Ranking hiding privacy protection method and system based on tensor

Abstract

The invention provides a tensor-based ranking hiding privacy protection method and system, and the method comprises the steps: carrying out the tensor modeling of an item set, a user set, a position set and a time set, so as to construct an original four-order tensor composed of four dimensions, and the content of the tensor is ranking data; laplace noise after square root compression is added to high-order data in the fourth-order tensor, and a mixed tensor is obtained; the method comprises the following steps: selecting target data with minimum privacy leakage and ranking loss from confused tensors by calculating ranking loss and mutual information privacy loss, and replacing data in a fourth-order tensor with the target data to obtain a final ranking hidden tensor, and specifically, after tensor representation is constructed for user multi-modal behavior data, obtaining a final ranking hidden tensor. According to the method, classification is firstly carried out according to ranking importance, and then differential hiding processing is implemented, so that the accuracy of an analysis result is maintained while privacy security is ensured.

Description

Technical Field

[0001] This invention belongs to the field of ranking privacy protection technology, specifically relating to a ranking privacy protection method and system based on tensors. Background Technology

[0002] With the rapid development of internet technology, users frequently generate multimodal behavioral data, including tags, ratings, and check-ins, on various online platforms. While the release and sharing of this data drives the development of personalized analytics services, it also brings the risk of user privacy leaks. There are often potential correlations between users' private data (such as gender, health status, and income level) and public data, allowing attackers to potentially infer sensitive information by analyzing public data. Therefore, before providing user data to third-party analytics platforms, it is crucial to implement effective privacy protection while ensuring data availability.

[0003] In the early stages of privacy protection technology development, Sweeney L. proposed the k-anonymity model, which achieves identity concealment by mixing an individual's records with at least k-1 other individuals. With the increasing prevalence of multi-source heterogeneous data, Wong RCW et al. further proposed the (α, k)-anonymity model, which protects attribute association privacy through local encoding. However, these methods are mostly designed for structured relational data and are difficult to directly apply to multimodal high-order data scenarios such as tensors, often resulting in decreased data utility due to excessive information loss.

[0004] To balance privacy and utility in high-dimensional data environments, Ahuja R. et al. proposed a distortion measurement mechanism based on Euclidean distance, while Janowicz K. used JS distance to protect semantic signature data. Gao C. further proposed semantic Euclidean distance, constructing a privacy-preserving cross-domain location analysis framework, and combining differential privacy and matrix factorization techniques to achieve secure data sharing. Preethi P. further proposed a privacy-preserving clustering method based on heterogeneous data distortion (PPDC). Although these methods perform well on scalar or matrix data, their distance measurement methods are insufficient to effectively characterize the complex relationships in tensor ranking data and cannot constrain higher-order distortions in ranking analysis tasks.

[0005] For ranking-based analytics scenarios, Yang D. et al. proposed the PrivRank framework, which achieves privacy protection by obfuscating user data globally and controlling ranking loss. However, this method employs a global order-preserving strategy, failing to differentiate the strength of different data's impact on the analysis results, which may lead to issues such as distorted rankings of key items and insufficient privacy protection.

[0006] In short, existing privacy protection methods have significant limitations when dealing with ranking-based analytics tasks. While traditional k-anonymity and its improved models provide basic privacy guarantees, they lack generality and severely compromise data utility. Distance-based protection mechanisms (such as Euclidean distance and JS divergence) primarily focus on the degree of data distortion and struggle to effectively constrain ranking losses in ranking analysis. Although the PrivRank framework, specifically designed for ranking privacy, can control overall ranking bias, its global obfuscation strategy ignores differences in user preferences, potentially leading to ranking drops in core items and insufficient protection strength. Summary of the Invention

[0007] Based on this, the present invention provides a tensor-based ranking privacy protection method and system, which aims to improve privacy and security by obfuscating multiple dimensions of data through tensor modeling, improve the accuracy of analysis by protecting the correlation between multiple dimensions, and finally improve computational efficiency by simplifying the privacy protection process.

