Financial data privacy protection method and system based on large model

By sorting, screening and sensitivity assessment of financial data, combining historical privacy records and differential privacy models, dynamic privacy protection is achieved, and dynamic budget allocation problems in financial data privacy protection is solved, and the privacy protection adaptability and data practicality are improved. .

CN118965441AActive Publication Date: 2024-11-15BEIJING WANGZHI TIANYUAN BIG DATA TECH CO LTD +1

Patent Information

Application Number
CN202411419450.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-12
Publication Date
2024-11-15
Estimated Expiration
2044-10-12

AI Technical Summary

Technical Problem

In the protection of financial data privacy, it is difficult for the existing technology to achieve dynamic privacy budget allocation, resulting in imbalance in computing resources and noise impacts, which in turn affects the practicality of data and the effectiveness of privacy protection.

Method used

By acquiring financial data, performing classification screening and sensitivity assessment, obtaining historical privacy records, calculating associated privacy losses and differential privacy constraints, determining differential privacy index and dynamic privacy boundaries, and achieving dynamic protection of financial data.

Benefits of technology

It effectively improves the adaptability of financial data privacy protection, ensures that privacy protection does not exceed acceptable limits, flexibly adjusts privacy protection policies, maximizes data practicality and effectiveness, and accurately controls privacy leakage risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965441A_ABST
    Figure CN118965441A_ABST
Patent Text Reader

Abstract

The invention provides a financial data privacy protection method and system based on a large model, and the method comprises the steps: determining the privacy sensitivity of each piece of financial privacy data in a financial privacy data cluster after obtaining the financial data in a current financial trade; determining the associated privacy loss of all financial privacy data in the financial privacy data cluster according to each privacy sensitivity and all privacy loss amounts, and determining the differential privacy constraint of the financial privacy data cluster according to all the associated privacy loss and the total privacy budget of a differential privacy model in the current financial trade; and determining a dynamic privacy boundary of each piece of financial privacy data in the financial privacy data cluster in the historical financial trade through the differential privacy index and each associated privacy loss, and further performing dynamic protection on the financial data in the current financial trade through the dynamic privacy boundary. According to the scheme, privacy protection is carried out on the financial data based on the dynamic adaptive distribution of the privacy budget, and the privacy protection adaptive capability during privacy protection of the financial data can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of financial privacy protection, and more specifically, to a method and system for financial data privacy protection based on a large model. Background Art

[0002] Privacy protection is the use of technical means, policies and practices to ensure the security and privacy of personal or organizational information during collection, storage, processing and transmission. Key measures include data encryption, access control, data desensitization and differential privacy. Data encryption uses symmetric and asymmetric algorithms to ensure the confidentiality of sensitive information during transmission and storage. Access control limits data access rights based on roles and attributes. Data desensitization technologies, such as pseudonymization and anonymization, prevent leaked data from being directly associated with individuals. In addition, federated learning allows multiple institutions to jointly train models without sharing original data, improving data utilization while protecting privacy. Through these technical measures and compliance strategies, privacy protection ensures data security while improving the transparency and trust of information processing.

[0003] Financial data privacy protection based on large models refers to the use of multiple privacy protection technologies to ensure data security and privacy when using large-scale pre-trained models for financial data analysis and processing. The main technologies include federated learning, dynamic encryption and data desensitization. Federated learning allows multiple institutions to jointly train models without sharing original data. Homomorphic encryption supports performing calculations on encrypted data to ensure data confidentiality. Data desensitization technologies, such as pseudonymization and anonymization, prevent data leakage and provide effective privacy protection for financial data. However, in the existing financial data privacy protection, financial data privacy processing methods based on large models are usually used to protect financial data. Privacy protection is carried out, but in this process, the imbalance of computing resource consumption and noise impact of large models often occur, which will cause protection delays to the privacy budget of financial data. For example, a management privacy budget that is too small will make the noise of financial data too large, affecting the practicality of financial data. A management privacy budget that is too large will make it difficult to achieve sufficient privacy protection and impossible to realize dynamic privacy budget allocation during financial data privacy protection, which in turn leads to a decrease in the privacy protection adaptability of financial data privacy protection. Therefore, how to realize dynamic privacy budget allocation during financial data privacy protection and effectively improve the privacy protection adaptability during financial data privacy protection is a problem faced by the industry. Summary of the Invention

[0004] This application provides a financial data privacy protection method and system based on a large model, which can realize dynamic privacy budget allocation during financial data privacy protection and effectively improve the privacy protection adaptability during financial data privacy protection.

[0005] In a first aspect, the present application provides a financial data privacy protection method based on a large model, comprising the following steps: Obtain financial data in current financial trade; Classifying and screening the financial data to obtain a financial privacy data cluster, and determining the privacy sensitivity of each financial privacy data in the financial privacy data cluster; Obtaining historical privacy records of the financial privacy data cluster, determining the privacy loss of the financial privacy data cluster at different time intervals based on the historical privacy records, and performing association constraints on each privacy sensitivity and all privacy loss amounts to obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster; Determine the differential privacy constraints of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in current financial trade, and then determine the differential privacy index of the financial privacy data under the differential privacy constraints; Determining the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in historical financial trade by using the differential privacy index and each associated privacy loss; Dynamically protect financial data in current financial trade based on all dynamic privacy boundaries.

[0006] In some embodiments, the financial data is classified and screened to obtain a financial privacy data cluster, specifically including: filtering out sensitive information from the financial data; Identify the financial data according to the sensitive information to obtain a plurality of financial privacy data; All financial privacy data are classified to obtain financial privacy data clusters.

