Privacy protection method based on difference model
The differential privacy model with Laplace noise injection addresses privacy leaks in data sharing by ensuring personal data protection and data utility, providing a scalable solution for diverse data analysis and sharing.
Patent Information
- Application Number
- CN202411425165.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2024-12-31
AI Technical Summary
Existing privacy protection methods, such as data desensitization and encryption, are inadequate in preventing privacy leaks during data sharing and publication, and there is a need for a more robust solution to protect individual privacy while allowing data analysis and sharing.
A privacy protection method based on differential privacy models, utilizing the Laplace mechanism to add noise to query results, combined with data cleaning, de-identification, encryption, and access control, ensures that personal data remains protected during analysis and sharing, while maintaining data utility.
This approach effectively safeguards personal privacy, ensuring data availability and credibility, achieving a balance between privacy and utility, and is scalable for various data analysis and sharing scenarios.
Smart Images
Figure CN119227113A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data encryption technology, and in particular to a privacy protection method based on differential models. It provides data privacy protection, privacy protocols, data analysis, and private data sharing and publication. It addresses privacy leaks during data sharing and publication, protects the privacy rights of data owners, and promotes trusted data sharing and collaborative research. The system and method described in the patent effectively protect the privacy of user sensitive data and meet relevant privacy protection requirements. Intelligent templates are used to generate privacy agreement policies corresponding to platforms and services. Background Art
[0002] With the advent of the big data era, privacy protection is a growing concern in the information society. The large-scale collection, storage, and processing of personal data exposes individuals to risks of privacy leakage and abuse. Traditional privacy protection methods, such as data desensitization and encryption, have demonstrated their limitations and shortcomings. Therefore, differential privacy, as an emerging privacy protection technology, offers a more powerful and reliable solution.
[0003] Differential privacy definition: Differential privacy is a privacy protection model that aims to protect individual privacy by adding carefully designed noise when querying and analyzing personal data. Simply put, differential privacy provides a mathematical guarantee that sensitive information related to a specific individual will not be revealed when analyzing data.
[0004] Privacy Budget and Sensitivity: The core concepts of differential privacy are privacy budget and sensitivity. The privacy budget is a parameter that controls the risk of privacy leakage and is used to limit the amount of noise added to query results. Sensitivity measures the impact of individual data on query results, thereby determining the required noise level.
[0005] Differential privacy mechanisms: Differential privacy mechanisms are specific algorithms and technologies that implement differential privacy. Common differential privacy mechanisms include the Laplace mechanism, the exponential mechanism, and the mixed noise mechanism. These mechanisms ensure the protection of individual privacy by introducing noise into query results.
[0006] Data publishing and sharing: Differential privacy can be applied to data publishing and sharing scenarios. When publishing private data, differential privacy mechanisms can be used to perturb the data to protect individual privacy. Furthermore, when sharing data, differential privacy mechanisms can be used to ensure that shared data does not lead to the leakage of individual privacy.
[0007] Privacy analysis and evaluation: Privacy analysis and evaluation in differential privacy systems involves analyzing and evaluating system performance and privacy protection effectiveness. This can include optimizing noise addition strategies and privacy budgets, as well as assessing the risk and probability of private information leakage in query results.
[0008] Applications of differential privacy: Differential privacy is widely used in various scenarios, including data mining, machine learning, and social network analysis. It provides a powerful tool for these application scenarios, which can not only protect personal privacy but also achieve effective use and analysis of data.
