A cross-domain multi-dimensional data trusted sharing processing method and platform

Abstract

The application relates to a cross-domain multi-dimensional data trusted sharing processing method and platform, and relates to the field of data processing. The method comprises the following steps: acquiring a cross-domain multi-dimensional data set and performing random trusted desensitization processing; combining data correlation degree analysis desensitization utility and data utility loss; calculating trusted processing fitness; and optimizing an optimal desensitization scheme. The technical problem that privacy protection and data utility maintenance cannot be considered at the same time when cross-domain multi-dimensional data is shared, and the data structure and meaning are damaged after desensitization due to neglect of data correlation is solved. Cross-domain data can be kept at a high data utility while protecting privacy, trusted data sharing is promoted, data security and utility are balanced, and the technical effect of maintaining data correlation is achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a cross-domain, multi-dimensional data trusted sharing processing method and platform. Background Technology

[0002] In today's digital age, data has become a key resource driving socio-economic development and technological progress. With the rapid development of information technology and the widespread adoption of internet applications, massive amounts of multidimensional data have been generated across different fields and industries. This data contains immense value and is of great significance to scientific research, business decision-making, public services, and many other areas. However, due to issues such as data privacy protection, data security, and data ownership, cross-domain data sharing has always faced numerous challenges.

[0003] Traditional data sharing methods often focus on simple data exchange or copying, neglecting privacy protection and utility maintenance during the sharing process. Especially in cross-domain data sharing scenarios, data from different domains have varying degrees of sensitivity and sharing value, making it difficult for traditional anonymization methods to balance data security and utility. On the one hand, excessive anonymization may cause data to lose its original value, failing to meet sharing needs; on the other hand, insufficient anonymization may lead to data leakage risks, threatening personal privacy and corporate security. Existing anonymization technologies are mostly designed for single-domain or single-type data, lacking comprehensive consideration of cross-domain and multi-dimensional data. In cross-domain data sharing, complex relationships often exist between data from different domains, which are crucial for data understanding and application. However, existing anonymization methods often ignore these relationships, making it difficult for anonymized data to maintain its original structure and meaning in cross-domain applications. Summary of the Invention

[0004] This invention addresses the technical problems in existing technologies that make it difficult to balance privacy protection and data utility maintenance when sharing multidimensional data across different domains, and that neglecting data correlation leads to damage to the data structure and meaning after de-identification. It provides a cross-domain multidimensional data trusted sharing processing method and platform to solve these problems.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] In a first aspect, the present invention provides a cross-domain multidimensional data trusted sharing processing method, the method comprising: acquiring multiple datasets of multidimensional data to be shared across domains; performing random trusted de-identification processing on the multiple datasets to obtain a first trusted processing scheme; performing de-identification utility analysis on the first trusted processing scheme based on the correlation between the multidimensional data to obtain a first de-identification utility parameter, wherein the de-identification utility analysis includes de-identification utility loss analysis; performing data utility loss analysis on the first trusted processing scheme based on the multidimensional data to obtain a first data utility loss parameter; calculating a first trusted processing fitness based on the first trusted processing fitness; optimizing the first trusted processing scheme based on the first trusted processing fitness to obtain an optimal trusted processing scheme; performing trusted de-identification processing on the multiple datasets to achieve cross-domain trusted data sharing.

[0007] Secondly, this invention provides a cross-domain multidimensional data trusted sharing processing platform, the platform comprising: a data desensitization module, used to acquire multiple datasets of multidimensional data to be shared across domains, and to perform random trusted desensitization processing on the multiple datasets to obtain a first trusted processing scheme; a utility analysis module, used to perform desensitization utility analysis on the first trusted processing scheme based on the correlation between the multidimensional data to obtain a first desensitization utility parameter, wherein the desensitization utility analysis includes desensitization utility loss analysis; a loss analysis module, used to perform data utility loss analysis on the first trusted processing scheme based on the multidimensional data to obtain a first data utility loss parameter, and to calculate a first trusted processing fitness based on the first desensitization utility parameter; and a trusted sharing module, used to optimize the first trusted processing scheme based on the first trusted processing fitness to obtain an optimal trusted processing scheme, and to perform trusted desensitization processing on the multiple datasets for cross-domain trusted data sharing.

[0008] The beneficial effects of this invention are as follows: by acquiring cross-domain multidimensional datasets and performing random and trustworthy de-identification processing, combining data correlation analysis to determine the de-identification utility and data utility loss, calculating the trustworthy processing fitness, and optimizing to obtain the optimal de-identification scheme, this invention achieves the technical effect of maintaining high data utility while protecting privacy of cross-domain data, promoting trustworthy data sharing, balancing data security and utility, and maintaining data correlation. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a cross-domain, multi-dimensional, trusted data sharing processing method provided by the present invention.