[0008] A first aspect of this invention provides a tensor-based ranking-hiding privacy protection method, applied in a cloud-edge architecture scenario. The cloud-edge architecture includes a trusted edge terminal and a cloud terminal communicatively connected to the trusted edge terminal. The trusted edge terminal performs privacy protection calculations, obfuscates the data, and uploads it to the cloud terminal. After analysis on the cloud terminal, the analysis results are sent to the user. The method includes: Tensor modeling is performed on the project set, user set, location set, and time set to construct the original fourth-order tensor consisting of these four dimensions, with the content of the tensor being the ranking data; Add square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain a scrambled tensor. By calculating the ranking loss and mutual information privacy loss, target data with the minimum privacy leakage and ranking loss is selected from the obfuscated tensor, and the target data is used to replace the data in the fourth-order tensor to obtain the final ranking hidden tensor.

[0009] Furthermore, in the step of adding square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain the obfuscated tensor, the expression for the noise is: ; in, Let be a uniformly distributed random variable. To adjust the parameters.

[0010] Furthermore, in the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hiding tensor, the expression for calculating the ranking loss is as follows: ; in, for and Ranking loss between , For tensor The total number of elements. For tensor The total number of elements, where SR represents the number of data points with the same rank, and DR represents the number of data points with different rank. , , for A collection of different elements in for A collection of different elements, auxiliary function , , Representing elements respectively , The number of times it appears in the corresponding tensor.

[0011] Furthermore, in the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hiding tensor, the expression for calculating mutual information privacy loss is as follows: ; ; ; ; ; ; in, Let the probability of occurrence be... for and Appeared and The number of times at the same location For an element of the fourth-order tensor, An element of the obfuscated tensor. For joint probability, For conditional probability, For information entropy based on tensors, For tensor-based conditional entropy, For mutual information based on tensors.

[0012] Furthermore, in the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor, obfuscation-constrained DC is used to balance ranking loss and privacy leakage.

[0013] Furthermore, the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor includes: Quantify ranking loss and mutual information privacy loss, and set privacy and security thresholds to constrain mutual information privacy loss; Based on the multidimensionality of the fourth-order tensor, a dimension-sensitive noise adjustment strategy is designed to make the highly sensitive, low-ranking-important dimensions receive stronger noise, while the low-sensitivity, high-ranking-important dimensions receive weaker noise. Generate candidate confusion tensors with multiple noise parameter combinations, and select the solution that satisfies the privacy constraints and minimizes the ranking loss through Pareto optimal selection; Local optimization is performed on the initial optimal solution to further reduce the ranking loss without breaking privacy constraints; The final ranking hidden tensor is deployed to a cloud-edge architecture to collect user feedback and is updated regularly.

[0014] Furthermore, in the step of designing a dimension-sensitive noise adjustment strategy based on the multidimensional characteristics of the fourth-order tensor, an independent noise parameter b is assigned to each dimension d. d The calculation formula is: ; Among them, S d For the privacy sensitivity of dimension d, R d The ranking importance of dimension d.

[0015] A second aspect of this invention provides a tensor-based ranking hiding privacy protection system for implementing the tensor-based ranking hiding privacy protection method provided in the first aspect of this invention. The system includes: The modeling module is used to perform tensor modeling on the itemset, user set, location set, and time set to construct the original fourth-order tensor consisting of these four dimensions, with the content of the tensor being ranking data. The noise addition module is used to add square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain the obfuscated tensor. The replacement module is used to select target data with minimum privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replace the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor.

[0016] A third aspect of the present invention provides a computer-readable storage medium, comprising: The readable storage medium stores one or more programs that, when executed by a processor, implement the tensor-based ranking-hiding privacy protection method as described in the first aspect.

[0017] A fourth aspect of the present invention provides an electronic device, the electronic device including a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the tensor-based ranking hiding privacy protection method as described in the first aspect.

