Data privacy rating methods, devices, equipment, media and products

Abstract

The present application provides a data privacy rating method, device, equipment, medium and product. The method includes: obtaining ciphertext user data; inputting the ciphertext user data into a privacy rating model, in which the forward reasoning process of the first M layers of the privacy rating model is calculated to obtain the ciphertext intermediate data of the first participant, the privacy rating model is a model trained based on a large language model, and M is a positive integer; when the ciphertext intermediate data of multiple second participants are obtained, the plaintext intermediate data is reconstructed based on the ciphertext intermediate data of the multiple second participants and the ciphertext intermediate data of the first participant; the plaintext intermediate data is input into the privacy rating model, and the remaining layers of the privacy rating model are calculated in plaintext to obtain the privacy rating result of the user data. The technical problem that the high robustness and low computational transmission overhead of data privacy rating cannot be simultaneously met in the related technology is solved.

Description

Technical Field

[0001] The present application belongs to the field of privacy computing technology, and in particular, relates to a data privacy rating method, device, equipment, medium and product. Background Art

[0002] In the data analysis of multiple parties, it is necessary to classify the privacy sensitivity of a single piece of user data based on the data provided by multiple parties to determine the subsequent use of user data. However, the privacy rating of multi-party data needs to be calculated based on the data of multiple parties without disclosing the data of each party.

[0003] In related technologies, privacy computing technologies such as secure multi-party computing are usually used to rate data privacy, or data privacy rating is performed based on machine learning. However, the traditional rating method based on secure multi-party computing has poor detection robustness and is difficult to handle new data formats and irregular data; while the detection method based on machine learning has high computational and transmission overhead for model training, and it is difficult to label multi-party training data.

[0004] Therefore, it is very important to reduce the computational transmission overhead of the rating while ensuring the high robustness of the data privacy rating. Summary of the invention

[0005] The embodiments of the present application provide a data privacy rating method, apparatus, device, storage medium and computer program product to solve the technical problem in the related art that both high robustness and low computational transmission overhead of data privacy rating cannot be satisfied simultaneously.

[0006] In a first aspect, an embodiment of the present application provides a data privacy rating method, which is applied to a first participant, and the method includes:

[0007] Obtaining ciphertext user data, the ciphertext user data including a private shared fragment of a first participant and a plurality of private shared fragments of second participants, the private shared fragments being generated based on user data of each participant, the participants including the first participant and the second participant;

[0008] Input the encrypted user data into the privacy rating model. In the privacy rating model, calculate the forward reasoning process of the first M layers of the privacy rating model to obtain the encrypted intermediate data of the first participant. The privacy rating model is a model trained based on a large language model, and M is a positive integer.

[0009] When the ciphertext intermediate data of the plurality of second participants are obtained, reconstructing the plaintext intermediate data according to the ciphertext intermediate data of the plurality of second participants and the ciphertext intermediate data of the first participant;

[0010] The plaintext intermediate data is input into the privacy rating model, and the remaining layer network of the privacy rating model is calculated in plaintext to obtain the privacy rating result of the user data.

[0011] In a second aspect, an embodiment of the present application provides a data privacy rating device, which is applied to a first participant, and the device includes:

[0012] An acquisition module, used to acquire ciphertext user data, the ciphertext user data including a private shared fragment of a first participant and a plurality of private shared fragments of second participants, the private shared fragments being generated based on user data of each participant, the participants including the first participant and the second participant;

[0013] A calculation module is used to input the encrypted user data into the privacy rating model, and in the privacy rating model, calculate the forward reasoning process of the first M-layer network of the privacy rating model to obtain the encrypted intermediate data of the first participant. The privacy rating model is a model obtained by training based on the large language model, and M is a positive integer;

[0014] A reconstruction module, used for reconstructing the plaintext intermediate data according to the ciphertext intermediate data of the plurality of second participants and the ciphertext intermediate data of the first participant when the ciphertext intermediate data of the plurality of second participants are obtained;

[0015] The calculation module is also used to input the plaintext intermediate data into the privacy rating model, calculate the remaining layer network of the privacy rating model in plaintext, and obtain the privacy rating result of the user data.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, the device comprising:

[0017] A processor and a memory storing programs or instructions;

[0018] The above method is implemented when the processor executes the program or instruction.