[0010] Figure 2 This is a schematic diagram of the structure of a cross-domain, multi-dimensional data trusted sharing and processing platform provided by the present invention.

[0011] Figure labeling: Data anonymization module 11, utility analysis module 12, loss analysis module 13, trusted sharing module 14. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] Example 1:

[0016] like Figure 1 As shown, this embodiment of the invention provides a cross-domain, multi-dimensional data trusted sharing processing method, the method comprising:

[0017] S10: Obtain multiple datasets of multidimensional data to be shared across domains, perform random and trustworthy desensitization processing on the multiple datasets, and obtain a first trustworthy processing scheme.

[0018] For example, in trusted data sharing, multiple datasets of multidimensional data to be shared across different domains are first acquired. These datasets may cover different domains, such as healthcare, finance, and education, and each dataset contains various types of data, such as sensitive information like age, salary, and education level. Subsequently, randomized trusted de-identification processing is implemented on these datasets. This step focuses on the randomness and trustworthiness of the de-identification processing parameters. Specifically, the de-identification processing parameters can be set as a floating range for sensitive fields such as age and salary, for example, using different de-identification processing margins such as X±10% or ±20%, to ensure that the data maintains a certain level of authenticity while effectively preventing the direct leakage of sensitive information. By randomly generating multiple first de-identification processing parameters for the multidimensional data within this parameter space, a first trusted processing scheme is constructed. This scheme aims to balance the needs of data privacy protection and data utility maintenance, laying the foundation for subsequent data sharing. For example, in a medical data sharing scenario, the age field may be randomly adjusted to ±10% of the actual age, protecting patient privacy while preserving the data's usability in disease analysis.

[0019] S20: Based on the correlation between the multidimensional data, perform a desensitization utility analysis on the first trusted processing scheme to obtain a first desensitization utility parameter, wherein the desensitization utility analysis includes a desensitization utility loss analysis.

[0020] Preferably, a desensitization utility analysis is performed on the initially generated first trusted processing scheme, focusing on the correlation between multidimensional data. This aims to ensure the consistency of related data during the desensitization process, preventing the risk of inferring original sensitive information from partial data. Correlation reflects the tightness of the inherent connection between data and is an indispensable factor in analyzing desensitization utility. Specifically, desensitization utility analysis includes desensitization utility loss analysis, which assesses the impact of desensitization on data correlation. For example, in a dataset containing two sensitive fields, age and salary, if the age field is desensitized by ±5%, while the salary field is desensitized by 20%, this differentiated desensitization strategy may disrupt the correlation between data. This could allow attackers to infer specific individuals from minor changes in the age field, thus threatening data privacy. Therefore, calculating the desensitization utility parameter can quantify the degree of this correlation disruption, providing a basis for subsequent optimization of the desensitization strategy. In practice, the calculation of the desensitization utility parameter takes into account the correlation between various data fields and the selection of desensitization processing parameters to ensure that the related data can still maintain the necessary ambiguity after desensitization, while maximizing the preservation of the shared value of the data.

[0021] S30: Based on the multidimensional data, perform data utility loss analysis on the first trusted processing scheme to obtain the first data utility loss parameter, and calculate the first trusted processing fitness by combining the first desensitization utility parameter.

[0022] Furthermore, performing data utility loss analysis on the generated first trusted processing scheme is a crucial step in optimizing the anonymization strategy. This step focuses on assessing the impact of anonymization on the reference value of multidimensional data, i.e., quantifying the degree of utility loss after data anonymization. Data utility loss parameters embody the core indicator of this assessment process, reflecting the attenuation of information content and analytical value of the data after anonymization compared to the original data. Specifically, this analysis process simulates the data state after anonymization for each data category in the multidimensional data, combined with the anonymization parameters set in the first trusted processing scheme, such as ±5% fluctuation for the age field or a 20% adjustment for the salary field. Subsequently, a pre-built data utility loss analyzer is used to input these simulated data and output the corresponding data utility loss parameters. These parameters not only consider the utility changes of individual data fields but also integrate the impact of the correlation between data fields on the overall data utility.