[0007] In some embodiments, determining the privacy sensitivity of each financial privacy data in the financial privacy data cluster specifically includes: Determining a privacy sensitivity standard for the financial privacy data cluster; A sensitivity evaluation is performed on each financial privacy data in the financial privacy data cluster according to the privacy sensitivity standard, thereby obtaining the privacy sensitivity of each financial privacy data in the financial privacy data cluster.

[0008] In some embodiments, determining the privacy loss amount of the financial privacy data cluster at different time intervals through the historical privacy records specifically includes: Determining a privacy status aggregation value of the financial privacy data cluster based on the historical privacy records; determining discrete values ​​of privacy status of the financial data cluster at different time intervals; The privacy loss amount of the financial privacy data cluster in different time intervals is determined according to the privacy state discrete value and the privacy state aggregate value.

[0009] In some embodiments, determining the differential privacy constraints of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in current financial trade specifically includes: Determining a relaxation factor of the financial privacy data cluster according to all associated privacy losses; Determining the differential privacy amount between adjacent data sets in the financial privacy data cluster according to the total privacy budget of the differential privacy model in current financial trade; The differential privacy constraint of the financial privacy data cluster is determined according to the relaxation factor and the differential privacy amount.

[0010] In some embodiments, determining the differential privacy index of the financial privacy data under the differential privacy constraint specifically includes: Determining the raw data utility of the financial privacy data; Determining the privacy protection amount of the financial privacy data under the differential privacy constraint; A differential privacy index of the financial privacy data under the differential privacy constraint is determined according to the original data utility and the privacy protection amount.

[0011] In some embodiments, financial data in current financial trades is obtained through a financial institution application programming interface and a market data application programming interface.

[0012] In a second aspect, the present application provides a financial data privacy protection system based on a large model, comprising: The acquisition module is used to obtain the financial data in the current financial trade; a processing module, configured to classify and screen the financial data to obtain a financial privacy data cluster, and determine the privacy sensitivity of each financial privacy data in the financial privacy data cluster; The processing module is further configured to obtain historical privacy records of the financial privacy data cluster, determine the privacy loss of the financial privacy data cluster at different time intervals based on the historical privacy records, and associate constraints between each privacy sensitivity and all privacy loss amounts to obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster; The processing module is further configured to determine the differential privacy constraints of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and further determine the differential privacy index of the financial privacy data under the differential privacy constraints; The processing module is further configured to determine the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in historical financial trades by using the differential privacy index and each associated privacy loss; An execution module is used to dynamically protect financial data in current financial trade based on all dynamic privacy boundaries.

[0013] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned large model-based financial data privacy protection method.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned large-model-based financial data privacy protection method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The large-scale model-based financial data privacy protection method and system provided herein first obtain financial data in current financial trade; then, classify and screen the financial data to obtain financial privacy data clusters, and determine the privacy sensitivity of each financial privacy data in the financial privacy data clusters; second, obtain historical privacy records of the financial privacy data clusters, determine the privacy loss of the financial privacy data clusters at different time intervals based on the historical privacy records, and associate constraints between each privacy sensitivity and all privacy losses to obtain the associated privacy loss of each financial privacy data in the financial privacy data clusters; third, determine the differential privacy constraints of the financial privacy data clusters based on all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and then determine the differential privacy index of the financial privacy data under the differential privacy constraints; then, determine the dynamic privacy bounds of each financial privacy data in the financial privacy data cluster in historical financial trades based on the differential privacy index and each associated privacy loss; finally, dynamically protect the financial data in the current financial trade based on all dynamic privacy bounds.

[0016] It can be seen that in this application, dynamic privacy budget allocation can be achieved during financial data privacy protection, thereby effectively improving the privacy protection adaptability during financial data privacy protection. First, the differential privacy constraints of the financial privacy data cluster are determined by all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade to ensure that each financial data cluster does not exceed the acceptable privacy leakage limit during the data privacy protection process, thereby protecting the privacy of the data subject; then, the degree of privacy protection paid for protecting specific financial privacy data is determined, and then the differential privacy index and each associated privacy loss are used to determine the historical financial privacy of each financial privacy data in the financial privacy data cluster. Dynamic privacy boundaries in trade can realize dynamic privacy budget allocation, which can flexibly adjust privacy protection strategies and improve the adaptability of privacy protection, thereby maximizing the practicality and utility of data while ensuring data privacy security; further accurately control privacy leakage risks, improve data utility, optimize resource allocation, enhance adaptability and flexibility; finally, dynamically protect financial data in current financial trade through dynamic privacy boundaries; in summary, the solution provided in this application can dynamically allocate privacy budgets for financial data when the privacy budget in the process of financial data privacy protection produces protection hysteresis, thereby improving the adaptability of privacy protection during financial data privacy protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is an exemplary flow chart of a method for protecting financial data privacy based on a large model according to some embodiments of the present application; Figure 2 is an exemplary schematic diagram of a differential privacy framework for financial data according to some embodiments of the present application; Figure 3 is an exemplary schematic diagram of a financial data privacy protection framework according to some embodiments of the present application; Figure 4 is a schematic diagram of exemplary hardware and / or software of a financial data privacy protection system based on a large model according to some embodiments of the present application; Figure 5 It is a structural diagram of a computer device for implementing a financial data privacy protection method based on a large model as shown in some embodiments of the present application. DETAILED DESCRIPTION

[0018] The core of this application is to obtain financial data from current financial trade; classify and filter the financial data to obtain financial privacy data clusters; determine the privacy sensitivity of each financial privacy data in the financial privacy data cluster; obtain historical privacy records for the financial privacy data cluster; determine the privacy loss of the financial privacy data cluster at different time intervals based on the historical privacy records; associate constraints between each privacy sensitivity and all privacy losses to obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster; and then determine the differential privacy constraints of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade; determine the differential privacy index of the financial privacy data under the differential privacy constraints; and determine the dynamic privacy bounds of each financial privacy data in the financial privacy data cluster in historical financial trade based on the differential privacy index and each associated privacy loss; and dynamically protect the financial data in the current financial trade based on all dynamic privacy bounds. This scheme protects the privacy of financial data based on the dynamic adaptive allocation of the privacy budget, which can improve the privacy protection adaptability of financial data privacy protection.