[0009] In summary, privacy protection systems and methods based on differential privacy have advanced the development of privacy protection technology. Related technologies include differential privacy definitions, privacy budgets and sensitivity, differential privacy mechanisms, data publishing and sharing, privacy analysis and assessment, and differential privacy applications. By employing differential privacy techniques, we can achieve secure processing and valuable analytical utilization of large-scale data while protecting individual privacy. Summary of the Invention
[0010] A privacy protection method based on a differential model includes the following steps: 1. Data collection and processing: Provide unified user privacy data collection, verification and processing services, including user identity information, resident information, and application information, and perform different processing based on different user privacy data directories, including data cleaning, de-identification and other operations; 1.1 Data cleaning: First, query the user data information and update it if it exists to ensure that the user information is unique. Then, perform data format verification to ensure the accuracy of the data format. Standardize data: standardize the data and save it according to the standard model of the platform user data directory. Statistical data: perform statistical analysis on the data to better understand the data and update the data directory. 1.2 Data De-identification: Anonymization: Replace personal information in the data with randomly generated unique user identification data, so that the data can no longer be identified to a specific user; Encryption: Encrypt the data so that it cannot be accessed by unauthorized personnel; 2. Data encryption and storage: Provides a unified data encryption portal, using different encryption and storage methods based on the user's private data directory and data confidentiality level. Identity information is encrypted using AES or SM2, and application information is obfuscated and encrypted using methods such as Base64. Through the user privacy data directory, the user's basic information, application information, device information, etc. are encrypted with Base64, and the user identity information is encrypted with AES; 3. Access control of private data: Provide access control through authentication and data authorization of the access end to control access to private data and memory data; 3.1. Authentication requirements: The access terminal registers as a data application according to different scenario applications and obtains permission access through authentication; 3.2. Directory authorization: Authorize access according to data privacy level; 4. Data Privacy Differential Mechanism: Introducing the LaplaceMechanism Laplace noise mechanism. By passing in the sensitivity and privacy budget, the generateNoise method is implemented to generate noise using a Laplace distribution. The random.nextDouble() method generates a uniformly distributed random number. The Laplace distributed noise value is then calculated using a formula. The unified addNoise method is implemented to add the generated noise to the original query result to obtain the query result after adding the noise. 4.1 User Identity Information After decrypting the encrypted stored user information query, a noise annotation method is used to add noise value to the facet after the query is completed and return it, and a verification method for noise identity data is provided; 5. Data publishing or sharing: Publish or share the data processed with differential privacy, allowing other platform services to subscribe to the corresponding user privacy data; 6. Dataset aggregation and privacy analysis and assessment: Evaluate and analyze the user's privacy data on the platform as a whole, aggregate and analyze the user's privacy dataset according to the data directory, and conduct a security assessment of the data; 6.1. Evaluate data quality and integrity by comparing data before and after data aggregation for different user data directories, and perform data consistency checks; 6.2. Privacy assessment mainly evaluates whether the aggregated data contains personal privacy information, personal identity information, and personal behavior information, corresponding to the risk assessment of privacy leakage, identity information leakage, and personal behavior tracking respectively.
[0011] The purpose of this patent is to provide a technical solution that effectively protects personal privacy and can be applied to various platforms to store and publish various user privacy information, analyze privacy policies, and provide privacy agreement templates. It also solves the problem of personal privacy leakage that may occur during data analysis and sharing, and ensures that the privacy of sensitive data is effectively protected.
[0012] 1. Privacy protection: Provide a method and system that can effectively protect personal sensitive information, ensuring that personal data will not be leaked or exposed during data analysis and sharing.
[0013] 2. Data availability: Ensure the availability and validity of data while minimizing interference and noise impact on the data to ensure the accuracy and credibility of data analysis results.
[0014] 3. Good balance between privacy and utility: Through the design and optimization of differential models, a good balance between personal privacy and data utility is achieved, so that data analysis and sharing projects can simultaneously meet the needs of privacy protection and data value.
[0015] 4. Scalability and Applicability: Provide a general and scalable differential model privacy protection system and method that is applicable to different fields and different types of data analysis and sharing scenarios.
[0016] 5. Provide a universal privacy agreement: Provide a unified user privacy agreement and personal information list based on the user's privacy data and personal information. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 This is a schematic diagram of the platform's private data storage; Figure 2 Schematic diagram of platform privacy data access and sharing; Figure 3 Dynamically generate diagrams for platform privacy agreements and user information lists; Figure 4 This is a schematic diagram of the user privacy data directory; Figure 5 This is a schematic diagram of the privacy assessment case process; Figure 6 Example diagram of privacy for public service apps. DETAILED DESCRIPTION
[0018] We will employ a differential privacy model to protect user privacy by adding Laplace noise. We will also provide a data validity algorithm to enhance the validity and usability of privacy in both data release and privacy protocols. The following is a detailed technical implementation.
