooperation in a certain specific scene, project, or system. In the context of cross-institution data collaborative analysis, the participant can be an institution or organization that has relevant data and is willing to share and jointly calculate the data on the premise of privacy protection. The commercial bank provides the financial account data of the enterprise debtor, such as the public account flow, loan records, repayment conditions, etc., to evaluate the liquidity and debt repayment ability of the enterprise. The tax bureau provides the tax data of the enterprise debtor, such as the amount of tax, value-added tax, income tax, etc., to analyze the operating status and profitability of the enterprise. The power grid company provides the energy consumption data of the enterprise debtor, such as the electricity consumption, electricity payment records, etc., to reflect the production scale and operation stability of the enterprise.

[0050] The enterprise debtor raw data refers to the unprocessed and most basic data set provided by each participant when assessing the credit status, debt repayment ability and operating risk of the enterprise debtor. These data are directly derived from the daily business activities and management systems of the enterprise, covering multiple dimensions such as basic information, financial data, tax data, energy consumption data, etc. of the enterprise, and are the underlying data basis for enterprise debtor portrait feature extraction and risk analysis. Among them, the enterprise basic information includes the basic information of enterprise name, registered address, legal representative, establishment time, business scope, etc., which is used to identify the identity and background of the enterprise. Financial account data includes transaction records of bank accounts, loan information, repayment records, etc., which is used to analyze the liquidity and credit status of the enterprise. Tax data includes the tax records, tax types, tax amounts, etc. of the enterprise, which is used to assess the operating compliance and profitability of the enterprise. Energy consumption data includes the power consumption of the enterprise, electricity payment records, etc., which is used to infer the production scale and operation status of the enterprise.

[0051] In specific implementation, the commercial bank extracts the financial account data of the enterprise debtor through the bank internal system, including the public account flow, loan records, repayment situation, etc., and ensures that the data format meets the input requirements of the multi-party secure computing system, wherein the bank internal system can be a core bank system or a credit management system, and the data format can be CSV, JSON. The tax bureau extracts the tax data of the enterprise debtor through the tax system, including tax records, tax types, tax amounts, etc., and performs desensitization processing on sensitive fields, wherein the tax system can be the third golden tax system, and the sensitive field can be the enterprise taxpayer identification number. The power grid company extracts the energy consumption data of the enterprise debtor through the power metering system, including power consumption, electricity payment records, etc., and performs standardization processing on the data, such as unifying the measurement unit to kilowatt-hour. Each participant stores the extracted raw data in the local encrypted database to ensure the security of the data in the transmission and storage process. Clean and preprocess the data, including removing duplicate values, filling missing values, and unifying data formats, etc. to improve data quality and computing efficiency, wherein the data formats are date format, currency unit, etc.

[0052] S20: Access the multi-party secure computing system through the secure communication channel by each participant, and when receiving the feature generation computing task instruction issued by the central dispatcher to each participant, generate the encryption key pair required by the multi-party secure computing protocol based on threshold cryptography or a trusted third party, the public key is distributed to each participant, and the private key is kept by the authorized result user.

[0053] It should be noted that the secure communication channel refers to a special data transmission channel constructed by using encryption technology to ensure the confidentiality, integrity and non-repudiation of data transmission between each participant and the multi-party secure computing system, prevent data from being stolen, tampered or forged, wherein the encryption technology is, for example, SSL / TLS protocol, VPN, etc. The multi-party secure computing system is a technical platform that supports multiple data holders to cooperatively complete joint calculation without directly sharing original data. Its core function is to realize data availability and invisibility through cryptographic protocols such as homomorphic encryption, secret sharing, and garbled circuits, to ensure that the privacy of each party's data is not leaked during the calculation process. The central scheduler is responsible for coordinating the whole process of multi-party secure computing, including task allocation, resource scheduling, key management, and computing node monitoring. Its role is to uniformly manage the computing tasks of each participant, ensuring the orderliness and efficiency of multi-party cooperation.

[0054] Threshold cryptography is a key fragmentation technology that divides a private key into multiple sub-keys (fragments) and distributes them to different participants. Only when not less than a preset threshold number of sub-keys are collected can the complete private key be reconstructed. For example, using a (t, n) threshold scheme, at least t participants in n participants cooperate to recover the private key, improving the security of the key and avoiding a single node holding the complete private key. The trusted third party is an entity in traditional multi-party secure computing that is responsible for generating and managing keys and coordinating the calculation process. Its existence relies on the assumption that each party is absolutely trustworthy, but in actual application, it can introduce single-point failure or trust risk, so the scheme includes threshold cryptography as an optional key generation method. The encryption key pair consists of a public key and a private key: the public key can be publicly distributed and used for encrypting data or verifying signatures. Anyone can encrypt data using the public key, but cannot reverse the private key from the public key. The private key is strictly confidential and only authorized parties hold it, which is used to decrypt data encrypted by the public key or generate signatures. The authorized result user refers to an entity authorized by each participant and entitled to receive and decrypt the final calculation result, such as financial institutions such as commercial banks. Its responsibility is to decrypt the plaintext portrait features using the private key after the completion of multi-party secure computing, which is used in business scenarios such as credit approval and risk assessment.