[0023] After obtaining the first data utility loss parameter, it needs to be combined with the previously calculated first de-identification utility parameter to comprehensively evaluate the adaptability of the first trusted processing scheme. The first trusted processing fitness, as the result of this comprehensive evaluation, measures the de-identification scheme's ability to balance protecting data privacy with maintaining data utility. For example, if a de-identification scheme performs well in protecting age privacy (high de-identification utility parameter), but simultaneously causes a significant decrease in the analytical value of wage data (high data utility loss parameter), its overall fitness may be low. This calculation process can identify shortcomings in the de-identification scheme and provide direction for subsequent optimization, such as adjusting de-identification processing parameters to reduce data utility loss, or improving the de-identification strategy to better balance privacy protection and data utility.

[0024] S40: Optimize the first trusted processing scheme according to the first trusted processing fitness to obtain the optimal trusted processing scheme, perform trusted desensitization processing on the multiple datasets, and perform trusted cross-domain data sharing.

[0025] Specifically, systematically optimizing the initially generated trusted processing scheme is a core step in achieving efficient data sharing. This stage focuses on iteratively adjusting the anonymization parameters to explore and determine the optimal trusted processing scheme that strikes the best balance between protecting data privacy and maintaining data utility. Optimization aims to reduce the loss of data utility while ensuring that the anonymized data still meets the needs of cross-domain sharing. Specifically, the optimization process involves applying different combinations of anonymization parameters to multiple datasets, such as adjusting the anonymization margin of the age field to ±3% or ±7%, or changing the anonymization ratio of the salary field to 15% or 25%, to evaluate the combined impact of different parameter settings on data utility and anonymization effectiveness.

[0026] By constructing an evaluation model and combining the obtained data utility loss parameter and desensitization utility parameter, the fitness of each trusted processing scheme is calculated, thereby selecting the scheme with the highest fitness as the optimal trusted processing scheme. For example, in data sharing scenarios in the medical and financial fields, if a scheme protects patient age privacy while only slightly affecting the analytical value of medical cost data, and its overall fitness is better than other schemes, then this scheme will be selected as the optimal scheme. Subsequently, this optimal scheme is used to perform trusted desensitization processing on multiple datasets, ensuring that the desensitized data can effectively protect personal privacy while providing valuable information support to the recipient during cross-domain sharing, ultimately achieving secure and efficient cross-domain data sharing.

[0027] In a preferred embodiment, multiple datasets of multidimensional data to be shared across domains are acquired, and random trusted de-identification processing is performed on the multiple datasets to obtain a first trusted processing scheme. The process includes: acquiring multiple datasets of multidimensional data to be shared across domains; acquiring de-identification processing parameter ranges for the multidimensional data to be de-identified, forming a trusted de-identification processing scheme space; and randomly generating multiple first de-identification processing parameters for the multidimensional data within the trusted de-identification processing scheme space to obtain a first trusted processing scheme.

[0028] Optionally, multiple datasets of multidimensional data to be shared across different domains can be acquired. These datasets may originate from various fields, such as healthcare, financial services, and education and research, and each dataset may contain multiple types of data, such as sensitive information like age, income, and education level. Subsequently, for each type of data, a parameter range for data anonymization is determined. These parameter ranges form the basis of a space for trusted data anonymization schemes. The data anonymization parameter range refers to the acceptable range of anonymization for different data types to protect data privacy. For example, the age field might be set to a floating range of ±5% or ±10%, while the income field might use methods such as scaling or range fuzzification.

[0029] After defining the anonymization parameter ranges for each data type, these parameters are combined to form a multi-dimensional space of reliable anonymization schemes, encompassing all possible combinations of anonymization parameters. Next, multiple initial anonymization parameter combinations are randomly generated within this scheme space, each representing a potential anonymization strategy. This random generation process yields an initial reliable processing scheme, providing a foundation for subsequent data utility analysis, anonymization utility analysis, and optimization. For example, in a cross-domain data sharing scenario involving medical and financial data, a set of anonymization parameters might be randomly generated, where the age field is anonymized by ±7%, and the income field is scaled to 80%-90% of its original value. This serves as a preliminary reliable processing scheme for evaluation and optimization.

[0030] In a preferred embodiment, the first trusted processing scheme is subjected to desensitization utility analysis based on the correlation between the multidimensional data to obtain a first desensitization utility parameter. This includes: analyzing the correlation between data categories of the multidimensional data based on cross-domain data sharing records within a historical period to obtain multiple correlation degrees; calculating the absolute difference magnitude of the first desensitization processing parameters between every two data categories based on the multiple first desensitization processing parameters within the first trusted processing scheme to obtain multiple first desensitization utility loss parameters; and performing a weighted calculation on the multiple first desensitization utility loss parameters based on the multiple correlation degrees to obtain a first fused desensitization utility loss parameter, and then calculating the first desensitization utility parameter.