[0018] This invention provides a tensor-based ranking hiding privacy protection method and system. It constructs an original fourth-order tensor by modeling the itemset, user set, location set, and time set using tensors, with the tensor content being ranking data. Square-root compressed Laplace noise is added to the higher-order data in the fourth-order tensor to obtain a confused tensor. By calculating ranking loss and mutual information privacy loss, target data with minimal privacy leakage and ranking loss is selected from the confused tensor, and this target data replaces the data in the fourth-order tensor to obtain the final ranking hiding tensor. Specifically, after constructing a tensor representation of user multimodal behavior data, it is first classified according to its ranking importance, and then differentiated hiding processing is implemented, thereby maintaining the accuracy of the analysis results while ensuring privacy and security. Attached Figure Description

[0019] Figure 1 The flowchart illustrates the implementation of a tensor-based ranking privacy protection method according to Embodiment 1 of the present invention. Figure 2 This is a structural block diagram of a tensor-based ranking hiding privacy protection system provided in Embodiment 3 of the present invention; Figure 3 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0020] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0021] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] Example 1 Embodiment 1 of this invention provides a tensor-based ranking-hiding privacy protection method, applied to scenarios with a cloud-edge architecture. The cloud-edge architecture includes a trusted edge terminal and a cloud terminal communicatively connected to the trusted edge terminal. The trusted edge terminal performs privacy protection calculations and obfuscation before uploading the results to the cloud terminal. After analysis on the cloud terminal, the analysis results are sent to the user. (See [link to relevant documentation]). Figure 1 This is a flowchart of a ranking-hiding privacy protection method based on tensors, specifically including steps S01 to S03.

[0024] Step S01: Perform tensor modeling on the project set, user set, location set, and time set to construct the original fourth-order tensor composed of these four dimensions. The content of the tensor is the ranking data.

[0025] In this process, a fourth-order tensor is constructed by collecting users' project sets, user sets, location sets, and time sets.

[0026] Step S02: Add square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain the obfuscated tensor.

[0027] Specifically, noise is added to each higher-order data point. This added noise may alter the data ranking, potentially preventing ranking privacy leaks. However, the added noise may cause data overflow in the four dimensions of the tensor, causing the data to lose its physical meaning. Therefore, obfuscated data needs to be constrained in the four dimensions of I (Item), U (User), L (Location), and T (Time). To reduce the possibility of data overflow, the noise formula adds a square root operation to the traditional Laplace inverse cumulative distribution function. The expression for the noise is: ; in, Let be a uniformly distributed random variable. To adjust the parameters.

[0028] In this embodiment of the invention, high-order data can be clearly defined from three dimensions: ranking priority, numerical influence, and business importance. Specifically, based on the definition of ranking position, it refers to high-ranking data that is in the top N% (e.g., top 20%) in the ranking dimension of the tensor; based on the definition of numerical size, it refers to data in the tensor with a large absolute value and a high contribution to the overall ranking; based on the definition of business importance, it refers to data that plays a key role in business logic or user experience, usually defined by business rules or domain knowledge.

[0029] Step S03: By calculating the ranking loss and mutual information privacy loss, select the target data with the minimum privacy leakage and ranking loss from the obfuscated tensor, and replace the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor.

[0030] It should be noted that the formula for calculating the ranking loss is as follows: ; in, for and Ranking loss between , For tensor The total number of elements. For tensor The total number of elements, where SR represents the number of data points with the same rank, and DR represents the number of data points with different rank. , , for A collection of different elements in for A collection of different elements, auxiliary function , , Representing elements respectively , The number of times it appears in the corresponding tensor.

[0031] Generally, the TRL value ranges from -1 to 1. A larger TRL indicates less ranking loss, but also more privacy leakage. Therefore, a confusion-limited DC is used to balance ranking loss and privacy leakage. Only when TRL > DC can both ranking loss and privacy leakage risks be considered acceptable. It should be noted that DC is a quantitative definition of the minimum acceptable ranking loss, combining business tolerance and privacy security baselines. In this embodiment of the invention, the business side sets the tolerance threshold based on the actual utility of the ranking results (such as recommendation click-through rate and decision accuracy). For example, in an e-commerce scenario, if a ranking accuracy of no less than 80% is required, the corresponding DC threshold can be derived through experimentation.

