Data processing method and device, and storage medium
The method and apparatus for data processing with LLMs address data privacy issues by concealing sensitive data categories, ensuring secure and efficient analysis through one-to-one correspondence and encryption, thereby enhancing user trust and utilization of LLM services.
Patent Information
- Application Number
- JP2025079675
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-05-20
- Filing Date
- 2025-05-12
- Publication Date
- 2025-12-03
AI Technical Summary
Existing large-scale language models (LLMs) face a contradiction between data analysis capabilities and data privacy protection, leading to concerns about data leakage and misuse when processing sensitive user data, which hinders their widespread use.
A method and apparatus that utilize a large-scale language model to determine and conceal sensitive data categories while maintaining data meaning and distribution, allowing for secure data processing and analysis by establishing a one-to-one correspondence between original and concealed data values, and using techniques like pseudonymous mapping, randomization, and encryption.
Enables efficient and accurate data processing while ensuring user data privacy, promoting the use of LLM services by addressing data security concerns.
Smart Images
Figure 2025175965000001_ABST
Abstract
Description
[Technical Field]
[0001] The present disclosure relates generally to the field of information services, and more particularly to a method and apparatus for processing data using large-scale language models, and a storage medium having an associated program stored thereon. [Background technology]
[0002] With the development of artificial intelligence technology, especially the emergence of large-scale language models (LLMs) such as GPT and Claude, machine understanding has reached and, in some respects, exceeded that of ordinary humans. LLMs offer powerful capabilities and have great potential for data analysis and understanding generation. LLMs can be used to analyze data, for example, to analyze data patterns and trends, to draw graphs to visualize data, and to automatically generate program code for processing data.
[0003] Currently, powerful LLMs are primarily provided to users through APIs based on cloud services. However, locally deployable open-source models often suffer from poor performance and are unable to meet user needs. Therefore, users typically need to send their data to commercial companies offering LLM services for processing. However, this data transmission raises data privacy concerns. In business environments, data often contains highly confidential business and personal information, such as customer information, product information, and transaction records. Sending data to a third-party company for processing can increase the risk of data leakage and misuse. These risks discourage users from sending their sensitive data to LLMs for processing, limiting their use of LLM services and hindering the further application of artificial intelligence technology.
[0004] As mentioned above, existing LLM services face a contradiction between data analysis capabilities and data privacy protection. How to fully utilize the capabilities of LLM for data analysis while protecting users' data privacy is an issue that needs to be resolved urgently. Summary of the Invention [Problem to be solved by the invention]
[0005] The present invention aims to propose a method for processing data using LLM, which can provide users with safer and more reliable artificial intelligence services by using LLM to perform efficient and accurate data processing and analysis while protecting users' data privacy. [Means for solving the problem]
[0006] In one aspect of the present disclosure, there is provided a computer-implemented method for processing data using a large-scale language model, the method including the steps of: inputting a plurality of data categories into the large-scale language model and instructing the large-scale language model to determine one or more data categories that need to be concealed from the plurality of data categories to protect data privacy, wherein each data category includes a plurality of data values; and for each data category among the one or more data categories that need to be concealed, inputting criteria for concealing the data category into the large-scale language model and instructing the large-scale language model to provide a plurality of candidate methods for concealing the data category according to the criteria, wherein the large-scale language model selects one of the plurality of candidate methods as a concealment method according to the criteria, wherein the criteria include at least that the concealed data values of the data category correspond one-to-one to original data values, and that the concealed data values of the data category match the original data values in data meaning and data distribution.
[0007]
[0013] In another aspect of the present disclosure, there is provided an apparatus for processing data in communication with a large-scale language model, the apparatus including: a memory having a program stored therein; and one or more processors, wherein the processor is configured to execute the program to perform a method including: inputting a plurality of data categories to the large-scale language model and instructing the large-scale language model to determine one or more data categories from the plurality of data categories that need to be concealed to protect data privacy, wherein each data category includes a plurality of data values; and, for each data category among the one or more data categories that need to be concealed, inputting criteria for concealing the data category to the large-scale language model and instructing the large-scale language model to provide a plurality of candidate methods for concealing the data category according to the criteria, wherein the large-scale language model selects one of the plurality of candidate methods as a concealment method according to the criteria, wherein the criteria include at least that there is a one-to-one correspondence between the concealed data values of the data category and original data values, and that the concealed data values of the data category and the original data values match in data semantics and data distribution.