[0055] In specific implementation, each participant connects to the multi-party secure computing system through a dedicated line or an encrypted network, and uses the SSL / TLS protocol to encrypt the transmitted data end-to-end, ensuring that the data is encrypted into ciphertext during transmission, and even if intercepted, it cannot be cracked. Participants need to be authenticated by digital certificates or dynamic tokens when accessing the system to prevent illegal institutions from impersonating access to the system, for example, tax authorities need to provide special certificates for government agencies, and commercial banks need to be authenticated by a financial authentication center. Each participant registers its data type and computing capability with the central scheduler to facilitate the central scheduler to subsequently allocate appropriate computing tasks, wherein the data type is, for example, tax data, financial data, etc., and the computing capability is, for example, whether it supports homomorphic encryption operations.

[0056] When the authorized result user needs to generate the enterprise debtor portrait, a feature generation request is submitted to the central dispatcher, such as "calculate the solvency feature vector of enterprise A", the central dispatcher parses the required data dimensions and calculation logic according to the request, and generates specific feature generation calculation task instructions, where the data dimensions are, for example, financial account data, tax data, and energy data, and the calculation logic is trained through a logistic regression model. The central dispatcher broadcasts the task instructions to all relevant participants through a secure communication channel, and the instruction content includes task unique identifier, required data field, calculation protocol type, data encryption requirement, etc., where all relevant participants are, for example, the tax bureau, the power grid company and the corresponding commercial bank involved in the enterprise data, and the calculation protocol type is, for example, threshold cryptography or trusted third party mode.

[0057] When generating a key pair using threshold cryptography, the central dispatcher starts the threshold cryptography algorithm, such as Shamir secret sharing, and randomly splits the private key into n pieces, where n is the number of participants, and assigns each participant a unique sub-key (piece). At the same time, set the minimum threshold t required to recover the private key, such as t = 2, that is, at least 2 participants must cooperate to reconstruct the private key. Each participant sends their sub-key to the central dispatcher through a secure channel, and the central dispatcher synthesizes the global public key according to the threshold algorithm. This public key can be used to encrypt data, but cannot be used to infer the private key pieces. When generating a key pair using a trusted third party, the trusted third party acts as a key generation center, generates an asymmetric encryption key pair, distributes the public key to all participants, and sends the private key directly to the authorized result user. The trusted third party can be an industry regulator or a certified neutral agency, and ensures the integrity of the key transmission through digital signature. It should be noted that the trusted third party mode relies on the trust of all parties in the trusted third party, and is suitable for scenarios where supervision is strict and the third party has strong public credibility, such as government data sharing platforms.

[0058] It should be understood that no matter which mode is adopted, the generated public key is securely distributed to all participants by the central dispatcher, and each participant uses the public key to encrypt the local raw data, such as encrypting the enterprise tax data into ciphertext using the public key. If threshold cryptography is used, the private key is composed of the sub-key pieces held by each participant, and does not need to be stored centrally. Only when decrypting does the authorized result user collect at least t pieces and reconstruct the complete private key, ensuring that the transmission process of the pieces is secure. If a trusted third party is used, the private key is directly managed by the authorized result user and stored in a hardware security module or an encrypted database, ensuring that multiple authentications are required to access the private key. Among them, after receiving the public key, each participant can verify the integrity of the public key through hash check or digital signature to prevent man-in-the-middle attacks from replacing malicious public keys. After receiving the "key reception confirmation" from all participants, the central dispatcher marks the task as "ready state" and enters the next stage of data encryption and secure calculation process.

[0059] S30: encrypt the enterprise debtor original data of each participant using the public key to obtain encrypted enterprise debtor data, and score the importance of the encrypted enterprise debtor data, and preferentially select the encrypted enterprise debtor data with a score greater than a preset score threshold to perform secure calculation using the corresponding multi-party secure calculation technology combination.

[0060] It should be noted that the importance score refers to quantitatively scoring each field or data subset of the enterprise debtor original data according to its influence on the final calculation target. The higher the score, the greater the contribution of the data to feature generation or model training. For example, the "tax amount" field in the tax data may be more important than the "enterprise registration address". The score is usually based on the business relevance, information gain rate, historical model contribution of the data. The preset score threshold is a critical value of importance that is pre-set by business requirements or historical experience. When the importance score of the data exceeds the threshold, it is determined to be "high-value data" and higher security or more accurate multi-party secure calculation technology needs to be used preferentially; "low-value data" that does not exceed the threshold can use lightweight technology to improve efficiency. The threshold setting needs to be combined with industry standards, such as the threshold of tax data in financial risk control is usually higher than that of basic information. The multi-party secure calculation technology combination uses multiple secure calculation technologies such as homomorphic encryption, secret sharing, garbled circuit, and federated learning according to data sensitivity, calculation complexity, and importance. For example, high importance data such as tax data and loan balance uses a combination of "garbled circuit + homomorphic encryption" to ensure privacy and calculation accuracy; low importance data such as enterprise business scope and establishment time uses a combination of "secret sharing + lightweight encryption" to reduce computational overhead.