[0031] Specifically, in trusted data sharing, a desensitization utility analysis is performed on the generated first trusted processing scheme to evaluate its effectiveness. This process first analyzes the correlation between different data categories in multidimensional data based on historical cross-domain data sharing records. These correlations reflect the inherent connections and mutual influence between the data. For example, there are significant correlations between age and disease type in medical data, and between income and credit score in financial data.

[0032] After obtaining multiple correlation degrees, the absolute difference magnitude of the desensitization parameters between each pair of data categories is calculated by combining multiple first desensitization parameters from the first trusted processing scheme. The absolute difference magnitude refers to the quantitative expression of the difference between the desensitization parameters of two data categories. For example, if the desensitization magnitude of one data category is 20%, and that of the other is 5%, then their absolute difference magnitude is calculated as (20% - 5%) / 20% = 75%. This value serves as a first desensitization utility loss parameter between these two data categories, reflecting the potential degree of damage to the correlation between data categories caused by the desensitization process.

[0033] Subsequently, the first de-identification utility loss parameter is weighted and calculated based on multiple obtained correlations to comprehensively consider the impact of the strength of correlations between different data categories on the overall de-identification utility, thus obtaining the first fusion de-identification utility loss parameter. Finally, the first de-identification utility parameter is obtained by subtracting this fusion de-identification utility loss parameter from 1. This parameter directly reflects the ability of the first trusted processing scheme to maintain the correlation between data categories and the overall data utility while protecting data privacy. For example, in a medical and financial data sharing scenario, if the above analysis shows a high first de-identification utility parameter, it indicates that the current de-identification processing scheme effectively preserves the correlation between data and its analytical value while protecting privacy.

[0034] In a preferred embodiment, based on cross-domain data sharing records over a historical period, the correlation between data categories of the multidimensional data is analyzed to obtain multiple correlation degrees, including: traversing and combining multiple data categories of the multidimensional data in pairs to obtain multiple data category groups; collecting multiple historical data sharing records based on cross-domain data sharing records over a historical period; collecting the number of times multiple data category groups appear in the multiple historical data sharing records to obtain multiple correlation occurrence counts; and allocating and calculating multiple correlation degrees based on the multiple correlation occurrence counts.

[0035] Furthermore, to assess the correlations between multidimensional data categories, the data categories are first iterated and paired to form multiple data category groups. For example, if the dataset contains three data categories: age, income, and education level, it may form three data category groups: "age-income," "age-education level," and "income-education level." Subsequently, a large number of historical data sharing records are collected as the basis for analysis, based on cross-domain data sharing records over a historical period. In these historical records, for each data category group, the frequency of its occurrence in all shared records is counted to obtain multiple correlation occurrence frequencies. For example, if the "age-income" combination appears 30 times in 100 shared records, its correlation occurrence frequency is 30. The correlation occurrence frequency reflects the co-occurrence frequency of the data category group in historical sharing. Finally, based on these correlation occurrence frequencies, multiple correlation degrees are calculated by assigning the ratio of each correlation occurrence frequency to the sum of all correlation occurrence frequencies. For example, if the total number of associations across all data category groups is 100, and the "age-income" combination appears 30 times, then its association degree is 30 / 100 = 0.3. This value reflects the close association between the "age" and "income" data categories in historical data sharing. This process systematically quantifies the associations between multidimensional data categories, providing an important basis for subsequent desensitization effectiveness analysis and processing scheme optimization.

[0036] In a preferred embodiment, performing data utility loss analysis on the first trusted processing scheme based on the multidimensional data to obtain a first data utility loss parameter includes: combining multiple data categories of the multidimensional data with multiple first desensitization processing parameters within the first trusted processing scheme to obtain multiple data utility loss analysis input data; inputting the multiple data utility loss analysis input data into a pre-built data utility loss analyzer to output multiple data utility loss parameters; and calculating the first data utility loss parameter based on the multiple data utility loss parameters.

[0037] In detail, to assess the impact of the first trusted processing scheme on data utility, a data utility loss analysis is required. Specifically, this involves combining multiple data categories within the multidimensional data, such as age, income, and education level, with multiple first anonymization parameters within the first trusted processing scheme. The input data for the data utility loss analysis refers to the data categories combined with specific anonymization parameters, which form the basis for subsequent analysis. For example, if a data category is age, and the anonymization parameter for age in the first trusted processing scheme is ±5%, then the input data for the data utility loss analysis would be the age field and its corresponding ±5% anonymization parameter.