[0019] In a fourth aspect, an embodiment of the present application provides a machine-readable storage medium, on which a program or instruction is stored, and the program or instruction implements the above method when executed by a processor.

[0020] In a fifth aspect, an embodiment of the present application provides a computer program product. When instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the above method.

[0021] This application introduces a split reasoning rating method to split the preset rating model, and only uses ciphertext calculations in the first M layers of the preset rating model. Then, the reasoning rating of the subsequent network layers of the preset rating model is completed in plain text, which reduces the number of network layers that require ciphertext calculations, thereby significantly reducing the computational overhead and transmission overhead of data privacy ratings. At the same time, since the original data cannot be inferred based on the ciphertext intermediate data, the privacy security of the data is also guaranteed. In addition, this application uses a large language model as a privacy rating model, which can effectively handle irregular data and changing data, as well as better understand the semantics of the privacy rating standard, and has high rating robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0023] Figure 1 is a flowchart of a privacy rating method applied to data of a first participant according to an embodiment of the present invention;

[0024] Figure 2 is a schematic diagram of the overall process of a data privacy rating method according to an embodiment of the present invention;

[0025] Figure 3 is a structural block diagram of a privacy rating device applied to data of a first participant according to an embodiment of the present invention;

[0026] Figure 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0028] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0029] In addition, it should be noted that the acquisition, storage, use, and processing of data in the embodiments of this application are in compliance with the relevant provisions of national laws and regulations. It should be noted that in the embodiments of this application, some existing solutions in the industry such as certain software, components, and models may be mentioned, and they should be considered as exemplary. Their purpose is only to illustrate the feasibility of the implementation of the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0030] The following are explanations of the terms in the embodiments of the present application:

[0031] (1) Secure Multi-party Computation (MPC) is a cryptographic communication protocol with two or more participants. Each participant provides a portion of private input to complete a certain computing goal (function), and the calculation result is made public to each participant. MPC can ensure that the original data privacy of each participant will not be leaked during the protocol interaction process.

[0032] (2) Private Set Intersection (PSI), a special type of secure multi-party computation protocol that allows two or more participants to determine the elements of their common sets without revealing the specific contents of their respective sets.

[0033] (3) Secret Share: A data encryption method that splits data into several random fragments (secret shares) and then distributes them to several participants; each fragment appears to be uniformly randomly distributed, but when combined, the original data information can be restored.

[0034] Secret sharing fragment (Share), a single data fragment (share) in secret sharing, individually appears to be uniformly randomly distributed.

[0035] (4) Large Language Model (LLM) is an AI technology that is trained through deep learning algorithms to understand and generate natural language text. These models typically have hundreds of millions of parameters, enabling them to capture the complexity and nuances of language. Large language models can perform a variety of tasks, such as language translation, text summarization, sentiment analysis, question-answering systems, and text generation.

[0036] (5) Prompt: An LLM input that guides or triggers the model to perform a specific task or generate a specific output. Prompts can be questions, instructions, sentences, or paragraphs. Their function is to stimulate the model to respond in the expected manner.

[0037] In order to solve the problems of the prior art, an embodiment of the present application provides a data privacy rating method, which can be applied in a scenario of data privacy rating with multiple participants. Figure 1 is a flow chart of a privacy rating method applied to data of a first participant according to an embodiment of the present invention. Figure 1 As shown, the privacy rating method of the data may include the following steps:

[0038] S110, obtaining encrypted user data.

[0039] In S110, the ciphertext user data includes private shared fragments of a first participant and private shared fragments of multiple second participants. The private shared fragments can be generated based on the user data of each participant. The participants may include the first participant and the second participant. By way of example, the participants may be platforms that hold user privacy data, such as e-commerce platforms and social platforms. The user privacy data may include the user's gender, age, contact information and other data.