[0061] In specific implementation, the enterprise debtor original data extracted by each participant from the local database is split by field or data block, such as "loan balance", "tax amount", and "monthly average electricity consumption" as independent data blocks. The participant uses the public key distributed in the step to encrypt each data block to generate encrypted enterprise debtor data. For example, the commercial bank encrypts the loan balance field of enterprise A using the public key to generate ciphertext C1, and the tax authority encrypts the tax amount to generate ciphertext C2. The encrypted data retains metadata tags such as data type, participating party, and associated enterprise ID, facilitating subsequent scoring and matching. The central dispatcher or independent scoring module generates a score of 0-10 for each encrypted data block according to the preset rules or machine learning model, such as a "tax amount" score of 9 and a "business scope" score of 3. The scoring process needs to record the weight allocation basis, such as 60% for business expert rules and 40% for historical data training, to ensure interpretability and compliance.

[0062] The score of each encrypted data block is compared with a preset score threshold (e.g., the threshold is set to 6 points): high-value data (score > 6 points): such as "loan balance" and "number of overdue payments" in financial account data, "value-added tax payment" in tax data, and "average monthly electricity consumption" in energy data. Low-value data (score ≤ 6 points): such as "legal representative name" and "registered phone number" in basic enterprise information, and "account opening time" in financial data. High-value data and low-value data are stored in different task queues, respectively. High-value data enters the "priority calculation queue", and low-value data enters the "regular calculation queue".

[0063] The central dispatcher assigns "obfuscation circuit + homomorphic encryption" combination to high-value data according to data grouping: first, use homomorphic encryption to perform addition / multiplication operation on tax amount ciphertext, such as calculating the total tax amount of the enterprise in the past 12 months, and then verify the logical conditions through obfuscation circuit, such as "whether the tax amount has decreased for 3 consecutive months". Low-value data uses secret sharing technology: the hash value of the enterprise registration address is divided into multiple participants, and only when de-duplication is needed, the threshold reconstruction is performed to avoid transmitting the complete plaintext. More computing nodes or higher priority computing power, such as GPU cluster, are allocated to high-value data tasks to ensure that their secure computing process is executed first. For example, 80% of the system resources are occupied by complex computing tasks of tax data, and only 20% of the resources are used for basic information processing. The central dispatcher tracks the progress of high-value data calculation in real time, and if there is a delay, such as too long time consumption of obfuscation circuit calculation, it automatically triggers resource expansion or technology degradation, such as switching from full homomorphic encryption to partial homomorphic encryption, to balance efficiency and security.

[0064] Further, in another embodiment, the priority selection of the encrypted enterprise debtor data with the importance score greater than the preset score threshold to perform secure calculation by using the corresponding multi-party secure calculation technology combination comprises:

[0065] The encrypted enterprise debtor data with the importance score greater than the preset score threshold is read by blocking mode, and the encrypted enterprise debtor data read by the blocking mode is prioritized to perform secure calculation by using the corresponding multi-party secure calculation technology combination.

[0066] It should be noted that the blocking mode is a synchronous data reading mechanism: when the program tries to read certain data (such as high-importance encrypted enterprise debtor data), it will wait until the data is ready or timeout, and during this period, it will not process other tasks. Unlike non-blocking mode (Non-Blocking Read, which allows read operation to return immediately regardless of whether the data is ready or not), blocking mode ensures that high-priority data is processed first by exclusive waiting, avoiding delays caused by low-priority tasks occupying resources, and is suitable for scenarios with extremely high real-time and priority requirements, and can ensure the security of data transmission.

[0067] In a specific implementation, when the secure computing module needs to obtain high-value data, it initiates a blocking read request to the high-priority queue proactively, and the computing thread enters a blocking state until there is a data block in the queue that meets the conditions, i.e., the data is encrypted and the importance score is completed, during which time the read request of the low-priority queue is not responded. A blocking timeout (e.g., 300 ms) is set, and if the data is not obtained within the timeout, the thread resource is released to avoid deadlock, which needs to be adjusted in combination with the business tolerance. The timeout time is usually short in the financial scenario. After reading the data, the data integrity and authority are immediately verified to prevent illegal data from being mixed in. By blocking mode reading high importance data, strong binding of data priority and computing resources is achieved, ensuring that the data critical to business decision-making is processed safely and preferentially.

[0068] In addition, after the encrypted enterprise debtor data with the importance score greater than the preset score threshold is preferentially selected to perform secure computation by using the corresponding multi-party secure computation technology combination, the method further includes:

[0069] reading, by using a non-blocking mode, the encrypted enterprise debtor data with the importance score less than the preset score threshold;

[0070] After the secure computation of the encrypted enterprise debtor data read in the blocking mode is completed, the secure computation of the encrypted enterprise debtor data read in the non-blocking mode is performed by using the corresponding multi-party secure computation technology combination.