[0038] Subsequently, multiple data points for data utility loss analysis are fed into a pre-built data utility loss analyzer. This analyzer, based on a preset algorithm or model, quantifies the impact of anonymization on the utility of each data category and outputs multiple data utility loss parameters. For example, the analyzer might assess that the age field loses 10% of its data utility with ±5% anonymization. Finally, based on these multiple data utility loss parameters, a first data utility loss parameter is calculated using a weighted average or other statistical methods. This parameter comprehensively reflects the impact of the first reliable processing scheme on the overall utility of the multidimensional data, providing an important basis for optimizing subsequent processing schemes.

[0039] In a preferred embodiment, the construction process of the data utility loss analyzer includes: collecting a set of sample data categories and a set of sample anonymization parameters based on data sharing records of different data categories; performing anonymization processing on different sample data categories and sample anonymization parameters; labeling the proportion of data utility loss after anonymization to obtain a set of sample data utility loss parameters; constructing a data utility loss analyzer using machine learning; and iteratively supervising the data utility loss analyzer using the set of sample data categories, the set of sample anonymization parameters, and the set of sample data utility loss parameters as input training data and supervised training data until convergence.

[0040] For example, in constructing a data utility loss analyzer, it is necessary to systematically collect a set of sample data categories and a set of sample anonymization parameters based on historical data sharing records of different data categories. The set of sample data categories covers various sensitive data categories such as age, income, and medical records, while the set of sample anonymization parameters contains the specific anonymization parameters applied to these data categories, such as ±5% fluctuation for the age field or 20% scaling for the income field.

[0041] Subsequently, for each combination of sample data category and sample anonymization parameters, anonymization is performed, and the percentage loss of data utility after anonymization is labeled, thus forming a set of sample data utility loss parameters. For example, for the age field, if the data utility is lost by 10% after applying ±5% anonymization, this percentage loss is labeled as the corresponding sample data utility loss parameter.

[0042] Next, using machine learning techniques, a data utility loss analyzer is constructed based on these labeled sample data. The data utility loss analyzer is essentially a predictive model designed to learn the intrinsic relationship between data categories, anonymization parameters, and data utility loss. Finally, the data utility loss analyzer is iteratively trained using the sample data category set, sample anonymization parameter set, and sample data utility loss parameter set as input training data and supervised training data. During training, the analyzer continuously adjusts its internal parameters to minimize the error between predicted and true values ​​until the model converges, reaching the preset prediction accuracy or iteration limit. For example, in medical and financial data sharing scenarios, through training with a large amount of sample data, the data utility loss analyzer can accurately predict the degree of utility loss for each data category under different anonymization parameters, providing strong support for the optimization of subsequent anonymization schemes.

[0043] In a preferred embodiment, calculating the first trustworthy processing fitness by combining the first desensitization utility parameter includes: calculating the first data utility parameter based on the first data utility loss parameter; and calculating the first trustworthy processing fitness based on the first desensitization utility parameter and the first data utility parameter.

[0044] Preferably, in evaluating the suitability of the first reliable processing scheme, the first data utility parameter needs to be calculated based on the obtained first data utility loss parameter. The first data utility parameter reflects the degree to which the dataset maintains its original analytical value and utility under a specific desensitization process. Its calculation usually involves the inverse operation or transformation of the data utility loss parameter. For example, if the first data utility loss parameter indicates that the data utility has been reduced by 30%, then the first data utility parameter can be set to 70% to quantify the retained data utility.

[0045] Subsequently, combining the calculated first de-identification utility parameter, which measures the effectiveness of the de-identification process in protecting data privacy, and comprehensively considering the first de-identification utility parameter and the first data utility parameter, the first trustworthy processing fitness is calculated. The first trustworthy processing fitness, as the result of this comprehensive evaluation, reflects the de-identification processing scheme's ability to balance privacy protection and data utility maintenance. For example, if the first de-identification utility parameter indicates excellent privacy protection performance, while the first data utility parameter shows a high degree of data utility retention, then the calculated first trustworthy processing fitness will be at a high level, indicating that the de-identification processing scheme has high applicability and effectiveness in cross-domain data sharing. This fitness index provides an important basis for the optimization and selection of subsequent processing schemes.

[0046] In a preferred embodiment, optimizing the first trusted processing scheme to obtain the optimal trusted processing scheme, performing trusted de-identification processing on the multiple datasets, and performing trusted cross-domain data sharing includes: continuing to perform random trusted de-identification processing on the multiple datasets to obtain a second trusted processing scheme, calculating the fitness of the second trusted processing, and optimizing the second trusted processing scheme; after optimization convergence, outputting the optimal trusted processing scheme with the largest trusted processing fitness, performing trusted de-identification processing on the multiple datasets, and performing trusted cross-domain data sharing.