[0032] Furthermore, the expression for calculating the privacy loss of mutual information is as follows: ; ; ; ; ; ; in, Let the probability of occurrence be... for and Appeared and The number of times at the same location For an element of the fourth-order tensor, An element of the obfuscated tensor. For joint probability, For conditional probability, For information entropy based on tensors, For tensor-based conditional entropy, i.e. and Conditional entropy between them For tensor-based mutual information, i.e. and Mutual information between them.

[0033] In summary, the tensor-based ranking hiding privacy protection method proposed in this invention constructs an original fourth-order tensor composed of these four dimensions by tensor modeling of the itemset, user set, location set, and time set. The content of the tensor is the ranking data. Square root compressed Laplace noise is added to the higher-order data in the fourth-order tensor to obtain a confused tensor. By calculating the ranking loss and mutual information privacy loss, target data with the minimum privacy leakage and ranking loss is selected from the confused tensor, and this target data replaces the data in the fourth-order tensor to obtain the final ranking hiding tensor. Specifically, after constructing a tensor representation of user multimodal behavior data, it is first classified according to its ranking importance, and then differentiated hiding processing is implemented, thereby ensuring privacy and maintaining the accuracy of the analysis results.

[0034] Example 2 Embodiment 2 of the present invention also provides a ranking hiding privacy protection method based on tensors. The difference between this method and the ranking hiding privacy protection method based on tensors provided in Embodiment 1 of the present invention is that, with mutual information privacy loss ≤ security threshold as a hard constraint, under the premise of satisfying privacy and security, the ranking loss is minimized by dynamically adjusting the noise strategy, hierarchically optimizing the tensor dimension, and introducing a loss compensation mechanism.

[0035] Specifically, the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor includes: The ranking loss and mutual information privacy loss are quantified, and a privacy security threshold is set to constrain the mutual information privacy loss. Specifically, the marginal mutual information of each dimension is calculated: , , , Where U represents the original user dimension data. I represents the obfuscated user dimension data, while I represents the original project dimension data. The data represents the obfuscated project dimension data, while L represents the original location dimension data. The location dimension data is obfuscated, and T is the original time dimension data. Given the obfuscated time-dimension data, the total mutual information privacy loss is: ; weight w d Based on the sensitivity settings, for example, location L and user U have high sensitivity, w L w U Take 0.3, and take 0.2 for Project I and Time T; The total ranking loss is: ; In this embodiment of the invention, γ=0.6, emphasizing overall ranking consistency. It can be understood that the previously defined TRL formula is used to calculate the fourth-order tensor as a whole and the sub-tensors of each dimension (such as user-item sub-tensor and location-time sub-tensor). Furthermore, based on differential privacy theory, Mapping to the privacy budget ε, for example, , To ensure strong privacy. This is a hard constraint; Based on the multidimensionality of fourth-order tensors, a dimension-sensitive noise adjustment strategy is designed to provide stronger noise for dimensions with high sensitivity and low ranking importance, and weaker noise for dimensions with low sensitivity and high ranking importance. Specifically, an independent noise parameter b is assigned to each dimension d. d The calculation formula is: ; Among them, S d R represents the privacy sensitivity of dimension d (0-1, manually labeled or output by a data anonymization risk assessment model). d The ranking importance of dimension d (0-1, calculated by the weight of the influence of the ranking of this dimension on user decision-making in the original data). Multiple candidate confusion tensors with various noise parameter combinations are generated. The solution that satisfies privacy constraints and minimizes ranking loss is selected through Pareto optimality screening. In this embodiment of the invention, firstly, a candidate set is generated; specifically, perturbations are applied around bd (e.g., ±20%) to generate 10-20 different sets. Combining, noise is added to the original tensor to obtain candidate tensors. Subsequently, a Pareto front is constructed, specifically for each candidate Y. i calculate and Plot a scatter plot and filter for Pareto optimal solutions, i.e., solutions that do not exist for Y. j Make and Finally, constraint screening is performed; specifically, constraints are selected from the Pareto front. If there are multiple candidates, choose... The smallest one is taken as the initial optimal solution. If it does not exist, b0 is increased (increasing the overall noise), and the candidate generation step is repeated until a solution that meets the privacy constraints is found. Local optimization is performed on the initial optimal solution to further reduce the ranking loss without violating privacy constraints. It should be noted that, firstly, the pairs of elements in the initial optimal solution are calculated... The contribution of each element is assessed (e.g., by using gradient back-calculation: the greater the change in an element's ranking, the higher its contribution). The top 20% of key elements by contribution are then selected (e.g., frequently accessed items, ranking data in core positions). Subsequently, noise compensation is adjusted; specifically, for key elements, noise is appropriately reduced (by decreasing b). d To reduce its ranking loss, for non-critical elements in the bottom 30% of contribution, increase their noise (increase b). d ), compensate for privacy losses, and ensure that the adjustments are made ; The final ranking hidden tensor is deployed to a cloud-edge architecture to collect user feedback and update it regularly. Specifically, if users report low ranking accuracy (TRL too high), R is fine-tuned. d (Increase the ranking importance of key dimensions), regenerate the candidate set; if privacy risks increase (such as attack incidents), decrease... (Tighten privacy constraints), increase S d This increases noise intensity.