[0071] It should be noted that the non-blocking mode reading is an asynchronous data reading mechanism: when the program attempts to read low-importance encrypted enterprise debtor data, it can return immediately without waiting for the data to be ready, allowing the thread to continue processing other tasks, such as computing high-priority data or system monitoring. When the data is ready, the subsequent processing is triggered through a callback mechanism or event notification. Unlike the exclusive waiting of the blocking mode, the non-blocking mode improves system resource utilization through asynchronous operation, which is suitable for low-priority tasks with low security requirements, such as format verification of enterprise basic information. Moreover, since the data can be returned immediately without waiting for the data to be ready, the processing speed of the data is improved.

[0072] In another embodiment, the encrypted enterprise debtor data with the importance score greater than the preset score threshold is preferentially selected to perform secure computation by using the corresponding multi-party secure computation technology combination, which includes:

[0073] determining the computing complexity, real-time requirement, and current network condition of the encrypted enterprise debtor data with the importance score greater than the preset score threshold;

[0074] determine the allocation proportion of the edge device and the cloud server computing resources based on the calculation complexity, real-time requirement and current network status of the encrypted enterprise debtor data;

[0075] perform secure computation based on the corresponding multi-party secure computation technology combination according to the allocation proportion of the edge device and the cloud server computing resources.

[0076] It should be noted that the calculation complexity refers to the algorithm complexity, data processing amount and computing power consumption degree required in the secure computation process of the encrypted enterprise debtor data. For example, high complexity involves ciphertext multiplication, division, logical judgment nesting, such as risk score model training under homomorphic encryption; low complexity only requires ciphertext addition, comparison or simple aggregation, such as enterprise number statistics under secret sharing. The real-time requirement is divided into levels according to the time sensitivity of the calculation result in the business scenario, usually divided into strong real-time and weak real-time, among which strong real-time is real-time risk assessment in loan approval; weak real-time is monthly enterprise credit report generation. The edge device is a distributed computing node deployed near the data collection or use terminal, such as an enterprise local server, a financial institution front-end machine, and an Internet of Things gateway, which has light computing power and low-delay data processing capability, is suitable for processing low-complexity and strong real-time tasks, and avoids the delay and bandwidth consumption of long-distance data transmission to the cloud. The cloud server computing resource allocation proportion dynamically divides the computing power allocation strategy of the edge device and the cloud server according to the task requirement, such as edge processing 60% simple calculation and cloud processing 40% complex calculation, to realize the collaborative architecture of "edge lightweight processing + cloud deep calculation", balance delay, computing power and security.

[0077] The method further includes:

[0078] The method further includes:

[0079] The method further includes:

[0080] It should be noted that the sub-task data packet is a lightweight task unit that splits the original encrypted enterprise debtor data and its calculation task according to function modules, calculation steps, and data sensitivity. It can contain task description, data parameters, resource requirements, and dependency relationships, wherein the task description is the specific calculation logic, such as "enterprise tax amount ciphertext addition" and "loan overdue number logical judgment"; the data parameters are encrypted sub-data blocks, such as "2024Q3 tax amount ciphertext" and "associated enterprise number ciphertext"; the resource requirements are computing power level, network transmission volume, and real-time constraint; and the dependency relationship is the association between upstream and downstream tasks, such as "sub-task B needs to wait for the ciphertext result of sub-task A". The essence is the minimization of task granularity encapsulation, which supports the distributed collaborative processing of edge devices and cloud servers. By splitting the encrypted enterprise debtor data into sub-task data packets with clear resource requirements and combining the dynamic calculation and allocation proportion of the edge-cloud capability portrait, the scheme realizes the accurate splitting, on-demand scheduling, and collaborative execution of secure calculation tasks. This fine-grained resource allocation mechanism not only takes advantage of the low-latency advantage of edge devices but also utilizes the powerful computing power of the cloud, providing a distributed solution that balances efficiency, cost, and compliance for secure calculation of high-value data, especially suitable for complex business scenarios involving cross-regional and multi-agency collaboration.

[0081] Specifically, the encrypted enterprise debtor data is split into a plurality of sub-task data packets based on the calculation complexity, real-time requirement, and current network status of the encrypted enterprise debtor data, including:

[0082] The encrypted enterprise debtor data is split into a plurality of function modules according to function or logic based on the calculation complexity, real-time requirement, and current network status of the encrypted enterprise debtor data.

[0083] The dependency relationship and data flow relationship between the function modules are determined.

[0084] A plurality of sub-task data packets are determined based on the dependency relationship and data flow relationship between the function modules.