[0047] Specifically, to obtain the optimal data processing strategy, it is necessary to continuously implement randomized trusted de-identification processing on multiple datasets to generate diverse trusted processing schemes. Randomized trusted de-identification processing means that in each processing step, de-identification parameters (such as the degree of data obfuscation and replacement rules) are randomly generated, thereby exploring different de-identification strategy spaces. For each newly generated trusted processing scheme, its corresponding trusted processing fitness needs to be calculated. This fitness comprehensively considers the performance of de-identification processing in protecting data privacy and maintaining data utility. Subsequently, based on the calculated trusted processing fitness, optimization algorithms (such as genetic algorithms and simulated annealing) are used to iteratively optimize the trusted processing scheme, aiming to gradually improve the fitness and find better de-identification strategies. In the specific optimization process, operations such as adjusting de-identification parameters and combining multiple de-identification techniques may be involved to balance the needs of privacy protection and data utility. For example, in the cross-domain sharing scenario of medical and financial data, the optimization algorithm may gradually adjust the de-identification degree of patient age and the obfuscation degree of income data to ensure that the data can still be used for effective risk assessment and analysis while protecting personal privacy. When the optimization process converges, i.e., when the fitness no longer significantly improves, the optimal trusted processing scheme with the highest fitness is output. Finally, this optimal scheme is used to perform trusted anonymization processing on multiple datasets to ensure that the processed data meets privacy protection requirements while retaining sufficient data utility, thereby supporting trusted data sharing across domains.

[0048] The cross-domain multidimensional data trusted sharing processing method provided by this invention has at least the following technical effects:

[0049] 1. By analyzing historical cross-domain data sharing records, the correlation between multi-dimensional data categories is quantified. Based on this, the effectiveness of the anonymization process is analyzed. Instead of viewing anonymization in isolation, the inherent connections between data are taken into consideration. This ensures that while protecting privacy, the anonymization process can retain the correlation and analytical value between data to the greatest extent, thereby improving the overall effectiveness of the anonymization process.

[0050] 2. A data utility loss analyzer based on machine learning was constructed. Through training on a large amount of sample data, it can accurately predict the degree of data utility loss under different de-identification parameters, realize the quantitative assessment of data utility loss, provide a scientific basis for the optimization of de-identification schemes, and enable de-identification to find a better balance between privacy protection and data utility.

[0051] 3. By calculating the fitness of trusted processing, the de-identification processing scheme is iteratively optimized until the optimal trusted processing scheme with the highest fitness is found. The fitness function is introduced as the optimization objective, and the privacy protection effect and data utility loss of de-identification processing are comprehensively considered. This ensures that the final output de-identification processing scheme not only meets the privacy protection requirements, but also maximizes the retention of data practicality and analytical value, thereby supporting trusted data sharing across domains.

[0052] Example 2:

[0053] like Figure 2 As shown, based on the same inventive concept as the cross-domain multidimensional data trusted sharing processing method provided in Embodiment 1, this embodiment of the invention also provides a cross-domain multidimensional data trusted sharing processing platform, the platform comprising:

[0054] The data desensitization module 11 is used to acquire multiple datasets of multidimensional data to be shared across domains, perform random and reliable desensitization processing on the multiple datasets, and obtain a first reliable processing scheme.

[0055] The utility analysis module 12 is used to perform desensitization utility analysis on the first trusted processing scheme based on the correlation between the multidimensional data, and obtain the first desensitization utility parameter, wherein the desensitization utility analysis includes desensitization utility loss analysis.

[0056] The loss analysis module 13 is used to perform data utility loss analysis on the first trusted processing scheme based on the multidimensional data, obtain the first data utility loss parameter, and calculate the first trusted processing fitness by combining the first desensitization utility parameter.

[0057] The trusted sharing module 14 is used to optimize the first trusted processing scheme according to the first trusted processing fitness to obtain the optimal trusted processing scheme, perform trusted desensitization processing on the multiple datasets, and perform trusted cross-domain data sharing.

[0058] Furthermore, the data desensitization module 11 is also used to perform the following steps:

[0059] Multiple datasets of multidimensional data to be shared across domains are obtained; the desensitization parameter ranges of the multidimensional data are obtained respectively to form a trusted desensitization scheme space; multiple first desensitization parameters of the multidimensional data are randomly generated in the trusted desensitization scheme space to obtain a first trusted processing scheme.