[0036] Example 3 Embodiment 3 of the present invention provides a tensor-based ranking hiding privacy protection system 200. Please refer to [link / reference]. Figure 2 Here is a structural block diagram of a tensor-based ranking hiding privacy protection system 200, which includes: Modeling module 21 is used to perform tensor modeling on the itemset, user set, location set and time set to construct the original fourth-order tensor composed of these four dimensions, and the content of the tensor is the ranking data; The noise addition module 22 is used to add square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain a confused tensor. The expression for the noise is: ; in, Let be a uniformly distributed random variable. To adjust the parameters; Replacement module 23 is used to select target data with minimum privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replace the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor. The expression for calculating the ranking loss is: ; in, for and Ranking loss between , For tensor The total number of elements. For tensor The total number of elements, where SR represents the number of data points with the same rank, and DR represents the number of data points with different rank. , , for A collection of different elements in for A collection of different elements, auxiliary function , , Representing elements respectively , The number of times it appears in the corresponding tensor; The formula for calculating mutual information privacy loss is: ; ; ; ; ; ; in, Let the probability of occurrence be... for and Appeared and The number of times at the same location For an element of the fourth-order tensor, An element of the obfuscated tensor. For joint probability, For conditional probability, For information entropy based on tensors, For tensor-based conditional entropy, For tensor-based mutual information; In addition, obfuscation-limited DCs are used to balance ranking loss and privacy breaches.

[0037] Furthermore, in some other embodiments of the present invention, the replacement module 23 includes: The quantification unit is used to quantify ranking loss and mutual information privacy loss, and to set a privacy security threshold to constrain mutual information privacy loss. The design unit is used to design dimension-sensitive noise adjustment strategies based on the multi-dimensional characteristics of the fourth-order tensor. This aims to provide stronger noise for highly sensitive dimensions with low ranking importance and weaker noise for dimensions with low sensitivity and high ranking importance, assigning an independent noise parameter b to each dimension d. d The calculation formula is: ; Among them, S d For the privacy sensitivity of dimension d, R d The ranking importance of dimension d; The generation unit is used to generate candidate confusion tensors with multiple noise parameter combinations, and the solution that satisfies the privacy constraint and minimizes the ranking loss is selected through Pareto optimality. Local optimization units are used to perform local optimization on the initial optimal solution, further reducing ranking loss without breaking privacy constraints; The deployment unit is used to deploy the final ranking hidden tensor to the cloud-edge architecture, collect user feedback, and update it regularly.

[0038] Example 4 Embodiment 4 of this invention proposes an electronic device; please refer to [link / reference]. Figure 3 This is a structural block diagram of an electronic device, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the tensor-based ranking hiding privacy protection method described above.

[0039] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0040] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, SmartMediaCard (SMC), SecureDigital (SD) card, FlashCard, etc., equipped on the electronic device. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0041] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the tensor-based ranking hiding privacy protection method described above.