[0085] It should be noted that the functional module refers to the smallest executable unit that splits the complete calculation task of encrypted enterprise debtor data according to business function attributes or calculation logic independence, has the characteristics of single responsibility, clear input and output, and technical adaptability, wherein the single responsibility refers to only realizing a specific function, such as “enterprise basic information ciphertext verification” and “loan overdue day ciphertext calculation”; the clear input and output refer to containing standardized ciphertext input parameters and ciphertext output results, wherein the ciphertext input parameters are, for example, “enterprise code ciphertext” and “loan contract number ciphertext”, and the ciphertext output results are, for example, “verification pass / fail ciphertext” and “overdue day ciphertext”; the technical adaptability refers to being able to independently match the calculation technology of an edge device or a cloud server, such as a lightweight encryption module adapting to an edge, and a complex algorithm module adapting to a cloud. The essence is the atomic encapsulation of business logic, providing a basic unit for subsequent dependency analysis and resource allocation. The dependency relationship refers to the execution sequence constraint between functional modules, including time sequence dependency, data dependency, and resource dependency, wherein the time sequence dependency refers to the fact that a subsequent module can be started only after the output result of a previous module, such as the fact that the “risk scoring model calculation” module depends on the output of the “feature ciphertext aggregation” module; the data dependency refers to the fact that the input parameters of a subsequent module must be provided by the output of a previous module, such as the fact that the “tax amount proportion ciphertext calculation” module depends on the outputs of the “tax amount ciphertext” and “operating income ciphertext” two preceding modules; the resource dependency refers to the mutual exclusion relationship when multiple modules compete for the same computing resource, such as the queuing order when two modules both need to call the SM4 hardware encryption unit of an edge device. Through the definition of the dependency relationship, it is ensured that the sub-task data packet conforms to the business logic sequence when distributedly executed. The data flow relationship refers to the ciphertext transmission path between functional modules, including data flow direction, data format, and data volume, wherein the data flow direction clearly indicates that the ciphertext data is transmitted from the output of which module to the input of which module, such as the fact that the “enterprise equity structure ciphertext” flows from the “industrial and commercial data analysis” module to the “correlation relationship analysis” module; the data format defines the ciphertext encoding format of the transmission data, such as the integer ciphertext after homomorphic encryption and the fragmented ciphertext after secret sharing; and the data volume quantifies the size of the ciphertext data that flows at a time, such as a 10KB ciphertext data packet and a 500KB ciphertext data packet, serving as the basis for network resource allocation. The data flow relationship directly affects the network transmission efficiency of the sub-task data packet and the selection of secure calculation technology.

[0086] It should be understood that by splitting the encrypted enterprise debtor data into the smallest executable unit, i.e. functional module, according to the function / logic, the originally coupled complex computing task is transformed into an independent controllable atomic operation, such as splitting the "enterprise risk score calculation" into "basic data verification", "feature ciphertext extraction", "model inference" and other modules, so that each module only undertakes a single computing function. Breaking the traditional black box task processing mode, realizing the explicit decomposition of computing logic, facilitating subsequent independent resource adaptation, performance optimization and security audit of each module. Unlike the problem of too coarse scheduling granularity caused by traditional whole task packaging processing, such as directly scheduling 10GB ciphertext data to the cloud, ignoring the lightweight processing capability of the edge device, the present embodiment splits the modules with controllable granularity, so that the system can dynamically adjust the processing unit size according to the real-time network condition, and when the network is congested, it is split into 10KB small packets for edge processing, and when the network is good, it is combined into 100KB large packets for cloud processing, and the scheduling flexibility is improved by more than 70%.

[0087] Specifically, the allocation ratio of the edge device and the cloud server computing resources is determined based on the sub-task data packet, including:

[0088] Based on the characteristics and resource requirements of the sub-task data packet, the encrypted enterprise debtor data node and its expansion module can undertake the corresponding sub-task;

[0089] Based on the sub-task that the encrypted enterprise debtor data node and its expansion module can undertake, the data computing execution sequence is determined;

[0090] Based on the data computing execution sequence, the allocation ratio of the edge device and the cloud server computing resources is determined.

[0091] It should be noted that the encrypted enterprise debtor data node refers to a physical or logical computing unit with encrypted data processing capability, which is a basic execution entity carrying sub-task data packets, and can be an edge device deployed locally in the enterprise, such as an industrial computer or an intelligent gateway, or a logical computing node in a cloud server cluster, supporting secure computation of ciphertext data, such as homomorphic encryption operation and secret sharing decryption; built-in secure computing engine, such as supporting SM2 / SM3 / SM4 national encryption algorithm, with data encryption transmission, access control and other security capabilities to ensure the security of encrypted enterprise debtor data during processing in the node; Each node defines a clear upper limit of computing power, storage and network bandwidth as a constraint condition for resource allocation. The expansion module refers to a functional enhancement component that can be dynamically loaded onto the data node to expand the encrypted data processing capability of the node, including a computing power expansion module, a storage expansion module and a network expansion module; Among them, the computing power expansion module is, for example, a GPU acceleration card or a dedicated encryption chip; the storage expansion module is, for example, a distributed cache component or a secure database plug-in; the network expansion module is, for example, a 5G communication module of an edge node or an RDMA network adapter of a cloud node. The expansion module can be dynamically mounted according to the special needs of the sub-task data packet, such as temporarily loading a GPU acceleration module for a cloud node when processing a high-complexity homomorphic encryption task, to achieve flexible adaptation of node capability.