[0019] In order to better understand the above technical solution, the following will be combined with the accompanying drawings and specific implementation methods to describe the above technical solution in detail. Figure 1 , which is an exemplary flow chart of a method for protecting financial data privacy based on a large model according to some embodiments of the present application. The method 100 for protecting financial data privacy based on a large model mainly includes the following steps: In step 101, financial data in current financial trade is obtained.

[0020] In specific implementation, financial data in current financial trade can be obtained through the financial institution application programming interface and the market data application programming interface, historical transaction data, customer information, etc. can be extracted from the company's internal database, or through cooperation with other financial institutions, desensitized and encrypted transaction data can be shared. In other embodiments, web crawler technology can also be used to collect information and data related to financial trade from financial news websites, social media platforms, professional financial forums, etc. This method is suitable for obtaining unstructured data, such as news reports, market sentiment, etc., and will not be repeated here.

[0021] It should be noted that the financial data in current financial trade in this application refers to various data generated and used in real time by financial markets and financial institutions in daily operations and transactions. Among them, current financial trade refers to financial trade activities within one month, including but not limited to transaction data (such as trading records of stocks, bonds, foreign exchange, futures, etc.), market data (such as prices, trading volumes, buy and sell orders, etc. real-time market information), customer information (such as account information, trading behavior, credit records, etc.), financial data (such as balance sheets, income statements, etc.) and related market information and analysis reports. These data are of great value in financial decision-making, risk management, market analysis and compliance monitoring.

[0022] In step 102, the financial data is classified and screened to obtain a financial privacy data cluster, and the privacy sensitivity of each financial privacy data in the financial privacy data cluster is determined.

[0023] In some embodiments, the financial data may be classified and screened to obtain a financial privacy data cluster by using the following steps: filtering out sensitive information from the financial data; Identify the financial data according to the sensitive information to obtain a plurality of financial privacy data; All financial privacy data are classified to obtain financial privacy data clusters.

[0024] In the specific implementation, first, personal privacy or sensitive information contained in all fields of financial data is selected, such as name, ID number, address, contact information, account balance, and transaction records. In the selection process, predefined rules can be used to filter out data records containing these sensitive information, and the record is used as sensitive information in the financial data, wherein the predefined rules can be personal identity information identification rules, financial information identification rules, and behavioral data identification rules, etc., which are not limited here; then, according to the category of sensitive information, all financial data are traversed, and the financial data corresponding to the same type of sensitive information are combined together, and the financial data corresponding to the combined sensitive information are used as financial privacy data. For example: determine the personal identity information contained in the fields of financial data, such as name, ID number, address, contact information, etc., and use the personal information as the sensitive information of the financial data, and then traverse the data field names of the financial data containing "name", Keywords such as "ID number," "address," and "telephone number" are used, and string matching or regular expression technology is used to identify and extract records containing this information. The extracted records are combined together, and the combined information records are used as financial privacy data. Finally, some of the identified sensitive data records are randomly selected and manually checked to ensure the accuracy of the identification, or a verification algorithm is used to automatically verify the identification results to detect potential misjudgments or omissions. The verified sensitive data is then labeled with its category (such as personal information, financial information, behavioral data) and privacy sensitivity (high, medium, and low). The labeled sensitive data is classified and organized, that is, all financial privacy data is classified. After classification, different categories of financial privacy data clusters are formed. Each data cluster is stored and protected separately to ensure data privacy and security. In other embodiments, other methods can also be used to determine financial privacy data clusters, which will not be repeated here.

[0025] It should be noted that the sensitive information in the financial data in this application refers to information that can identify an individual or reveal an individual's financial status. If this information is leaked, it may affect personal privacy and security; financial privacy data refers to sensitive information generated, collected and stored in financial transactions and business processing, which can directly or indirectly identify an individual's identity, reveal the financial status, transaction behavior and credit record of an individual or institution; a financial privacy data cluster refers to a collection of multiple financial privacy data formed by classifying and screening financial data in financial data processing. Each data item in the financial privacy data cluster involves sensitive information of an individual or institution, such as identity information, financial status, transaction behavior, etc.

[0026] In some embodiments, determining the privacy sensitivity of each financial privacy data in the financial privacy data cluster may be achieved by using the following steps: Determining a privacy sensitivity standard for the financial privacy data cluster; A sensitivity evaluation is performed on each financial privacy data in the financial privacy data cluster according to the privacy sensitivity standard, thereby obtaining the privacy sensitivity of each financial privacy data in the financial privacy data cluster.

[0027] In a specific implementation, first, the sensitivity level of the financial privacy data is determined, and the sensitivity level of the financial privacy data is used as a privacy sensitivity standard. In other embodiments, other methods can also be used to determine the privacy sensitivity standard, which is not limited here. Generally, sensitivity can be divided into three categories: high, medium, and low. For example, high sensitivity: information that directly identifies an individual or is highly private, such as name, ID number, bank account number, credit card number, etc.; medium sensitivity: information that indirectly identifies an individual or information related to financial status but not completely private, such as transaction records, account balances, etc.; low sensitivity: information that is difficult to identify an individual or public information, such as summary statistics, market data, etc.; then, each financial privacy data in the financial privacy data cluster is evaluated according to the determined privacy sensitivity standard. And assign corresponding sensitivity labels, that is, the sensitivity on the sensitivity label is used as the privacy sensitivity of each financial data in the financial privacy data cluster. For example, high-sensitivity data evaluation: identification of key fields: such as name, ID number, bank account number, credit card number, etc., evaluation criteria: any data item containing these key fields is automatically classified as high sensitivity; medium-sensitivity data evaluation: identification of important financial information: such as transaction records, account balances, loan records, etc., evaluation criteria: this information usually cannot identify an individual's identity alone, but still involves personal financial privacy, and is classified as medium sensitivity; low-sensitivity data evaluation: identification of general information: such as summary statistics, market data, etc., evaluation criteria: this information does not involve personal privacy or is difficult to identify an individual, and is classified as low sensitivity, which will not be repeated here.