[0040] Through decentralized storage and encryption processing, it is ensured that no single party can decrypt or obtain complete user data alone, thus protecting user data privacy.

[0041] S120, inputting the encrypted user data into the privacy rating model, in which the forward reasoning process of the first M layers of the privacy rating model is calculated to obtain the encrypted intermediate data of the first participant, where M is a positive integer.

[0042] In S120, the privacy rating model may be a model trained based on a large language model, and the large language model may be an open source basic large language model; the first M layers of the privacy rating model are network layers that can process encrypted data or perform calculations in an encrypted state; the obtained ciphertext intermediate data is also data in an encrypted state.

[0043] In this way, sufficient information can be extracted from the ciphertext user data for subsequent privacy rating without exposing the original user data.

[0044] S130, when the ciphertext intermediate data of multiple second participants are obtained, reconstruct the plaintext intermediate data according to the ciphertext intermediate data of the multiple second participants and the ciphertext intermediate data of the first participant.

[0045] In S130, a multi-party computing protocol (such as secure multi-party computing, homomorphic encryption, etc.) can be used to first combine the ciphertext intermediate data of the first participant and multiple second participants, and perform decryption processing to reconstruct the plaintext intermediate data.

[0046] In an example, the ciphertext intermediate data of the first participant is z0, and the ciphertext intermediate data of the second participant is z1. z0 and z1 are combined to obtain z=z0+z1.

[0047] In this way, it can be ensured that all participants can collaborate to generate plaintext intermediate data without decrypting their own original data, thereby maintaining the privacy of the data.

[0048] S140, input the plaintext intermediate data into the privacy rating model, calculate the remaining layer network of the privacy rating model in plaintext, and obtain the privacy rating result of the user data.

[0049] In S140, since plaintext calculation can make more full use of existing computing resources and perform algorithm optimization, the remaining layer network of the privacy rating model is calculated by plaintext, so the privacy rating result obtained can be more accurate. At the same time, plaintext calculation takes less time and has higher computational efficiency than ciphertext calculation.

[0050] In this way, the accuracy and efficiency of privacy rating results are improved while ensuring data privacy.

[0051] In an embodiment of the present application, a split inference rating method is introduced to split the preset rating model, and ciphertext calculation is only used in the first M layers of the preset rating model. Then, the inference rating of the subsequent network layers of the preset rating model is completed in plain text, which reduces the number of network layers that require ciphertext calculation, thereby significantly reducing the computational overhead and transmission overhead of data privacy rating. At the same time, since the original data cannot be inferred based on the ciphertext intermediate data, the privacy security of the data is also guaranteed. In addition, the present application uses a large language model as a privacy rating model, which can effectively handle irregular data and changing data, as well as better understand the semantics of the privacy rating standard, and has a high rating robustness.

[0052] In some embodiments, obtaining the encrypted user data includes:

[0053] Matching the user data of the first participant with the user data of the plurality of second participants through a privacy intersection protocol to determine the intersection user data that is in the same intersection as the user data of the first participant;

[0054] Acquire private sharing fragments of multiple second participants, where the private sharing fragments of the multiple second participants are generated based on private sharing of intersection user data by the multiple second participants;

[0055] The private sharing fragments of multiple second participants are spliced ​​with the private sharing fragments of the first participant to obtain ciphertext user data. The private sharing fragments of the first participant are privately shared and generated based on the user data of the first participant.

[0056] In this embodiment, the Private Set Intersection (PSI) protocol can be used to match the user data of the first participant and the user data of multiple second participants. The Private Set Intersection protocol allows both parties to discover common elements (i.e., intersection user data) between them without exposing their respective complete data sets. In this way, only user data that exists in the data sets of both parties or multiple parties will be further processed, thereby protecting the privacy of non-intersection user data.

[0057] Then, for the intersection user data determined by the privacy intersection protocol, each participant can generate private shared fragments based on the intersection user data it owns. These private shared fragments can be generated through some form of secret sharing scheme (such as Shamir secret sharing, additive secret sharing, etc.), which cannot restore the original data alone, but when all related fragments are correctly combined, the original data or some form of useful information can be restored.