[0092] It should be understood that by analyzing the computing power requirement, real-time requirement and data size of the sub-task data packet, it is allocated to the most suitable node and expansion module. Low computing power requirement and strong real-time task are directly processed by the edge node to avoid network delay of cloud remote call, and the response time is shortened by 70%; Local compliance tasks are allocated to enterprise local edge nodes, wherein local compliance tasks such as "tax data ciphertext verification" need to meet the requirement of not leaving the park. High computing power requirement and batch processing task are scheduled to the cloud, which utilizes its elastic computing power and distributed storage capability, and the computing efficiency is improved by more than 5 times, wherein high computing power requirement and batch processing task such as "historical default data ciphertext modeling" requires 100-core parallel computing; Cross-regional collaborative tasks are coordinated by cloud nodes to solve the problem of scattered computing power of edge nodes, wherein cross-regional collaborative tasks such as "cross-border enterprise correlation ciphertext analysis".

[0093] In addition, the special needs of the sub-tasks are loaded with extension modules, breaking through the inherent capacity limit of the node, for example, when the sub-tasks contain a large number of ciphertext matrix multiplication, temporarily load the GPU acceleration module for the cloud node, so that the calculation speed is increased by 300%, from 500ms / time of CPU processing to 150ms / time of GPU; for example, when the edge node processes a large batch of ciphertext data storage, dynamically mount the distributed cache extension module, and the ciphertext reading speed is increased from 10MB / s to 50MB / s, solving the bottleneck of insufficient memory in the edge. The execution order is determined based on the dependency relationship of the sub-task data packet, avoiding logical confusion in distributed computing, for example, when the pre-sub-task is not completed, the post-sub-task will not be scheduled for execution, and the global execution sequence is generated through the dependency graph traversal algorithm to ensure that the ciphertext data flows according to the business logic, wherein the pre-sub-task is, for example, the "basic data ciphertext verification package", and the post-sub-task is, for example, the "risk model reasoning package"; the post-node only receives the complete output ciphertext of the pre-node, avoiding the calculation error caused by incomplete data transmission in the traditional distributed system, and the task failure rate is reduced from 5% in the traditional scheme to below 0.1%. Through execution sequence planning, invalid waiting between nodes is reduced, for example, for sub-tasks without dependency relationship, such as "enterprise basic information verification package" and "financial statement ciphertext analysis package", they are scheduled to different edge nodes for parallel processing, which saves 50% time compared with serial execution; after the pre-node completes the sub-task, the resource pre-allocation of the post-node is triggered immediately, such as applying for GPU resources for the cloud node in advance, so that the task connection delay between nodes is reduced from 100ms in the traditional scheme to within 10ms.

[0094] S40: Based on the ciphertext calculation result output by the multi-party secure calculation protocol, an enterprise debtor ciphertext portrait feature vector containing multi-dimensional features is integrated, and the enterprise debtor ciphertext portrait feature vector is output to the authorized result user, so that the authorized result user obtains the enterprise debtor plaintext portrait feature with multi-dimensional features through private key decryption.

[0095] It should be noted that the multi-party secure calculation protocol allows multiple participants to cooperatively complete joint calculation of ciphertext data without leaking their own original data. The ciphertext calculation result refers to the intermediate or final encrypted result output after joint calculation of encrypted enterprise debtor data, such as financial statements, tax records, and credit information, based on the multi-party secure calculation protocol. The enterprise debtor ciphertext portrait feature vector refers to a high-dimensional encrypted vector formed by integrating multi-dimensional encrypted features in a standardized format, which is used to depict the comprehensive risk characteristics of the enterprise debtor. The enterprise debtor ciphertext portrait feature vector with multi-dimensional features can include tax and financial cross-feature, energy consumption and operating condition feature, cross-institution financial attribute feature, and joint risk assessment feature. The authorized result user refers to an entity authorized by the data provider, such as a bank risk control system or a guarantee institution decision platform, which has the decryption permission.

[0096] It should be understood that each participant only inputs encrypted local data, such as enterprise tax record ciphertext, credit record ciphertext, to the system without exposing the original plaintext. The finally integrated ciphertext portrait feature vector remains encrypted to any intermediate link before being decrypted by the authorized user, fundamentally avoiding the risk of data leakage. Based on the mathematical properties of the multi-party secure computing protocol, the calculation process only operates on ciphertext, and the result decryption needs to rely on the private key of the authorized user, forming a secure closed loop where data does not leave the local and the result is available. Taking the financial risk control scene as an example, the traditional mode requires the enterprise to submit a complete set of plaintext financial data to the bank, which has the risk of data abuse; while through this step, the bank can only obtain the integrated ciphertext feature vector such as "asset-liability ratio ciphertext value" and "tax compliance level ciphertext value", and after authorized decryption, the plaintext features used for risk control modeling are obtained, which not only meets the business demand for multi-dimensional data, but also avoids the excessive exposure of original data.