[0028] It should be noted that the privacy sensitivity standard of the financial privacy data cluster in this application represents the specification for evaluating and classifying the privacy sensitivity of data items in the financial privacy data cluster. The standard divides data into different sensitivity levels based on the nature and content of the data and its potential impact on the privacy of individuals or institutions to guide the implementation of data protection and privacy management measures. The standard usually includes three or more sensitivity levels of high, medium and low, and each level corresponds to different data protection needs and strategies; privacy sensitivity indicates the potential risk level to personal privacy or data security when financial data or information is accessed, used or disclosed. The privacy sensitivity measures the degree of privacy infringement that may be caused to individuals or organizations by the leakage of information. The level of privacy sensitivity determines the strictness of the protection measures that need to be taken for the data.

[0029] In step 103, historical privacy records of the financial privacy data cluster are obtained. The privacy loss amount of the financial privacy data cluster in different time intervals is determined based on the historical privacy records. Association constraints are performed on each privacy sensitivity and all privacy loss amounts to obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster.

[0030] In some embodiments, reference Figure 2 As shown, this figure is an exemplary schematic diagram of a differential privacy framework for financial data shown in some embodiments of the present application. In the differential privacy framework for financial data, the current privacy protection of financial data is mainly based on data perturbation, gradient perturbation and objective function perturbation methods to achieve differential privacy protection based on large models. Data perturbation directly adds noise to the data, which destroys the correlation between the data and inevitably leads to utility loss of the training data; in gradient perturbation, only the gradient of the discriminator is randomly perturbed by differential privacy noise, without destroying the training sample data and its correlation, nor destroying the objective function of the training. Because the gradient is randomly perturbed, it may take more iterations than data perturbation and objective function perturbation to make the objective function of differential privacy converge and stabilize. In addition, it is necessary to calculate the cumulative privacy budget loss estimate of the iterative training of the differential privacy framework to ensure that the training of the differential privacy framework is terminated after the cumulative privacy budget loss estimate is completely consumed.

[0031] In the specific implementation of this application, obtaining the historical privacy records of the financial privacy data cluster can be achieved in the following manner, namely: accessing the system and database that stores the financial privacy data records, which includes a data management system, a logging system, a compliance audit system, etc., and using query tools or scripts to extract the historical privacy records of each financial privacy data in the financial privacy data cluster from the relevant data source. In other embodiments, according to specific needs, records can be filtered by time range, data category or event type, and then historical records are extracted from the data source to ensure that the records include data access, processing, sharing and other privacy-related operation logs. These records include the time, user, operation type and sensitive data items involved in each operation. The extracted historical privacy records are then verified to ensure their accuracy and completeness, and the records are organized into a highly readable format, such as time series data, reports or audit logs. The organized privacy records are used as the historical privacy records of the financial privacy data cluster.

[0032] It should be noted that the historical privacy records of the financial privacy data cluster in this application represent detailed logs of all historical operations and events related to each financial privacy data in the financial privacy data cluster during the financial data management process. These records include the timestamps of operations such as data access, processing, sharing, modification, and deletion, the operator identity, operation type, and the specific data items involved. The records are used to review and verify the compliance of data processing and to ensure data privacy and security.

[0033] In some embodiments, determining the privacy loss amount of the financial privacy data cluster at different time intervals through the historical privacy records can be achieved by using the following steps: Determining a privacy status aggregation value of the financial privacy data cluster based on the historical privacy records; determining discrete values ​​of privacy status of the financial data cluster at different time intervals; The privacy loss amount of the financial privacy data cluster in different time intervals is determined according to the privacy state discrete value and the privacy state aggregate value.

[0034] In a specific implementation, determining the aggregate privacy status value of the financial privacy data cluster based on the historical privacy records can be achieved in the following manner: first, obtaining historical privacy records including logs of operations such as access, modification, and sharing; then identifying key privacy events from the historical privacy records, for example, operation types that affect the privacy status, such as data access, data modification, and data sharing; then, based on the historical privacy records, counting the number of privacy events at each time point; using a weighted or scoring mechanism, scoring the privacy events at each time point; and aggregating the privacy event scores at each time point to obtain an overall privacy status aggregate value. In other embodiments, other methods may also be used to determine the aggregate privacy status value of the financial privacy data cluster, for example, counting the total number of privacy events in all financial privacy data clusters and using this total as the aggregate privacy status value of the financial privacy data cluster. This will not be discussed in detail here. For example, within one month, the privacy event records of a financial institution are as follows: access to highly sensitive data (score 10): 50 times, access to moderately sensitive data (score 5): 100 times, modification of highly sensitive data (score 15): 20 times, modification of moderately sensitive data (score 10): 40 times. Calculate the privacy status aggregate value for that month: that is, privacy status aggregate value = (50 times * 10 points) + (100 times * 5 points) + (20 times * 15 points) + (40 times * 10 points), aggregate value = 500 + 500 + 300 + 400, privacy status aggregate value = 1700.

[0035] It should be noted that the privacy status aggregation value of the financial privacy data cluster in this application represents the result of a comprehensive assessment and quantification of the privacy status of all financial privacy data within a period of time, and is used to reflect the overall privacy risk level within that period of time.