[0008] Another aspect of the present disclosure provides a non-transitory computer-readable storage medium having a program stored thereon, the program causing the computer to perform the above-described method for processing data when executed by the computer. [Brief explanation of the drawings]
[0009] [Figure 1] 1 is a block diagram illustrating a schematic diagram of an apparatus for determining a concealment method according to the present disclosure; [Figure 2] 10 is a flowchart of a process for determining a concealment method according to the present disclosure. [Figure 3] 1 is a block diagram illustrating a schematic diagram of an apparatus for processing data using an LLM according to the present disclosure. [Figure 4]1 is a flowchart of a process for processing data using an LLM according to the present disclosure. [Figure 5] 1 is a flowchart of a process for automatically verifying a hidden program according to the present disclosure. [Figure 6] FIG. 1 is a block diagram illustrating an exemplary configuration of computer hardware capable of implementing the present disclosure. DETAILED DESCRIPTION OF THE INVENTION
[0010] In order to enable users to process data using LLM while protecting the privacy of the data, this disclosure proposes that the user first conceals the original data that he or she wishes to process, provides the concealed data to the LLM for processing, and then performs the reverse operation of the concealment method to restore the original data from the processing result provided by the LLM, thereby obtaining the processing result for the original data.
[0011] 1 is a block diagram illustrating an apparatus for determining a concealment method for concealing data according to the present disclosure. As shown in FIG. 1, the apparatus 100 includes a data category determination module 110, a concealment method determination module 120, and a preferred concealment program generation module 130, which are implemented based on a large-scale language model (LLM). Note that these modules refer to logical functional modules, not physical entities, and it may be understood that the LLM is included in these modules. These modules will be described in detail below.
[0012] As shown in FIG. 1 , a user inputs multiple data categories into device 100. For ease of explanation, this disclosure uses the retail industry as an example and lists the following exemplary data categories: consumer identity, gender and age, product name and sale price, and products viewed or purchased by the consumer. Each data category may include multiple data values. For example, the data category "gender" may include data values of "male" and "female." However, this disclosure is not limited to the retail industry and is not limited to these exemplary data categories and data values.
[0013] The user may also input criteria for hiding data categories into the device 100. The following describes the criteria in more detail.
[0014] Preferably, the user may also input other specific requests into the device 100 .
[0015] The data category determination module 110 provides a plurality of data categories input by the user to the LLM, and instructs the LLM to determine from the plurality of data categories data categories that need to be concealed to protect data privacy. Preferably, the data category determination module 110 may then present the data categories that need to be concealed, as determined by the LLM, to the user.
[0016] Such a scheme is particularly advantageous when the user is not very familiar with the domain to which the data belongs (e.g., retail) and is therefore unable to determine which data categories are important or sensitive and therefore need to be protected. In this case, the LLM's knowledge of the relevant domain may be utilized to recommend to the user data categories that need to be protected. As an example, the data category determination module 110 may instruct the LLM, as an expert in retail, to determine which of the data categories provided by the user need to be protected, and present the data categories determined by the LLM to the user.
[0017] Preferably, the data category determination module 110 may provide other specific requests input by the user to the LLM, instructing the LLM to determine the data categories that need to be obscured based on the specific requests. For example, if the user indicates that consumer information needs to be obscured, the LLM determines that data categories that include at least the consumer information need to be obscured.
[0018] In the present disclosure, data categories are concealed in accordance with at least the following criteria: (1) The data values of a data category after concealment correspond one-to-one to the original data values; (2) The data values of a data category after concealment match the original data values in terms of data meaning and data distribution. Criterion (1) can ensure that the original data values can be uniquely restored based on the data values after concealment; and Criterion (2) can ensure that the data values after concealment do not deviate from the meaning and distribution of the original data values.
[0019] Preferably, other rules that must be followed for concealment may also be set, for example, one or more of the following rules may be set:
[0020] The data value after hiding and the original data value match in data type. In the present disclosure, the data type may include, for example, numbers, alphabets, symbols, character strings, dates and times, or combinations thereof.
[0021] - The data value after concealment deviates from the original data value by a fixed amount.
[0022] The original data values are divided into multiple groups, and in each group, the data values after hiding correspond one-to-one to the original data values.
[0023] It should be noted that, when setting the rules that need to be followed for concealment, the present disclosure is not limited to the above-listed criteria (1) and (2) and other exemplary rules, and a person skilled in the art may design appropriate rules according to the actual situation.
[0024] In addition, in the present disclosure, methods for hiding data values may include, for example, pseudonymous mapping, randomization, encryption, hashing, tokenization, substitution, format-preserving encryption (FPE), character shuffling, etc. Note that the present disclosure is not limited to these hiding methods listed as examples, and those skilled in the art may adopt appropriate hiding methods according to actual situations as long as the above-mentioned criteria or rules are met.