[0019] 1. Data Collection and Processing: Provides unified user privacy data collection, verification, and processing services. This includes user identity information, resident information, application information, etc. Different processing methods are performed based on the user privacy data directory, including data cleaning and de-identification operations.
[0020] Data cleaning: 1) First query the user data information, and update it if it exists to ensure that the user information is unique.
[0021] 2) Perform data format verification to ensure the accuracy of the data format.
[0022] 3) Standardized data: Standardize the data and save it according to the standard model of the platform user data directory.
[0023] 4) Statistics: Statistical analysis of data is performed to better understand the data and update the data catalog.
[0024] Data de-identification: 1) Anonymization: Replace personal information in the data with randomly generated user unique identification data so that the data can no longer be identified to a specific user.
[0025] 2) Encryption: Encrypt data so that it cannot be accessed by unauthorized personnel.
[0026] 2. Data encryption and storage: Provide a unified data encryption entry and adopt different encryption and storage methods based on the user's privacy data directory and data confidentiality level. For example, identity information is encrypted in the form of AES or SM2, and application information is obfuscated and encrypted using methods such as base64.
[0027] By using the user privacy data directory, user basic information, application information, device information, etc. are encrypted with Base64, which enables better data analysis and use. User identity information is encrypted with AES to ensure data security.
[0028] 3. Access control of private data: Provide authentication of the access end and data authorization to control access to private data and memory data, ensuring the security of data access.
[0029] 3.1. Authentication requirements: The access end registers as a data application according to different scenario applications and obtains permission access through authentication.
[0030] 3.2. Directory authorization: Authorize access according to the data privacy level.
[0031] 4. Data Privacy Differential Mechanism: A Laplace noise mechanism is implemented without sacrificing data availability or analytical effectiveness. In this algorithm, we input the sensitivity and privacy budget. The generateNoise method generates noise using a Laplace distribution. The random.nextDouble() method generates a uniformly distributed random number, and the Laplace noise value is calculated using a formula. A unified addNoise method is implemented to add the generated noise to the original query result to obtain the noised result.
[0032] 4.1. Example: User Identity Information After decrypting the query of encrypted stored user information, through a noise annotation method, noise is added after the aspect of the query ends and then returned, and a verification method for noise identity data is provided to ensure the availability of the noise data.
[0033] 5. Data publishing or sharing: Publish or share the data after differential privacy processing, so that other platform services can subscribe to the corresponding user privacy data.
[0034] 6. Dataset aggregation and privacy analysis evaluation: In order to evaluate and analyze the overall privacy data of users on the platform, the privacy datasets of users will be aggregated and analyzed according to the data directory, and the data will be subject to security evaluation.
[0035] 6.1 Privacy analysis description: Mainly evaluate the data quality and integrity by comparing the data before and after data aggregation for different user data directories, and perform data consistency checks. 6.2 Privacy evaluation mainly evaluates whether the data after data aggregation contains personal privacy information, personal identity information, and personal behavior information, corresponding to the risk assessments of privacy leakage, identity information leakage, and personal behavior tracking respectively.
[0036] 6.3 Figure 5 It is a schematic diagram of the privacy evaluation case process; 6.4 Case: The aggregation of user data is provided in the form of data json, for example: {"userName":"xxz you are a test","name":"xxz Wang xxz Xiaoming","phone":"yyz1810002020202","idcard":"zzz3040032030230302303zzz","sign":"signValue"}, through the user data directory list, the user's account, name, phone number, and ID card information are assembled and encrypted and compared with the signValue of sign. If it is successful, the privacy data is complete; if the comparison is unsuccessful, the data is incomplete.
[0037] The encryption KEY for user data is generated algorithmically as 16 bits through the unique KEY corresponding to the user, which can ensure the consistency of the encrypted data. Privacy evaluation is based on the user's directory list. If the aggregated data of the user contains account information, name, and phone number, it is evaluated as level 2; if there is ID card information, etc., it is evaluated as level 3; if there is user behavior data, it is level 4.