[0042] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0043] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0044] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0045] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0046] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A ranking-hiding privacy protection method based on tensors, characterized in that, This method is applied to scenarios with a cloud-edge architecture, where the cloud-edge architecture includes a trusted edge terminal and a cloud terminal communicating with the trusted edge terminal. The trusted edge terminal performs privacy protection calculations and obfuscation before uploading the results to the cloud terminal. After analysis on the cloud terminal, the analysis results are sent to the user. Tensor modeling is performed on the project set, user set, location set, and time set to construct the original fourth-order tensor consisting of these four dimensions, with the content of the tensor being the ranking data; Add square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain a scrambled tensor. By calculating the ranking loss and mutual information privacy loss, target data with the minimum privacy leakage and ranking loss is selected from the obfuscated tensor, and the target data is used to replace the data in the fourth-order tensor to obtain the final ranking hidden tensor.

2. The tensor-based ranking privacy protection method according to claim 1, characterized in that, In the step of adding square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain the obfuscated tensor, the expression for the noise is: ； in, Let be a uniformly distributed random variable. To adjust the parameters.

3. The tensor-based ranking privacy protection method according to claim 2, characterized in that, In the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor, the expression for calculating the ranking loss is as follows: ； in, for and Ranking loss between , For tensor The total number of elements. For tensor The total number of elements, where SR represents the number of data points with the same rank, and DR represents the number of data points with different rank. , , for A collection of different elements in for Auxiliary function for a set of different elements , , Representing elements respectively , The number of times it appears in the corresponding tensor.

4. The tensor-based ranking privacy protection method according to claim 3, characterized in that, In the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor, the expression for calculating mutual information privacy loss is as follows: ； ； ； ； ； ； in, Let the probability of occurrence be... for and Appeared and The number of times at the same location For an element of the fourth-order tensor, An element of the obfuscated tensor. For joint probability, For conditional probability, For information entropy based on tensors, For tensor-based conditional entropy, For mutual information based on tensors.

5. The tensor-based ranking privacy protection method according to claim 4, characterized in that, In the step of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor, obfuscation-constrained DC is used to balance ranking loss and privacy leakage.

6. The tensor-based ranking privacy protection method according to claim 5, characterized in that, The steps of selecting target data with minimal privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replacing the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor include: Quantify ranking loss and mutual information privacy loss, and set privacy and security thresholds to constrain mutual information privacy loss; Based on the multidimensionality of the fourth-order tensor, a dimension-sensitive noise adjustment strategy is designed to make the highly sensitive, low-ranking-important dimensions receive stronger noise, while the low-sensitivity, high-ranking-important dimensions receive weaker noise. Generate candidate confusion tensors with multiple noise parameter combinations, and select the solution that satisfies the privacy constraints and minimizes the ranking loss through Pareto optimal selection; Local optimization is performed on the initial optimal solution to further reduce the ranking loss without breaking privacy constraints; The final ranking hidden tensor is deployed to a cloud-edge architecture to collect user feedback and is updated regularly.

7. The tensor-based ranking privacy protection method according to claim 6, characterized in that, In the step of designing a dimension-sensitive noise adjustment strategy based on the multidimensional characteristics of the fourth-order tensor, an independent noise parameter b is assigned to each dimension d. d The calculation formula is: ； Among them, S d For the privacy sensitivity of dimension d, R d The ranking importance of dimension d.

8. A ranking-hiding privacy protection system based on tensors, characterized in that, For implementing the tensor-based ranking-hiding privacy protection method as described in any one of claims 1-7, the system comprises: The modeling module is used to perform tensor modeling on the itemset, user set, location set, and time set to construct the original fourth-order tensor consisting of these four dimensions, with the content of the tensor being ranking data. The noise addition module is used to add square root compressed Laplace noise to the higher-order data in the fourth-order tensor to obtain the obfuscated tensor. The replacement module is used to select target data with minimum privacy leakage and ranking loss from the obfuscated tensor by calculating ranking loss and mutual information privacy loss, and replace the data in the fourth-order tensor with the target data to obtain the final ranking hidden tensor.

9. A computer-readable storage medium, characterized in that, include: The readable storage medium stores one or more programs that, when executed by a processor, implement the tensor-based ranking-hiding privacy protection method as described in any one of claims 1-7.

10. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the tensor-based ranking hiding privacy protection method according to any one of claims 1-7.

Images