[0060] Furthermore, the utility analysis module 12 is also used to perform the following steps:

[0061] Based on cross-domain data sharing records over a historical period, the correlation between data categories of the multidimensional data is analyzed to obtain multiple correlation degrees; based on multiple first desensitization processing parameters within the first trusted processing scheme, the absolute difference magnitude of the first desensitization processing parameters between every two data categories is calculated to obtain multiple first desensitization utility loss parameters; based on the multiple correlation degrees, the multiple first desensitization utility loss parameters are weighted and calculated to obtain a first fusion desensitization utility loss parameter, and the first desensitization utility parameter is calculated.

[0062] Furthermore, the utility analysis module 12 is also used to perform the following steps:

[0063] The multidimensional data is traversed and combined in pairs to obtain multiple data category groups; multiple historical data sharing records are collected based on cross-domain data sharing records within a historical time period; the number of times multiple data category groups appear in multiple historical data sharing records is collected to obtain multiple association occurrence counts; multiple association degrees are calculated and assigned based on the multiple association occurrence counts.

[0064] Furthermore, the loss analysis module 13 is also used to perform the following steps:

[0065] The multiple data categories of the multidimensional data are combined with multiple first desensitization processing parameters within the first trusted processing scheme to obtain multiple data utility loss analysis input data; the multiple data utility loss analysis input data are input into a pre-built data utility loss analyzer to output multiple data utility loss parameters; and a first data utility loss parameter is calculated based on the multiple data utility loss parameters.

[0066] Furthermore, the loss analysis module 13 is also used to perform the following steps:

[0067] Based on data sharing records of different data categories, a set of sample data categories and a set of sample anonymization parameters are collected. Anonymization is performed on different sample data categories and parameters, and the proportion of data utility loss after anonymization is labeled to obtain a set of sample data utility loss parameters. Machine learning is used to construct a data utility loss analyzer. The set of sample data categories, the set of sample anonymization parameters, and the set of sample data utility loss parameters are used as input training data and supervised training data to iteratively supervise and train the data utility loss analyzer until convergence.

[0068] Furthermore, the loss analysis module 13 is also used to perform the following steps:

[0069] Based on the first data utility loss parameter, the first data utility parameter is calculated; based on the first desensitization utility parameter and the first data utility parameter, the first trustworthy processing fitness is calculated.

[0070] Furthermore, the trusted sharing module 14 is also used to perform the following steps:

[0071] Continue to perform random trusted desensitization processing on the multiple datasets to obtain a second trusted processing scheme, calculate the fitness of the second trusted processing, and optimize the second trusted processing scheme; after optimization convergence, output the optimal trusted processing scheme with the largest trusted processing fitness, perform trusted desensitization processing on the multiple datasets, and carry out trusted cross-domain data sharing.

[0072] Through the foregoing detailed description of a cross-domain multidimensional data trusted sharing processing method, those skilled in the art can clearly understand that this embodiment provides a cross-domain multidimensional data trusted sharing processing platform. As the platform disclosed in the embodiment corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant details can be found in the method section.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cross-domain, multi-dimensional data trusted sharing processing method, characterized in that, The method includes: Multiple datasets of multidimensional data to be shared across domains are obtained, and random trusted de-identification processing is performed on the multiple datasets to obtain a first trusted processing scheme. Based on the correlation between the multidimensional data, a desensitization utility analysis is performed on the first trusted processing scheme to obtain a first desensitization utility parameter. The desensitization utility analysis includes a desensitization utility loss analysis, comprising: Based on cross-domain data sharing records over a historical period, the correlation between data categories of the multidimensional data is analyzed to obtain multiple correlation degrees; Based on the multiple first desensitization processing parameters in the first trusted processing scheme, the absolute difference magnitude of the first desensitization processing parameters between each two data categories is calculated to obtain multiple first desensitization utility loss parameters. Based on the multiple correlation degrees, the multiple first desensitization utility loss parameters are weighted and calculated to obtain the first fusion desensitization utility loss parameter, and the first desensitization utility parameter is calculated. Based on the multidimensional data, a data utility loss analysis is performed on the first trusted processing scheme to obtain a first data utility loss parameter. Combined with the first desensitization utility parameter, the first trusted processing fitness is calculated. The first trusted processing scheme is optimized based on the first trusted processing fitness to obtain the optimal trusted processing scheme. Trusted desensitization processing is then performed on the multiple datasets to achieve trusted cross-domain data sharing.

2. The cross-domain multidimensional data trusted sharing processing method according to claim 1, characterized in that, Multiple datasets of multidimensional data to be shared across domains are obtained, and random, trustworthy de-identification processing is performed on the multiple datasets to obtain a first trustworthy processing scheme, including: Acquire multiple datasets of multidimensional data to be shared across different domains; The desensitization parameter ranges of the multidimensional data are obtained respectively to form a reliable desensitization scheme space; Multiple first desensitization processing parameters of multidimensional data are randomly generated within the space of the trusted desensitization processing scheme to obtain the first trusted processing scheme.