[0058] Finally, after obtaining all the relevant private shared fragments, these fragments can be spliced ​​together. Since each fragment contains part of the original data, and this information is complementary between fragments, when all fragments are correctly spliced ​​together, the complete ciphertext user data can be restored. The ciphertext user data is still encrypted, and only the party with the correct key can decrypt and access the plaintext form of the data.

[0059] In this way, multiple parties can share and use data while protecting their respective data privacy.

[0060] In some embodiments, before inputting the ciphertext user data into the privacy rating model, the method further comprises:

[0061] Generate simulated user data according to preset privacy rating standards;

[0062] The simulated user data is input into the large language model to fine-tune the large language model in plaintext to obtain a privacy rating model.

[0063] In this embodiment, the preset privacy rating standard can be a public privacy grading standard, which is not limited. The large language model can be directly fine-tuned to obtain a fine-tuned privacy rating model, or the plaintext fine-tuning parameters of the large language model can be fine-tuned and output, and the plaintext fine-tuning parameters are sent to each participant, and each participant chooses whether to perform plaintext fine-tuning on the large language model according to the plaintext fine-tuning parameters.

[0064] In this way, by generating simulated user data according to the preset privacy rating standards and using the simulated user data to fine-tune the large language model, the large language model can learn the relevant concepts in the field of data privacy and the paradigm of data privacy classification standards. And by using simulated data, the large language model can be calculated in plain text, which will not pose a threat to the data privacy of the participants. In addition, due to the robustness of the large language model, there is no need for massive training data, thus reducing the cost of manual data annotation and the risk of privacy leakage in multi-party scenarios.

[0065] In some embodiments, the encrypted user data is input into a privacy rating model, and in the privacy rating model, the forward reasoning process of the first M-layer network of the privacy rating model is calculated to obtain the encrypted intermediate data of the first participant, including:

[0066] Acquire prompt word data, where the prompt word data is used to represent a rating prompt for user data;

[0067] The ciphertext user data and prompt word data are input into the privacy rating model, and the first M layers of the privacy rating model are forward-reasoned through the secure multi-party computing protocol to obtain the ciphertext intermediate data of the first participant.

[0068] In this embodiment, the prompt word data can be a set of keywords or phrases used to guide the privacy rating model to rate user data. It can be formulated based on factors such as user preferences or business needs, and there is no limitation on this. In one example, when evaluating the sensitivity of user data, the prompt words may include "ID number", "bank account information", etc.

[0069] During the forward reasoning process of the privacy rating model, a secure multi-party computation protocol can be used to ensure the privacy of the calculation. Secure multi-party computation allows multiple parties to jointly calculate the result of a function without exposing their respective input data.

[0070] In this way, through the secure multi-party computing protocol, the first M layers of the model can be securely calculated without leaking the private data of any party.

[0071] In some embodiments, the plaintext intermediate data is input into the privacy rating model, and the remaining layer network of the privacy rating model is calculated in plaintext to obtain the privacy rating result of the user data, including:

[0072] The plaintext intermediate data is input into the privacy rating model, and the remaining layer network of the privacy rating model is calculated in plaintext to obtain the privacy rating result of the user data and the privacy risk prompt corresponding to the privacy rating result.

[0073] In this embodiment, in order to balance privacy protection and computational efficiency, and because the subsequent layers of the model may require more complex data computation or data processing, which may not be effectively performed directly on the encrypted data, the ciphertext intermediate data can be decrypted to obtain the plaintext intermediate data, and the plaintext intermediate data is input into the privacy rating model to calculate the remaining layers of the privacy rating model in plaintext.

[0074] Thus, since encryption calculations are usually more resource-intensive than plaintext calculations, the solution of this embodiment can reduce the pressure of encryption calculations on system resources by transferring part of the calculation tasks to the plaintext stage.