[0036] In addition, in a specific implementation, determining the discrete values ​​of the privacy status of the financial data cluster at different time intervals can be achieved in the following manner, namely: dividing the timeline in the financial data privacy protection into a number of equally spaced time periods, such as by day, week, month, etc., within each time interval, counting the number and type of privacy events, and then calculating the privacy status value of each time interval, such as the sum or weighted average of the number of events, and then comparing the privacy status values ​​of adjacent time intervals, calculating the difference or change, and using the difference or change as the discrete value of the privacy status. The process of calculating the difference or change can use statistical methods such as standard deviation and variance to measure the discreteness; in other embodiments, the number of privacy events other than a certain privacy event can also be used. The sum of the quantities is used as the discrete value of the privacy status at different time intervals. For example, in addition to data access, the sum of events such as data modification and data sharing is not limited here. For example, if a month is divided into four weeks, the privacy status value of a financial institution in each week is as follows: Week 1: Privacy status value = 500, Week 2: Privacy status value = 700, Week 3: Privacy status value = 650, Week 4: Privacy status value = 800. The discrete value of the privacy status for each time interval is calculated as follows: the discrete value of Week 2 relative to Week 1 = |700-500| = 200, the discrete value of Week 3 relative to Week 2 = |650-700| = 50, and the discrete value of Week 4 relative to Week 3 = |800-650| = 150.

[0037] It should be noted that the discrete value of the privacy status in this application represents the degree to which the privacy status of financial privacy data changes within different time intervals. It reflects the fluctuation or discreteness of the privacy status and quantifies the changes in privacy risks over time. The discrete value of the privacy status measures the indicators of changes in privacy risks over time, which helps financial institutions to dynamically manage and adjust privacy protection strategies.

[0038] In addition, it should be noted that determining the amount of privacy loss of the financial privacy data cluster in different time intervals based on the privacy state discrete value and the privacy state aggregate value can be achieved in the following manner, namely: calculating the ratio of the privacy state discrete value and the privacy state aggregate value, and then determining the amount of privacy loss in each time interval based on the product of the calculation result and the sensitivity coefficient, that is: taking the product of the calculation result and the sensitivity coefficient as the amount of privacy loss in each time interval, for example: privacy loss = (privacy state discrete value / privacy state discrete value) * sensitivity coefficient, where the sensitivity coefficient is a numerical indicator used to quantify and measure the impact of a certain data item or operation on overall privacy protection. It reflects the sensitivity of the data and its importance in the privacy protection strategy. It can be obtained through expert experience and will not be repeated here.

[0039] It should be noted that the amount of privacy loss in this application represents a quantitative indicator of the privacy risk or degree of privacy leakage caused by data access, processing, sharing, or leakage operations within a predetermined time period. This indicator is used to evaluate the effectiveness of privacy protection measures and the level of privacy risk management.

[0040] In some embodiments, the following steps may be used to associate constraints between each privacy sensitivity and all privacy loss amounts to obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster: Obtain the privacy sensitivity of all financial privacy data; Determining the global sensitivity of the financial privacy data in privacy protection according to the privacy sensitivity; Determine the privacy constraint gradient of financial privacy data in privacy protection based on all privacy loss amounts; The global sensitivity is mapped to the privacy constraint gradient to obtain the associated privacy loss of all financial privacy data in the financial privacy data cluster.

[0041] In a specific implementation, first, the global sensitivity is determined based on the average privacy sensitivity of all financial privacy data. That is, the privacy sensitivities of all financial privacy data are averaged, and the result of the average is used as the global sensitivity. The average privacy sensitivity of all financial privacy data = (the sum of the sensitivities of all data items) / (the number of data items). In other embodiments, other methods can also be used to determine the global sensitivity, which is not limited here. Then, the privacy loss of all data items in different time periods is obtained, and the maximum privacy loss is found. The privacy constraint gradient is determined based on the privacy loss of each data item and the maximum privacy loss. The privacy constraint gradient can be determined in the following manner, namely: privacy constraint gradient = privacy loss / maximum privacy loss. Finally, the global sensitivity is mapped to the privacy constraint gradient, that is, a budget weight is assigned to each financial privacy data item based on the global sensitivity and the privacy constraint gradient. The process of assigning the budget weight can be implemented in the following manner: first, the budget weight is calculated, for example: budget weight = global sensitivity × privacy constraint gradient, and then the associated privacy loss is allocated based on the budget weight, for example: the associated privacy loss of each financial privacy data item = total budget × budget weight / total budget weight. In other embodiments, other methods can also be used to determine the associated privacy loss of all financial privacy data, for example, using an adaptive privacy budget allocation algorithm or a differential privacy algorithm for high-dimensional data release to determine the associated privacy loss of all financial privacy data, which is not limited here.

[0042] It should be noted that in this application, global sensitivity refers to the sensitivity index of financial privacy data when privacy protection is performed, which is used to measure the sensitivity of the entire data set (such as a financial privacy data cluster) in privacy protection; the privacy constraint gradient represents the priority and resource requirements of financial privacy data in privacy protection, and the privacy constraint gradient reflects the protection intensity or attention required for different data items in the privacy protection process; the associated privacy loss represents a parameter that quantifies the degree of privacy protection, and the associated privacy loss is used to measure the acceptable degree of information leakage in the process of financial data privacy protection, ensuring that a suitable point is found between balancing the usefulness of the data and privacy protection. A smaller associated privacy loss provides stronger privacy protection, but may reduce the accuracy of data analysis, while a larger budget allows more precise analysis results but a relatively low level of privacy protection.

[0043] In step 104, the differential privacy constraints of the financial privacy data cluster are determined by all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and then the differential privacy index of the financial privacy data under the differential privacy constraints is determined.