[0097] In another embodiment, after the authorized result user decrypts the multi-dimensional feature of the enterprise debtor plaintext portrait feature through the private key, the method further comprises:

[0098] risk interpretation of the multi-dimensional feature of the enterprise debtor plaintext portrait feature, and outputting a risk control decision suggestion, wherein the risk interpretation is generated based on correlation analysis of the multi-dimensional feature, including risk level, explanation basis and business suggestion.

[0099] It should be noted that risk interpretation refers to the process of converting data features into business understandable risk semantic information through a combination of quantitative analysis and qualitative judgment based on multi-dimensional plaintext features. Specifically, numerical features such as asset-liability ratio 0.65 are converted into risk implications such as "moderate to low debt repayment ability"; the interaction of multiple features such as "high asset-liability ratio" + "low tax level" derives "cash flow tightness risk"; and the interpretation rules are adjusted according to the industry to which the enterprise belongs, such as manufacturing / Internet, with the interpretation rules such as the "R&D investment proportion" of Internet companies being higher than that of manufacturing. Correlation analysis refers to the technical process of mining the correlation, causality or risk transmission path between multi-dimensional features through statistical methods or machine learning models. Risk level refers to the standardized grading result of the comprehensive risk degree of the enterprise debtor, usually represented by ordinal level or score interval, including grading standards and dynamic adjustment, two core elements, wherein the grading standards refer to risk thresholds based on industry benchmarks and historical default data; dynamic adjustment refers to real-time calibration of grade boundaries according to macroeconomic data to ensure the timeliness of risk assessment.

[0100] The explanation according to the risk level and the risk control decision includes key feature contribution, rule triggering path and model decision logic; the key feature contribution refers to quantifying the influence of each feature on the risk level; the rule triggering path refers to recording the specific business rules triggered if a rule engine is used; and the model decision logic refers to explaining the core basis of model judgment by using SHAP value, LIME and other explainability technologies if a machine learning model is used. The business suggestion is formulated according to different risk levels and has a clear business direction, for example, the suggestion for a low-risk enterprise is to "simplify the credit granting process and give interest rate concessions"; the suggestion for a medium-risk enterprise is to "require additional collateral and shorten the post-loan inspection period"; and the suggestion for a high-risk enterprise is to "refuse to grant credit and include it in the risk warning list".

[0101] In addition, with reference to Figure 3 The embodiment of the present application also provides an enterprise debtor cross-domain portrait feature extraction system combined with multi-party secure calculation, which comprises an enterprise debtor cross-domain portrait feature extraction device combined with multi-party secure calculation, and the device comprises a memory, a processor, and an enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure calculation stored in the memory and capable of running on the processor, wherein the enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure calculation is configured to implement the steps of the enterprise debtor cross-domain portrait feature extraction method combined with multi-party secure calculation according to any one of the above.

[0102] Other embodiments or specific implementation manners of the enterprise debtor cross-domain portrait feature extraction system combined with multi-party secure calculation can refer to the above method embodiments, and details are not described herein.

[0103] It should be noted that, in this document, the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or system. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0104] The above embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. In the unit claims of several devices, several of these devices can be embodied by the same hardware item. The use of the words first, second, and third does not represent any order, and these words can be interpreted as names.

[0105] Those skilled in the art can clearly understand the above-mentioned example method can be realized by means of software and the necessary general hardware platform, of course, also can be through hardware, but in many cases the former is a better implementation. Based on such understanding, the technical solutions of the present application essentially or say the part of the prior art contribution can be embodied in the form of software products, the computer software product is stored in a storage medium (such as read only memory image (Read Only Memory image, ROM) / random access memory (Random Access Memory, RAM), disk, optical disk), including a number of instructions to make a terminal user equipment (may be a mobile phone, computer, server, air conditioner, or network user equipment, etc.) executes the method described in various embodiments of the present application.

[0106] The above is only the preferred embodiment of the present application, not therefore limit the patent scope of the present application, any equivalent structure or equivalent flow transformation using the content of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An enterprise debtor cross-domain profiling feature extraction method combined with multi-party secure computation, characterized in that, The method comprises: obtaining enterprise debtor original data stored in a local database by each participant, wherein the participants include commercial banks, tax authorities and power grid companies, and the enterprise debtor original data includes enterprise basic information, financial account data, tax data and energy consumption data; accessing a multi-party secure computing system by the participants through a secure communication channel, and generating an encryption key pair required by a multi-party secure computing protocol based on threshold cryptography or a trusted third party when a feature generation computing task instruction issued by a central dispatcher to the participants is received, a public key is distributed to the participants, and a private key is kept by an authorized result user; encrypting the enterprise debtor original data of the participants using the public key to obtain encrypted enterprise debtor data, and performing importance scoring on the encrypted enterprise debtor data, and preferentially selecting encrypted enterprise debtor data with an importance score greater than a preset score threshold to perform secure computing using a corresponding multi-party secure computing technology combination; integrating an enterprise debtor ciphertext portrait feature vector containing multi-dimensional features based on a ciphertext computing result output by the multi-party secure computing protocol, and outputting the enterprise debtor ciphertext portrait feature vector to the authorized result user to enable the authorized result user to obtain an enterprise debtor plaintext portrait feature with multi-dimensional features by decrypting using the private key.