[0075] In some embodiments, after obtaining the privacy rating result of the user data and the privacy risk prompt corresponding to the privacy rating result, the method further includes:

[0076] obtaining calibration data input by the first party in response to the privacy risk prompt;

[0077] Correct the privacy rating results based on the calibration data.

[0078] In this embodiment, the system can display the privacy risk prompt to the first participant. After viewing the privacy risk prompt, the first participant can input calibration data to calibrate the privacy rating result according to its own business logic or actual situation. For example, the calibration data may include an explanation of the specific content of the user data, an explanation of the user data usage scenario, or an understanding of the privacy rating standard. After the system receives the calibration data, it can proofread the previous privacy rating results based on the calibration data. For example, the privacy rating model first analyzes the calibration data provided by the first participant, and then re-evaluates the previous privacy rating process based on the calibration data. This may include reconsidering the sensitivity of certain data, adjusting the privacy rating standards or weights, etc.

[0079] In this way, the privacy rating model can not only provide objective and accurate privacy rating results and privacy risk warnings, but also be dynamically adjusted and optimized based on the feedback from the first participant, thereby improving the practicality and accuracy of the privacy rating model.

[0080] Figure 2FIG. 1 is a schematic diagram of the overall process of a data privacy rating method according to an embodiment of the present invention. Figure 2 As shown, including:

[0081] S210, each participant performs plaintext fine-tuning on the open source Large Language Model (LLM) based on the open user data privacy concept, sample privacy data, and grading paradigm. During the fine-tuning, simulated sample privacy data generated based on the public privacy grading standard may be used.

[0082] In S210 , the preset privacy rating standard may include a public user data privacy concept, sample privacy data, and a grading paradigm.

[0083] S220, all participating parties collaborate to run a Private Set Intersection (PSI) protocol to match the user data they hold and filter out the user data that is in the public intersection.

[0084] In S220, the user data located in the common intersection is screened out and may correspond to the intersection user data of the present application.

[0085] S230, each participant sends its own user data in the intersection, i.e., the data to be classified for privacy, to other participants in the form of Secret Share;

[0086] S240, for each user in the intersection, each participant combines the secret sharing fragments received from other participants in the previous step with the local secret sharing fragments to obtain the user data in the complete secret sharing state;

[0087] In S240, the user data in the complete secret sharing state may correspond to the ciphertext user data of the present application.

[0088] S250, each participant runs the secure multi-party computation (MPC) protocol to calculate the forward reasoning process of the first m layers of the LLM network, where the input of the calculation is divided into four parts: the multi-party user data of the secret sharing obtained in the previous step, the privacy classification prompt, the privacy classification standard prompt, and the potential privacy risk prompt corresponding to the privacy classification result. Output the hidden representation of the data in the middle layer under the secret sharing state;

[0089] In S250, the secret shared multi-party user data may correspond to the ciphertext user data of the present application, the privacy grading prompt words and the privacy grading standard prompt words may correspond to the prompt word data of the present application, and the potential privacy risk prompt words may be used to enable the privacy rating model to output a privacy risk prompt corresponding to the privacy rating result.

[0090] S260, each participant sends the secret shared fragments of the sample hidden representation output by the first m layers of the network to each other, and reconstructs the sample hidden representation of the plaintext;

[0091] In S260, the secret sharing fragment of the sample hidden representation may correspond to the ciphertext intermediate data of the present application, and the plaintext sample hidden representation may correspond to the plaintext intermediate data of the present application.

[0092] S270, each participant locally completes the calculation of the remaining LLM network layers based on the plaintext sample hidden representation of the previous step, and outputs the privacy sensitivity classification results and data privacy risk prompts.

[0093] In S270, the privacy sensitivity grading result may correspond to the privacy rating result of the present application.