[0044] In some embodiments, reference Figure 3 As shown, this figure is an exemplary schematic diagram of the financial data privacy protection framework shown in some embodiments of the present application. Traditional differential privacy technology requires a trusted circuit to centralize the original data to a data center for processing, but in actual applications, it is difficult to find a trusted third party, which greatly limits the application of this differential privacy technology; in the financial data privacy protection framework, the data collection center based on the large model, in the scenario of untrusted third-party data collectors, inherits the traditional differential privacy protection and refines the protection of user-sensitive financial trade privacy data. The financial trade data owned can be privatized on the local side to avoid privacy attacks that may be caused by untrusted third-party data collectors.

[0045] In financial data privacy protection based on large models, the noise mechanism is always the core of achieving differential privacy protection. In differential privacy protection, as the privacy budget decreases, the disturbance to the data becomes greater and the privacy protection of the data becomes stronger, but at the same time the availability of the data also deteriorates. Balancing data privacy protection and the high quality of its protection becomes a problem that needs to be solved. In privacy protection, anonymization technology is prone to leakage when responding to complex attacks and affects the availability of published data. Encryption technology consumes a lot of system resources, generates a large computing overhead, and is difficult to implement. At the same time, once the attacker obtains the key during data sharing, it will cause complete leakage of the data. However, the financial data privacy protection framework based on large models has a strict privacy definition and can resist attacks with arbitrary background knowledge. It is necessary to control the privacy budget well, otherwise it may cause serious distortion of financial data and invalidate the availability of financial data.

[0046] In some embodiments, determining the differential privacy constraints of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in current financial trade can be achieved by using the following steps: Determining a relaxation factor of the financial privacy data cluster according to all associated privacy losses; Determining the differential privacy amount between adjacent data sets in the financial privacy data cluster according to the total privacy budget of the differential privacy model in current financial trade; The differential privacy constraint of the financial privacy data cluster is determined according to the relaxation factor and the differential privacy amount.

[0047] In a specific implementation, first, the associated privacy loss of each data item in all financial privacy data clusters is read, and then the relaxation factor of each data item is calculated to obtain the relaxation factor of the financial privacy data cluster. The relaxation factor can be calculated in the following way: the quotient of the sum of the associated privacy losses of all data items divided by the total number of data items is calculated, and the quotient is used as the relaxation factor of the financial privacy data cluster. For example, if the associated privacy loss of data item A is 0.4, the associated privacy loss of data item B is 0.3, and the associated privacy loss of data item C is 0.3, then the relaxation factor of the financial privacy data cluster is (0.4+0.3+0.3) / 3≈0.33. In other embodiments, other methods can also be used to determine the relaxation factor, which is not used here. I will elaborate on this; then, the total privacy budget of the differential privacy model in the current financial trade is determined based on business needs analysis, legal and compliance requirements, data sensitivity assessment, historical data leakage incidents, and the cost of privacy protection technology. The total privacy budget is then divided by the sum of the number of data items, and the quotient obtained is used as the differential privacy amount. In other embodiments, convolutional neural networks, long short-term memory networks, convolutional long short-term memory networks and other models can also be used to calculate the differential privacy amount, which is not limited here; finally, the relaxation factor and the differential privacy amount are combined to determine the differential privacy constraint of each data item, that is, the relaxation factor and the differential privacy amount are multiplied, and the product result is used as the differential privacy constraint of the financial privacy data cluster, which will not be elaborated on here.

[0048] It should be noted that the relaxation factor of the financial privacy data cluster in this application represents the coefficient used to measure the amount of additional privacy leakage that can be tolerated in the process of allocating associated privacy losses in financial data privacy protection. In the privacy protection of financial data based on large models, the privacy protection intensity can be adjusted and the budget allocation can be optimized. The differential privacy amount between adjacent data sets in the financial privacy data cluster represents the maximum privacy leakage allowed by the differential privacy mechanism between two adjacent data sets (i.e., data sets that differ by only one data item). In the privacy protection process, it measures the similarity of the analysis results of two adjacent data sets. In the privacy protection process, privacy leakage can be measured in the financial data privacy protection based on large models. Risk; The differential privacy constraint of the financial privacy data cluster means that in the process of financial data privacy protection, restrictions are imposed on the financial privacy data cluster through the differential privacy mechanism to ensure that the privacy leakage of each data item does not exceed the predetermined privacy budget when processing and analyzing data. This constraint ensures that when using a large model for data privacy protection, even if a single data item changes, the change of the overall analysis result is strictly controlled, thereby protecting individual privacy. That is, the differential privacy constraint sets the boundary of privacy protection. By controlling the amount of privacy leakage, it ensures that each financial data cluster will not exceed the acceptable privacy leakage limit during the data privacy protection process, thereby protecting the privacy of the data subject.

[0049] In some embodiments, determining the differential privacy index of the financial privacy data under the differential privacy constraint may be achieved by using the following steps: Determining the raw data utility of the financial privacy data; Determining the privacy protection amount of the financial privacy data under the differential privacy constraint; A differential privacy index of the financial privacy data under the differential privacy constraint is determined according to the original data utility and the privacy protection amount.