[0025] The following describes the concealment standards / rules and concealment methods according to the present disclosure with reference to some specific examples.
[0026] Example 1: The data category is "Consumer Identification Information (ID)" and the original data value is a combination of numbers and alphabets. In this case, it may be required that the data value after concealment maintains its data type. Therefore, the data value after concealment must be a combination of numbers and alphabets. As a concealment method, for example, pseudonym mapping may be adopted.
[0027] Example 2: The data category is "consumer's age" and the original data value is a number. In this case, it may be required that the data value after concealment maintains the consistency of the data type and the data meaning (age). Therefore, the data value after concealment needs to be a number within the reasonable range of human age. For example, the following methods may be adopted as concealment methods.
[0028] (1) Divide the numerical range from 0 to N (where N is the maximum scientifically verified human age) into multiple intervals, each of which has a length of K (e.g., 10).
[0029] (2) Classify the original data value into the corresponding interval. For example, classify the original age "35 years old" into the interval [30, 39].
[0030] (3) The order of the original age values in each interval is changed to obtain a sequence of age values after concealment. For example, the original age sequence that corresponds one-to-one to consumer 1, consumer 2, consumer 3, and consumer 4 is "32 years old, 35 years old, 33 years old, 38 years old", and after shuffling, the sequence after concealment becomes "35 years old, 38 years old, 32 years old, 33 years old".
[0031] This method can ensure that the data value after concealment is still within the range of natural human age, i.e., the data meaning of the original data value is not changed. In addition, the original data value is only changed within an interval of, for example, 10 years, so that the data value after concealment and the original data value do not differ significantly in terms of the overall distribution of all ages.
[0032] Example 3: The data category is "product name" and the original data value is a word. In this case, the data value after hiding may be required to maintain the consistency of the data type. For this reason, the data value after hiding must be a word. For example, the following hiding method may be adopted.
[0033] (1) Generate a list of random words. The length of the list depends on the total number of products. For example, list length x (list length - 1) / 2 > total number of products.
[0034] (2) Extract two words from the generated list to form the concealed product name.
[0035] Returning to FIG. 1, the concealment method determination module 120 includes a candidate method determination module 1201 and a concealment method selection module 1202 .
[0036] For one of the data categories that need to be protected as determined by the above data category determination module 110, the candidate method determination module 1201 provides the LLM with the concealment criteria or rules for that data category provided by the user, and instructs the LLM to provide multiple candidate concealment methods that meet the criteria or rules.
[0037] When the LLM provides multiple candidate concealment methods, the concealment method selection module 1202 instructs the LLM to select one of the multiple candidate methods as a concealment method according to the concealment criteria or rules. In this process, the LLM re-evaluates the multiple candidate concealment methods according to the concealment criteria or rules for the data category, and determines the optimal concealment method for the data category from the multiple candidate concealment methods.
[0038] In addition to the concealment criteria or rules, the concealment method determination module 120 may also provide other specific requirements from the user to the LLM, instructing the LLM to provide a concealment method taking the specific requirements into consideration. For example, if the user specifies a data format that the data value after concealment should conform to, the LLM will provide a concealment method that meets the requirements.
[0039] In addition, the LLM may not have knowledge about the distribution of the original data values. In this case, the concealment method determination module 120 may provide the LLM with distribution information of the original data values provided by the user, so that the LLM determines the concealment method according to the criterion (2) of consistency of data distribution.
[0040] 1 preferably further includes a concealment program generation module 130. The concealment program generation module 130 instructs the LLM to generate a concealment program for the corresponding data category according to the concealment method determined by the concealment method determination module 120. The concealment program generation module 130 may also instruct the LLM to generate program code that meets other specific requirements of the user.
[0041] In particular, in the present disclosure, a concealment program may be manually generated according to the concealment method determined by the concealment method determination module 120. For example, if the concealment method is complex, the success rate of automatic program code generation by the LLM may be reduced, and therefore the program code must be manually generated. In this case, the concealment program generation module 130 in FIG. 1 may be omitted.
[0042] Once a corresponding concealment program has been generated for each determined data category that needs to be concealed, these concealment programs may be stored for use in the processes described below with reference to FIGS.
[0043] 2 is a flowchart of a process for determining a concealment method according to the present disclosure. As shown in FIG. 2, in step S210, a plurality of data categories are input to the LLM, and the LLM is instructed to determine a data category that needs to be concealed from the plurality of data categories. This step may be performed, for example, by the data category determination module 110 in FIG. 1.