[0094] In one example, e-commerce platforms and social networking platforms can be used to conduct multi-party user data privacy ratings, where the e-commerce platform contributes the user's shopping history text information, the social networking platform contributes the user's historical public speech records, and the user is indexed by phone number. It is understandable that when the user's e-commerce platform data or social networking platform data is disclosed separately, it is not easy to cause the user's privacy to be leaked, because this information cannot be accurately matched to the individual. When the two aspects of information are combined, the individual matching range may be greatly reduced, causing serious user privacy leakage. Therefore, it is necessary for multiple parties to jointly conduct user data privacy ratings to determine the subsequent user data analysis and processing methods.

[0095] The following is an explanation of this solution based on this case: Based on the privacy intersection protocol, the user mobile phone numbers submitted by the e-commerce platform and the social platform are matched to obtain the users in the intersection; for each user in the intersection, the e-commerce platform splits the data into two secret sharing fragments (x0, x1), and sends one of the fragments x1 to the social platform; correspondingly, the social platform splits the data into two secret fragments (y0, y1), and sends one of the fragments y0 to the e-commerce platform. After local data splicing, the e-commerce platform holds (x0, y0) and the social platform holds (x1, y1). That is, both parties hold the complete user data in the secret sharing state; using the user data in the secret sharing state obtained in the previous step as input, the two platforms collaborate to run the MPC protocol, calculate the first m layers of the LLM model, and output the middle-layer hidden representation of the user data in the secret sharing state, where the e-commerce platform holds z0 and the social platform holds z1. The e-commerce platform sends z0 to the social platform, and the social platform sends z1 to the e-commerce platform. Both parties complete the local merging of the secret sharing fragments z=z0+z1, that is, both obtain the middle-layer hidden representation of the plaintext user data. Both parties complete the calculation of the remaining LLM model layers in local plaintext state, and output the privacy classification results and the corresponding privacy risk prompts.

[0096] Figure 3 A schematic structural diagram of a privacy rating device for data of a first participant provided in another embodiment of the present application is shown. For ease of explanation, only the portion related to the embodiment of the present application is shown.

[0097] Reference Figure 3 , the data privacy rating device 300 may include:

[0098] An acquisition module 301 is used to acquire ciphertext user data, where the ciphertext user data includes a private shared fragment of a first participant and a plurality of private shared fragments of second participants, where the private shared fragments are generated based on user data of each participant, where the participants include the first participant and the second participant;

[0099] A calculation module 302 is used to input the encrypted user data into a privacy rating model, and in the privacy rating model, calculate the forward reasoning process of the first M-layer network of the privacy rating model to obtain the encrypted intermediate data of the first participant, where the privacy rating model is a model obtained by training based on a large language model, and M is a positive integer;

[0100] The reconstruction module 303 is used to reconstruct the plaintext intermediate data according to the ciphertext intermediate data of the plurality of second participants and the ciphertext intermediate data of the first participant when the ciphertext intermediate data of the plurality of second participants are obtained;

[0101] The calculation module 302 is also used to input the plaintext intermediate data into the privacy rating model, and to calculate the remaining layer network of the privacy rating model in plaintext to obtain the privacy rating result of the user data.

[0102] In some embodiments, the acquisition module 301 may include the following units:

[0103] A matching unit, configured to match the user data of the first participant with the user data of the plurality of second participants through a privacy intersection protocol, and determine intersection user data that is in the same intersection as the user data of the first participant;

[0104] An acquisition unit, configured to acquire private sharing fragments of a plurality of second participants, wherein the private sharing fragments of the plurality of second participants are generated based on private sharing of intersection user data by the plurality of second participants;

[0105] The splicing unit is used to splice the private sharing fragments of multiple second participants with the private sharing fragments of the first participant to obtain ciphertext user data. The private sharing fragments of the first participant are privately shared and generated based on the user data of the first participant.

[0106] In some embodiments, the data privacy rating device 300 may further include the following modules:

[0107] A generation module, used to generate simulated user data according to a preset privacy rating standard before inputting the ciphertext user data into the privacy rating model;

[0108] The fine-tuning module is used to input simulated user data into the large language model to perform plaintext fine-tuning on the large language model to obtain a privacy rating model.