[0038] 7. Automatic Generation of Privacy Policies and Personal App Lists: Automatically generate or assist in the generation of an organization's privacy policy based on aggregated user privacy data, cataloged, and analyzed to identify sensitive information and regulatory requirements. This privacy policy should clearly define how users' personal information is collected, used, shared, and protected, and should be updated promptly to adapt to regulatory changes and organizational practices.
[0039] 7.1、 Figure 6 As shown in a certain city’s public service APP, the application template, user agreement template, third-party information data sharing template, personal information collection list template, and privacy policy template are formed according to regulatory requirements. Based on the application information, a compliant user agreement, third-party information data sharing, personal information collection list, and privacy policy are formed.
[0040] The technical points involved in this plan include: 1. Based on the privacy differential model constructed from the platform's user data directory, we perform sensitivity processing for different privacy data types. By perturbing query results, we control the degree to which the differential privacy system affects individual sensitive information. We can improve the utility of data by adjusting noise levels and performing aggregation operations.
[0041] 2. Privacy budget management: Set a reasonable privacy budget for different data query operations to balance the relationship between privacy protection and data analysis utility.
[0042] 3. Intelligent privacy agreement model: Use artificial intelligence technology to automatically generate privacy laws and regulations to ensure that personal privacy agreements and information lists comply with the requirements of relevant laws and regulations.
[0043] Glossary Differential Privacy: A privacy protection model that introduces noise into query results and encrypts data to hide individual information and protect the privacy of data owners.
[0044] Laplace Noise: A common noise injection technique, often used in differential privacy to increase the randomness of query results and protect data privacy.
Claims
1. A privacy protection method based on a differential model, characterized in that The steps include:
1. Data collection and processing: Provide unified user privacy data collection, verification and processing services, including user identity information, resident information, and application information, and perform different processing according to different user privacy data directories, including data cleaning, de-identification and other operations; Data cleaning: First query the user data information, and update it if it exists to ensure that the user information is unique; then perform data format verification to ensure the accuracy of the data format; standardize data: standardize the data and save it according to the standard model of the platform user data directory; statistical data: perform statistical analysis on the data to better understand the data and update the data directory; Data De-identification: Anonymization: Replace personal information in the data with randomly generated unique user identification data so that the data can no longer be identified to a specific user; Encryption: Encrypt the data so that it cannot be accessed by unauthorized persons; 2. Data encryption and storage: Provide a unified data encryption portal, and use different encryption and storage methods based on the user's privacy data directory and data confidentiality level. Identity information is encrypted in the form of AES or SM2, and application information is obfuscated and encrypted using methods such as base64; Through the user privacy data directory, the user's basic information, application information, device information, etc. are encrypted with base64, and the user identity information is encrypted with AES; 3. Access control of private data: Provide access control through authentication and data authorization of the access end to control access to private data; 3.
1. Authentication requirements: The access end registers as a data application according to different scenario applications and obtains permission access through authentication; 3.
2. Directory authorization: Authorize access according to the data privacy level; 4. Data privacy differential mechanism: Introduce the LaplaceMechanism Laplace noise mechanism, pass in the sensitivity and privacy budget, and use the Laplace distribution to generate noise by implementing the generateNoise method, where random.nextDouble() generates a uniformly distributed random number, and then calculates the noise value of the Laplace distribution through the formula. Implement a unified addNoise method to add the original query result to the generated noise to obtain the query result after adding the noise; 4.1 User Identity Information After decrypting the encrypted stored user information query, a noise annotation method is used to add noise value to the facet after the query is completed and then return it, and a verification method for noise identity data is provided; 5. Data publishing or sharing: Publish or share the data after differential privacy processing, so that other platform services can subscribe to the corresponding user privacy data; 6. Dataset aggregation and privacy analysis and evaluation: Evaluate and analyze the user's privacy data on the platform as a whole, aggregate and analyze the user's privacy data set according to the data directory, and conduct security assessment on the data; 6.
1. Mainly evaluate data quality and integrity by comparing data before and after data aggregation for different user data directories, and perform data consistency checks; 6.
2. Privacy assessment mainly evaluates whether the aggregated data contains personal privacy information, personal identity information, and personal behavior information, corresponding to the risk assessment of privacy leakage, identity information leakage, and personal behavior tracking.