[0044] In step S220, for the determined data category that needs to be concealed, input the concealment criteria or rules for the data category to the LLM, and instruct the LLM to provide multiple candidate methods for concealing the data category according to the criteria or rules. This step may be performed, for example, by the candidate method determination module 1201 in FIG. 1.
[0045] In step S230, the LLM instructs the concealment method to select one of the candidate methods according to the concealment criteria or rules, which may be performed by, for example, the concealment method selection module 1202 in FIG.
[0046] According to the method shown in FIG. 2, the LLM can be used to obtain the data categories that need to be concealed and obtain the concealment methods for concealing the data values of the data categories.
[0047] 3 is a block diagram illustrating a device for processing data using an LLM according to the present disclosure. As shown in FIG. 3, the device 300 includes a concealment module 310, a processing module 320 implemented based on the LLM, and a restoration module 330. Note that these modules refer to logical functional modules and not physical entities, and the processing module 320 may be understood to include an LLM.
[0048] A user inputs data to be processed by the LLM and a processing request into the device 300. The processing request indicates how the user intends to process the data in the LLM or what processing results (e.g., analysis results of sales data) the user intends to obtain.
[0049] If the data to be processed input by the user corresponds to the above-mentioned data category that needs to be concealed, the concealment module 310 may use the concealment method or concealment program for the above-mentioned data category to conceal the data value provided by the user and obtain the concealed data value. As an example, if the data to be processed by the user relates to a product name that needs to be concealed, the concealment module 310 may use the concealment method for the product name to conceal the original product name (e.g., "XX") provided by the user and obtain the concealed product name (e.g., "YY").
[0050] The processing module 320 provides the concealed data values generated by the concealment module 310 and the user's processing request to the LLM, and instructs the LLM to perform processing based on the provided concealed data values in accordance with the user's processing request. For example, if the user wants the LLM to collect statistics on the sales amounts of various products, the LLM will perform statistics based on the received concealed product names and output the sales amount for each product, for example, 100,000, which is the sales amount of the "YY" product.
[0051] The restoration module 330 obtains the processing result output by the processing module 320, which includes the concealed data value. The restoration module 330 uses a restoration method to restore the concealed data value included in the processing result to the original data value, thereby obtaining the processing result for the original data value. As an example, the restoration module 330 restores the product name "YY" in the statistical result output by the LLM to the original product name "XX", thereby obtaining the statistical result for the true product "XX" (sales amount is 100,000).
[0052] In the present disclosure, the concealment method establishes a one-to-one mapping from original data values to concealed data values, and the restoration method is based on the inverse mapping of this mapping. As an example, the mapping relationship from original data values to concealed data values may be stored in a table, and the original data value corresponding to the concealed data value may be obtained by looking up the table during restoration. Also, in the present disclosure, the restoration method may include the inverse operation of the operation in the concealment method.
[0053] 4 is a flowchart of a process for processing data using an LLM according to the present disclosure. As shown in FIG. 4, in step S410, a user inputs a data value to be processed by the LLM. As shown in step S420, if the data value corresponds to a data category that needs to be concealed, the data value is concealed according to the concealment method for the data category, and the concealed data value is obtained. This step may be performed, for example, by the concealment module 310 in FIG. 3.
[0054] Next, in step S430, the concealed data value is input to the LLM, and the LLM is instructed to perform processing based on the provided concealed data value in response to a user processing request. This step may be performed by, for example, the processing module 320 in FIG. 3.
[0055] In step S440, the processing result of the LLM is obtained, and the concealed data value included in the processing result is restored to the original data value by a restoration method, thereby obtaining the processing result for the original data value. This step may be performed, for example, by the restoration module 330 in FIG. 3.
[0056] In particular, in the present disclosure, the concealed data values generated in step S420 may be used to further perform data augmentation. Specifically, since the concealed data values have a different but consistent data distribution from the original data values, the concealed data values may be used to augment the user's set of original data values. For example, the augmented data set may be used as a training sample or a test sample for training or testing the device according to the present disclosure or other devices or models.
[0057] In addition, in the present disclosure, the generated concealment program may be verified. The verification may include automatic verification and manual verification.
[0058] 5 is a flowchart of a process for automatically verifying a hidden program. As shown in FIG. 5, in step S510, for a data category (e.g., consumer age) corresponding to the hidden program to be verified, one sample data value (e.g., "38 years old") of the data category is generated. The generated sample data value does not need to be the user's real data because it is used only to verify the authenticity of the hidden program.
[0059] In step S520, the concealment module 310 executes the concealment program to be verified, conceals the generated sample data value, and obtains the concealed data value (for example, "33 years old").