[0050] In the specific implementation, first, collect and organize the original financial data set that has not been processed for privacy protection, apply the large model to the original data set for data analysis, record the analysis results, and calculate the baseline utility value of the original data set based on the selected utility indicators, such as the accuracy in the classification task, the mean square error in the regression task, etc., and use the baseline utility value as the utility of the original data. Among them, the choice of utility indicators can be classification accuracy, mean square error, F1 score, etc., which will not be repeated here; then, the Laplace mechanism can be used to add Laplace noise to the financial privacy data, so that the output difference between the two adjacent data sets is within the parameter control range of differential privacy protection, among which differential privacy protection The parameters of can be determined by machine learning or other methods, and then the privacy protection amount of the financial privacy data under the differential privacy constraint is determined by linear regression, which will not be repeated here. Finally, the differential privacy index of the financial privacy data under the differential privacy constraint is determined based on the difference between the original data utility of the financial privacy data and the privacy protection amount of the financial privacy data under the differential privacy constraint. That is, the difference between the original data utility of the financial privacy data and the privacy protection amount of the financial privacy data under the differential privacy constraint is used as the differential privacy index of the financial privacy data under the differential privacy constraint. In other embodiments, other methods can also be used to determine the differential privacy index of the financial privacy data under the differential privacy constraint, which will not be repeated here.

[0051] It should be noted that the raw data utility of financial privacy data in this application represents the effectiveness and information value of the data in actual analysis and application without any privacy protection measures. The raw data utility measures the performance indicators of the data when meeting the analysis objectives (such as prediction, classification, clustering, etc.); the privacy protection amount of financial privacy data under differential privacy constraints represents the privacy protection strength on two adjacent data sets (that is, data sets with only one data item different). This privacy protection amount measures the privacy protection strength, determines the associated privacy loss, and optimizes the balance between privacy and data utility; the differential privacy index represents the effectiveness and information value of the data in actual analysis and application after applying privacy protection measures (such as differential privacy).

[0052] In step 105 , the dynamic privacy limit of each financial privacy data in the financial privacy data cluster in historical financial trade is determined by using the differential privacy index and each associated privacy loss.

[0053] In some embodiments, determining the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in historical financial trade by using the differential privacy index and each associated privacy loss can be achieved by using the following steps: Obtaining the raw data utility of financial privacy data; Obtaining a differential privacy index of the financial privacy data under the differential privacy constraint; determining a utility loss based on the original data utility and the differential privacy index; The dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in historical financial trade is determined according to the utility loss and each associated privacy loss.

[0054] In specific implementation, first, the utility loss is determined based on the original data utility and the differential privacy index. This can be done by subtracting the original data utility from the differential privacy index and using the difference as the utility loss. In other embodiments, other methods can also be used to determine the utility loss, which will not be repeated here. Then, the dynamic privacy bound of each data item under privacy protection is calculated, that is, the utility loss is ratioed with each associated privacy loss. Finally, the obtained ratio is used as the dynamic privacy bound of each financial privacy data in the financial privacy data cluster in historical financial trade. This will not be repeated here.

[0055] It should be noted that the utility loss in this application represents the reduction in data analysis or application effects due to the application of privacy protection measures (such as differential privacy). This utility loss is used to measure the degree of performance degradation of financial data in analysis tasks (such as prediction, classification, regression, etc.) after privacy protection compared to the original data; the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in historical financial trade represents the degree of privacy protection paid to protect specific financial privacy data in historical financial trade, which is used to measure the degree of data utility degradation, analysis accuracy reduction, etc. caused by ensuring data privacy (by applying privacy protection measures).

[0056] In step 106 , the financial data in the current financial trade is dynamically protected based on all dynamic privacy boundaries.

[0057] In specific implementation, dynamic protection of financial data in current financial trade based on all dynamic privacy boundaries can be achieved in the following way: first, calculate the privacy quantity index of each data cluster to quantify its privacy leakage risk and protection effect; then, based on these indicators, dynamically adjust the privacy budget allocation to ensure that highly sensitive data receives more protection while ensuring the actual utility of the data. Optimization algorithms, such as linear programming algorithms, can be used to reallocate the privacy budget by minimizing the overall privacy leakage risk; finally, use linear programming or other optimization algorithms to adjust the privacy budget of each data cluster based on the privacy quantity index. The new privacy budget should reduce the privacy leakage risk of highly sensitive data while meeting the total privacy budget. The adjusted privacy budget is applied to the privacy protection process of financial data to ensure a balance between privacy protection and data utility, thereby achieving dynamic protection of financial data in current financial trade.

[0058] In addition, in another aspect of the present application, in some embodiments, the present application provides a financial data privacy protection system based on a large model, referring to Figure 4 This figure is a schematic diagram of exemplary hardware and / or software of a large-model-based financial data privacy protection system according to some embodiments of the present application. The large-model-based financial data privacy protection system 400 includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described as follows: Acquisition module 401, in this application, acquisition module 401 is mainly used to acquire financial data in current financial trade; Processing module 402, in this application, is mainly used to classify and filter the financial data to obtain financial privacy data clusters, and determine the privacy sensitivity of each financial privacy data in the financial privacy data clusters; The processing module 402 is further configured to obtain historical privacy records of the financial privacy data cluster, determine the privacy loss of the financial privacy data cluster at different time intervals based on the historical privacy records, and associate constraints between each privacy sensitivity and all privacy loss amounts to obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster; The processing module 402 is further configured to determine the differential privacy constraints of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and further determine the differential privacy index of the financial privacy data under the differential privacy constraints; The processing module 402 is further configured to determine the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in historical financial trades by using the differential privacy index and each associated privacy loss; Execution module 403: In this application, execution module 403 is mainly used to dynamically protect financial data in current financial trade based on all dynamic privacy boundaries.

[0059] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory storing code, and the processor being configured to obtain the code and execute the above-mentioned large model-based financial data privacy protection method.

[0060] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing a financial data privacy protection method based on a large model according to some embodiments of the present application. The financial data privacy protection method based on a large model in the above embodiment can be achieved by Figure 5 The computer device 500 shown in FIG. 5 is implemented as shown in FIG. 5 . The computer device 500 includes at least one processor 501 , a communication bus 502 , a memory 503 , and at least one communication interface 504 .