[0109] In some embodiments, the calculation module 302 may include the following units:

[0110] The acquisition unit is further used to acquire prompt word data, where the prompt word data is used to represent a rating prompt for the user data;

[0111] The computing unit is used to input the ciphertext user data and the prompt word data into the privacy rating model, and perform forward reasoning calculations on the first M-layer network of the privacy rating model through a secure multi-party computing protocol to obtain the ciphertext intermediate data of the first participant.

[0112] In some embodiments, the calculation module 302 may further include the following units:

[0113] The computing unit is also used to input the plaintext intermediate data into the privacy rating model, plaintext calculate the remaining layer network of the privacy rating model, obtain the privacy rating result of the user data, and the privacy risk prompt corresponding to the privacy rating result.

[0114] In some embodiments, the data privacy rating device 300 may include the following modules:

[0115] The acquisition module 301 is further used to acquire calibration data input by the first participant in response to the privacy risk prompt;

[0116] The calibration module is used to calibrate the privacy rating results based on the calibration data.

[0117] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application, and are devices corresponding to the privacy rating method of the above-mentioned data. All implementation methods in the above-mentioned method embodiment are applicable to the embodiment of the device. Its specific functions and technical effects can be found in the method embodiment part, which will not be repeated here.

[0118] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0119] Figure 4 is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Figure 4 As shown:

[0120] The device may include a processor 401 and a memory 402 storing programs or instructions.

[0121] When the processor 401 executes the program, the steps in any of the above method embodiments are implemented.

[0122] Exemplarily, the program may be divided into one or more modules / units, one or more modules / units are stored in the memory 402, and executed by the processor 401 to complete the present application. One or more modules / units may be a series of program instruction segments capable of completing a specific function, and the instruction segments are used to describe the execution process of the program in the device.

[0123] Specifically, the processor 401 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0124] The memory 402 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 402 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 402 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 402 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 402 is a non-volatile solid-state memory.

[0125] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical or other physical / tangible memory storage devices. Thus, typically, the memory includes one or more tangible (non-transitory) machine-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to an aspect of the present disclosure.

[0126] The processor 401 implements any one of the methods in the above embodiments by reading and executing the program or instruction stored in the memory 402 .

[0127] In one example, the electronic device may further include a communication interface 403 and a bus 404. The processor 401, the memory 402, and the communication interface 403 are connected via the bus 404 and communicate with each other.

[0128] The communication interface 403 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0129] Bus 404 includes hardware, software or both, and couples the components of online data traffic billing equipment to each other. For example, but not limitation, the bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnect (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. Where appropriate, bus 404 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the present application considers any suitable bus or interconnection.

[0130] In addition, in combination with the method in the above embodiment, the embodiment of the present application may provide a machine-readable storage medium for implementation. The machine-readable storage medium stores a program or instruction; when the program or instruction is executed by a processor, any one of the methods in the above embodiment is implemented. The machine-readable storage medium can be read by a machine such as a computer.

[0131] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0132] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0133] An embodiment of the present application provides a computer program product, which is stored in a machine-readable storage medium. The program product is executed by at least one processor to implement the various processes of the above-mentioned method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0134] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0135] The functional modules shown in the above block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer grid such as the Internet, an intranet, etc.

[0136] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0137] Aspects of the present disclosure are described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by a computer program or instruction. These programs or instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the functions / actions specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0138] The above are only specific implementation methods of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A data privacy rating method, characterized in that: Applied to a first participant, the method comprises: Acquire ciphertext user data, the ciphertext user data including a private shared fragment of a first participant and a plurality of private shared fragments of second participants, the private shared fragments being generated based on user data of each participant, the participants including the first participant and the second participant; Input the ciphertext user data into a privacy rating model, and in the privacy rating model, calculate the forward reasoning process of the first M-layer network of the privacy rating model to obtain the ciphertext intermediate data of the first participant, wherein the privacy rating model is a model obtained by training based on a large language model, and M is a positive integer; When the ciphertext intermediate data of the plurality of second participants are obtained, reconstructing the plaintext intermediate data according to the ciphertext intermediate data of the plurality of second participants and the ciphertext intermediate data of the first participant; Inputting the plaintext intermediate data into the privacy rating model, calculating the remaining layer network of the privacy rating model in plaintext, and obtaining a privacy rating result of the user data; The obtaining of ciphertext user data includes: Matching the user data of the first participant with the user data of the plurality of second participants through a privacy intersection protocol to determine intersection user data that is in the same intersection as the user data of the first participant; Acquire private sharing fragments of the plurality of second participants, where the private sharing fragments of the plurality of second participants are generated based on private sharing of the intersection user data by the plurality of second participants; The private sharing fragments of the multiple second participants are concatenated with the private sharing fragments of the first participant to obtain the ciphertext user data, and the private sharing fragments of the first participant are privately shared and generated based on the user data of the first participant.