2. The enterprise debtor cross-domain profile feature extraction method of claim 1, wherein, The preferential selection of the encrypted enterprise debtor data with the importance score greater than the preset score threshold to perform secure computing using the corresponding multi-party secure computing technology combination comprises: reading the encrypted enterprise debtor data with the importance score greater than the preset score threshold through a blocking mode, and preferentially selecting the encrypted enterprise debtor data read through the blocking mode to perform secure computing using the corresponding multi-party secure computing technology combination.

3. The enterprise debtor cross-domain profile feature extraction method of claim 2, wherein, After the preferential selection of the encrypted enterprise debtor data with the importance score greater than the preset score threshold to perform secure computing using the corresponding multi-party secure computing technology combination, the method further comprises: reading the encrypted enterprise debtor data with the importance score less than the preset score threshold through a non-blocking mode; after the preferential selection of the encrypted enterprise debtor data read through the blocking mode to perform secure computing using the corresponding multi-party secure computing technology combination is completed, preferentially selecting the encrypted enterprise debtor data read through the non-blocking mode to perform secure computing using the corresponding multi-party secure computing technology combination.

4. The enterprise debtor cross-domain profile feature extraction method of claim 1, wherein, The preferential selection of the encrypted enterprise debtor data with the importance score greater than the preset score threshold to perform secure computing using the corresponding multi-party secure computing technology combination comprises: determining the calculation complexity, real-time requirement and current network condition of the encrypted enterprise debtor data with the importance score greater than the preset score threshold; determining the allocation proportion of edge device and cloud server computing resources based on the calculation complexity, real-time requirement and current network condition of the encrypted enterprise debtor data; preferentially selecting the corresponding multi-party secure computing technology combination to perform secure computing based on the allocation proportion of edge device and cloud server computing resources.

5. The enterprise debtor cross-domain profile feature extraction method of claim 4, wherein, The allocation ratio of the edge device and the cloud server computing resources is determined based on the calculation complexity, real-time requirement and current network condition of the encrypted enterprise debtor data, comprising: The encrypted enterprise debtor data is split into multiple sub-task data packets based on the calculation complexity, real-time requirement and current network condition of the encrypted enterprise debtor data. The allocation ratio of the edge device and the cloud server computing resources is determined based on the sub-task data packets.

6. The enterprise debtor cross-domain profile feature extraction method of claim 5, wherein, The encrypted enterprise debtor data is split into multiple sub-task data packets based on the calculation complexity, real-time requirement and current network condition of the encrypted enterprise debtor data, comprising: The encrypted enterprise debtor data is split into multiple functional modules according to functions or logics based on the calculation complexity, real-time requirement and current network condition of the encrypted enterprise debtor data; The dependency relationship and data flow relationship between the functional modules are determined; Multiple sub-task data packets are determined based on the dependency relationship and data flow relationship between the functional modules.

7. The enterprise debtor cross-domain profile feature extraction method of claim 5 in connection with secure multi-party computation, characterized in that, The allocation ratio of the edge device and the cloud server computing resources is determined based on the sub-task data packets, comprising: The corresponding sub-tasks that can be undertaken by the encrypted enterprise debtor data nodes and their extension modules are determined based on the characteristics and resource requirements of the sub-task data packets; The data calculation execution sequence is determined based on the corresponding sub-tasks that can be undertaken by the encrypted enterprise debtor data nodes and their extension modules; The allocation ratio of the edge device and the cloud server computing resources is determined based on the data calculation execution sequence.

8. The enterprise debtor cross-domain profile feature extraction method of claim 1 integrated with secure multi-party computation, wherein, The multi-dimensional feature enterprise debtor ciphertext portrait feature vector includes tax and financial cross-feature, energy consumption and operating condition feature, cross-institution financial attribute feature and joint risk assessment feature.

9. The enterprise debtor cross-domain profile feature extraction method of claim 1 integrated with secure multi-party computation, wherein, After the authorized result user obtains the multi-dimensional feature enterprise debtor plaintext portrait feature by decrypting with a private key, the method further comprises: The multi-dimensional feature enterprise debtor plaintext portrait feature is risk interpreted, and a risk control decision suggestion is output, wherein the risk interpretation is generated based on multi-dimensional feature correlation analysis, including risk level, explanation basis and business suggestion.

10. An enterprise debtor cross-domain profiling feature extraction system incorporating multi-party secure computation, the system comprising: The system comprises an enterprise debtor cross-domain portrait feature extraction device combined with multi-party secure calculation, which comprises a memory, a processor and an enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure calculation stored on the memory and executable on the processor, and the enterprise debtor cross-domain portrait feature extraction program combined with multi-party secure calculation is configured to implement the steps of the enterprise debtor cross-domain portrait feature extraction method combined with multi-party secure calculation according to any one of claims 1 to 9.

Images