[0061] The processor 501 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more processors for controlling the execution of the large-model-based financial data privacy protection method in this application.

[0062] The communication bus 502 may be used to transmit information between the aforementioned components.

[0063] The memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory 503 may be independent and connected to the processor 501 via the communication bus 502. The memory 503 may also be integrated with the processor 501.

[0064] Memory 503 is used to store program code for executing the solution of the present application, and is controlled by processor 501 for execution. Processor 501 is used to execute the program code stored in memory 503. The program code may include one or more software modules. The financial data privacy protection method based on the large model in the above embodiment can be implemented by processor 501 and one or more software modules in the program code in memory 503.

[0065] The communication interface 504 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0066] In a specific implementation, as an example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0067] The aforementioned computer device can be a general-purpose computer device or a dedicated computer device. In a specific implementation, the computer device can be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of this application do not limit the type of computer device.

[0068] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned large model-based financial data privacy protection method.

[0069] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0070] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A financial data privacy protection method based on a large model, characterized in that: The steps include: Obtain financial data in current financial trade; Classifying and screening the financial data to obtain a financial privacy data cluster, and determining the privacy sensitivity of each financial privacy data in the financial privacy data cluster; Acquire historical privacy records of the financial privacy data cluster, determine the privacy loss amount of the financial privacy data cluster in different time intervals through the historical privacy records, perform association constraints on each privacy sensitivity and all privacy loss amounts, and obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster; Determine the differential privacy constraint of the financial privacy data cluster by all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and then determine the differential privacy index of the financial privacy data under the differential privacy constraint; Determine the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in the historical financial trade through the differential privacy index and each associated privacy loss; Dynamically protect financial data in current financial trade based on all dynamic privacy boundaries.

2. The method according to claim 1, characterized in that The financial data is classified and screened to obtain a financial privacy data cluster including: filtering out sensitive information from said financial data; Identify the financial data according to the sensitive information to obtain a plurality of financial privacy data; All financial privacy data are classified to obtain financial privacy data clusters.

3. The method according to claim 1, characterized in that Determining the privacy sensitivity of each financial privacy data in the financial privacy data cluster specifically includes: Determining a privacy sensitivity standard for the financial privacy data cluster; A sensitivity evaluation is performed on each financial privacy data in the financial privacy data cluster according to the privacy sensitivity standard, thereby obtaining the privacy sensitivity of each financial privacy data in the financial privacy data cluster.

4. The method according to claim 1, characterized in that Determining the privacy loss amount of the financial privacy data cluster in different time intervals through the historical privacy records specifically includes: Determine the privacy status aggregation value of the financial privacy data cluster according to the historical privacy records; Determining discrete values ​​of privacy status of the financial data cluster at different time intervals; The privacy loss amount of the financial privacy data cluster in different time intervals is determined according to the privacy state discrete value and the privacy state aggregate value.

5. The method according to claim 1, characterized in that The differential privacy constraints of the financial privacy data cluster are determined by all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and specifically include: Determining a relaxation factor of the financial privacy data cluster according to all associated privacy losses; Determine the differential privacy amount between adjacent data sets in the financial privacy data cluster according to the total privacy budget of the differential privacy model in the current financial trade; The differential privacy constraint of the financial privacy data cluster is determined according to the relaxation factor and the differential privacy amount.

6. The method according to claim 1, characterized in that Determining the differential privacy index of the financial privacy data under the differential privacy constraint specifically includes: Determining the raw data utility of the financial privacy data; Determining the privacy protection amount of the financial privacy data under the differential privacy constraint; A differential privacy index of the financial privacy data under the differential privacy constraint is determined according to the original data utility and the privacy protection amount.

7. The method according to claim 1, characterized in that Access financial data from current financial trades through the Financial Institutions API and Market Data API.

8. A financial data privacy protection system based on a large model, characterized in that: include: The acquisition module is used to obtain the financial data in the current financial trade; A processing module, configured to classify and screen the financial data to obtain a financial privacy data cluster, and determine the privacy sensitivity of each financial privacy data in the financial privacy data cluster; The processing module is further configured to obtain historical privacy records of the financial privacy data cluster, determine the privacy loss amount of the financial privacy data cluster in different time intervals through the historical privacy records, associate constraints between each privacy sensitivity and all privacy loss amounts, and obtain the associated privacy loss of each financial privacy data in the financial privacy data cluster; The processing module is further used to determine the differential privacy constraint of the financial privacy data cluster based on all associated privacy losses and the total privacy budget of the differential privacy model in the current financial trade, and then determine the differential privacy index of the financial privacy data under the differential privacy constraint; The processing module is further used to determine the dynamic privacy boundary of each financial privacy data in the financial privacy data cluster in the historical financial trade through the differential privacy index and each associated privacy loss; An execution module is used to dynamically protect financial data in current financial trade based on all dynamic privacy boundaries.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores codes, and the processor is configured to obtain the codes and execute the financial data privacy protection method based on a large model as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the financial data privacy protection method based on a large model as described in any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • LBS service privacy protection method based on differential privacy

    CN111797433A

  • Differential privacy protection clustering method based on K-means

    CN112199722A

  • Internet of vehicles privacy data protection method based on block chain and differential privacy

    CN113609523A

  • Utility-enhanced conditional feature selection differential privacy data publishing method

    CN116186757A

  • Multi-party information security calculation method and system based on differential privacy

    CN117874821A

Cited By

  • Data encryption method for realizing multi-party secure sharing of financial data

    CN119670152A

  • A data encryption method for realizing multi-party secure sharing of financial data

    CN119670152B

  • Software development application data processing method based on AI large model

    CN120086878A

  • A data processing method for software development application based on large AI models

    CN120086878B

  • Method and system for estimating economic data according to bank historical investment data

    CN120162519A