2. The method according to claim 1, characterized in that Before inputting the encrypted user data into a privacy rating model, the method further includes: Generate simulated user data according to preset privacy rating standards; The simulated user data is input into the large language model to perform plaintext fine-tuning on the large language model to obtain the privacy rating model.

3. The method according to claim 1, characterized in that: The step of inputting the ciphertext user data into a privacy rating model, and calculating the forward reasoning process of the first M-layer network of the privacy rating model in the privacy rating model to obtain the ciphertext intermediate data of the first participant includes: Acquire prompt word data, where the prompt word data is used to represent a rating prompt for user data; The ciphertext user data and the prompt word data are input into the privacy rating model, and the first M layers of the privacy rating model are forward-reasoned through a secure multi-party computing protocol to obtain the ciphertext intermediate data of the first participant.

4. The method according to claim 1, characterized in that: The step of inputting the plaintext intermediate data into the privacy rating model, calculating the remaining layer network of the privacy rating model in plaintext, and obtaining the privacy rating result of the user data includes: The plaintext intermediate data is input into the privacy rating model, and the remaining layer network of the privacy rating model is calculated in plaintext to obtain a privacy rating result of the user data and a privacy risk prompt corresponding to the privacy rating result.

5. The method according to claim 4, characterized in that After obtaining the privacy rating result of the user data and the privacy risk prompt corresponding to the privacy rating result, the method further includes: obtaining calibration data input by the first participant in response to the privacy risk prompt; The privacy rating result is calibrated according to the calibration data.

6. A data privacy rating device, characterized in that: Applied at a first participant, the device comprises: an acquisition module, configured to acquire ciphertext user data, wherein the ciphertext user data includes a private shared fragment of a first participant and private shared fragments of a plurality of second participants, wherein the private shared fragments are generated based on user data of each participant, wherein the participants include the first participant and the second participant; a calculation module, configured to input the ciphertext user data into a privacy rating model, and calculate, in the privacy rating model, a forward reasoning process of a first M-layer network of the privacy rating model to obtain the ciphertext intermediate data of the first participant, wherein the privacy rating model is a model obtained by training based on a large language model, and M is a positive integer; a reconstruction module, configured to reconstruct the plaintext intermediate data according to the ciphertext intermediate data of the plurality of second participants and the ciphertext intermediate data of the first participant when the ciphertext intermediate data of the plurality of second participants are obtained; The calculation module is further used to input the plaintext intermediate data into the privacy rating model, calculate the remaining layer network of the privacy rating model in plaintext, and obtain the privacy rating result of the user data; Wherein, the acquisition module includes the following units: A matching unit, configured to match the user data of the first participant with the user data of the plurality of second participants through a privacy intersection protocol, and determine intersection user data that is in the same intersection as the user data of the first participant; An acquisition unit, configured to acquire private sharing fragments of a plurality of second participants, wherein the private sharing fragments of the plurality of second participants are generated based on private sharing of intersection user data by the plurality of second participants; The splicing unit is used to splice the private sharing fragments of multiple second participants with the private sharing fragments of the first participant to obtain ciphertext user data. The private sharing fragments of the first participant are privately shared and generated based on the user data of the first participant.

7. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 5.

9. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method according to any one of claims 1 to 5.

Images