3. The cross-domain multidimensional data trusted sharing processing method according to claim 1, characterized in that, Based on historical cross-domain data sharing records, the correlation between data categories of the multidimensional data is analyzed to obtain multiple correlations, including: The multiple data categories of the multidimensional data are traversed and combined in pairs to obtain multiple data category groups; Based on cross-domain data sharing records over a historical period, collect multiple historical data sharing records; Collect the frequency of occurrence of multiple data category groups within multiple shared historical data records to obtain multiple associated occurrence counts; Based on the frequency of occurrence of the multiple associations, multiple association degrees are calculated and assigned.

4. The cross-domain multidimensional data trusted sharing processing method according to claim 1, characterized in that, Based on the multidimensional data, a data utility loss analysis is performed on the first trusted processing scheme to obtain a first data utility loss parameter, including: The multiple data categories of the multidimensional data are combined with multiple first desensitization processing parameters in the first trusted processing scheme to obtain multiple data utility loss analysis input data. The multiple data utility loss analysis input data are input into a pre-built data utility loss analyzer, and multiple data utility loss parameters are output. The first data utility loss parameter is calculated based on the multiple data utility loss parameters.

5. The cross-domain multidimensional data trusted sharing processing method according to claim 4, characterized in that, The construction process of the data utility loss analyzer includes: Based on the data sharing records of different data categories, a set of sample data categories and a set of sample desensitization parameters are collected. Desensitization processing is performed on different sample data categories and sample desensitization parameters. The proportion of data utility loss after desensitization is labeled to obtain a set of sample data utility loss parameters. We use machine learning to build a data utility loss analyzer; The data utility loss analyzer is iteratively trained using the sample data category set, the sample desensitization parameter set, and the sample data utility loss parameter set as input training data and supervised training data until convergence.

6. The cross-domain multidimensional data trusted sharing processing method according to claim 1, characterized in that, The calculation of the first trustworthy processing fitness, based on the first desensitization utility parameter, includes: calculating the first data utility parameter based on the first data utility loss parameter; The first trustworthy processing fitness is calculated based on the first desensitization utility parameter and the first data utility parameter.

7. The cross-domain multidimensional data trusted sharing processing method according to claim 1, characterized in that, The first trusted processing scheme is optimized to obtain the optimal trusted processing scheme. Trusted de-identification processing is then performed on the multiple datasets to achieve trusted cross-domain data sharing, including: Continue to perform random and trusted de-identification processing on the multiple datasets to obtain a second trusted processing scheme, calculate the fitness of the second trusted processing scheme, and optimize the second trusted processing scheme; After optimization and convergence, the optimal trusted processing scheme with the highest trusted processing fitness is output, and trusted desensitization processing is performed on the multiple datasets to achieve trusted data sharing across domains.

8. A cross-domain, multi-dimensional data trusted sharing processing device, characterized in that, The apparatus for implementing the cross-domain multidimensional data trusted sharing processing method according to any one of claims 1-7, the apparatus comprising: The data desensitization module is used to acquire multiple datasets of multidimensional data to be shared across domains, perform random and reliable desensitization processing on the multiple datasets, and obtain a first reliable processing scheme. The utility analysis module is used to perform desensitization utility analysis on the first reliable processing scheme based on the correlation between the multidimensional data, and obtain a first desensitization utility parameter. The desensitization utility analysis includes desensitization utility loss analysis, comprising: Based on cross-domain data sharing records over a historical period, the correlation between data categories of the multidimensional data is analyzed to obtain multiple correlation degrees; Based on the multiple first desensitization processing parameters in the first trusted processing scheme, the absolute difference magnitude of the first desensitization processing parameters between each two data categories is calculated to obtain multiple first desensitization utility loss parameters. Based on the multiple correlation degrees, the multiple first desensitization utility loss parameters are weighted and calculated to obtain the first fusion desensitization utility loss parameter, and the first desensitization utility parameter is calculated. The loss analysis module is used to perform data utility loss analysis on the first trusted processing scheme based on the multidimensional data, obtain the first data utility loss parameter, and calculate the first trusted processing fitness by combining the first desensitization utility parameter. The trusted sharing module is used to optimize the first trusted processing scheme based on the first trusted processing fitness to obtain the optimal trusted processing scheme, perform trusted desensitization processing on the multiple datasets, and perform trusted cross-domain data sharing.

Images