[0060] In step S530, the restoration module 330 restores the original data value from the concealed data value according to a restoration method.
[0061] In step S540, it is determined whether the data value restored by the restoration module 330 is the same as the sample data value generated in step S510. If the determination result is "YES", it is determined that the hidden program is available.
[0062] If the determination result is "NO," it means that there is a problem with the concealment program and it is unusable. In this case, a warning is displayed indicating that there is an error and that the concealment program needs to be regenerated, and preferably specific information about the error, such as the cause of the error and the location where the error occurred, may also be provided. More specifically, in step S550, the concealment program generation module 130 in FIG. 1 instructs the LLM to regenerate the concealment program according to the concealment method for the data category (e.g., consumer age) and preferably according to the information about the error, or the concealment program may be manually modified or re-created.
[0063] Next, the method may return to step S510 and execute the verification process shown in Figure 5 for the new hidden program. Preferably, the number of times the hidden program is regenerated may be limited. For example, if all of the N (e.g., 10) regenerated hidden programs fail the verification shown in Figure 5, the process ends.
[0064] The above describes the process of automatic verification of the hidden program. For manual verification, the operation is the same as that of Fig. 5, except that the operation of generating sample data values in step S510 may be omitted, and the user's true data values may be input to the hiding module 310 in step S520.
[0065] The technology according to the present disclosure has been described above with reference to specific embodiments. The present disclosure allows users to efficiently process or analyze data using LLM while protecting data privacy, thereby promoting the use of LLM services by users.
[0066] The methods according to the above embodiments may be realized by software, hardware, or a combination of software and hardware. The programs included in the software may be pre-stored in a storage medium installed inside or outside the device. For example, during execution, these programs are written into a random access memory (RAM) and executed by a processor (e.g., a CPU) to realize each method and process described herein.
[0067] 6 is a block diagram showing an example of a computer hardware configuration capable of implementing the present disclosure. The computer hardware is an example of a device for processing data according to the present disclosure. The neural network according to the present disclosure may also be implemented based on the computer hardware.
[0068] As shown in FIG. 6, in a computer 600, a central processing unit (CPU) 601, a read-only memory (ROM) 602, and a random access memory (RAM) 603 are interconnected by a bus 604.
[0069] An input / output interface 605 is further connected to the bus 604. Connected to the input / output interface 605 are an input unit 606 including a keyboard, mouse, microphone, etc., an output unit 607 including a display, speakers, etc., a storage unit 608 including a hard disk, nonvolatile memory, etc., a communication unit 609 including a network interface card (such as a local area network (LAN) card or modem), and a driver 610 that drives a removable medium 611. The removable medium 611 is, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory.
[0070] In a computer having the above configuration, the CPU 601 loads a program stored in the storage unit 608 into the RAM 603 via the input / output interface 605 and the bus 604, and executes the program to perform the above method.
[0071] The program executed by the computer (CPU 601) may be recorded on a removable medium 611, which is a package medium. The package medium may be formed of, for example, a magnetic disk (including a floppy disk), an optical disk (including a compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), etc.), a magneto-optical disk, or a semiconductor memory. The program executed by the computer (CPU 601) may also be provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.
[0072] When the removable medium 611 is installed in the driver 610, the program can be installed in the storage unit 608 via the input / output interface 605. The program can also be received by the communication unit 609 via a wired or wireless transmission medium and installed in the storage unit 608. Alternatively, the program may be pre-installed in the ROM 602 or the storage unit 608.
[0073] The program executed by the computer may be a program that performs operations in the order described in this specification, or may be a program that performs operations in parallel, or may perform operations as needed (e.g., when called).
[0074] The devices or units described in this specification are logical and are not limited to physical devices or entities. For example, the functions of each unit described in this specification may be realized by multiple physical entities, or the functions of multiple units described in this specification may be realized by a single physical entity. Furthermore, features, components, elements, steps, etc. described in one embodiment are not limited to that embodiment and may be applied to other embodiments, for example, and may be used in place of or combined with specific features, components, elements, steps, etc. of other embodiments.
[0075] The scope of the present disclosure is not limited to the specific examples described herein. As can be understood by those skilled in the art, various modifications or variations may be made to the examples herein depending on design requirements and other factors without departing from the principles and spirit of the present disclosure. The scope of the present disclosure is limited by the appended claims and their equivalents.
[0076] Furthermore, the present disclosure further discloses the following supplementary notes regarding the embodiments including the above-described examples, but is not limited to these supplementary notes. (Appendix 1) 1. A computer-implemented method for processing data using a large-scale language model, comprising: inputting a plurality of data categories into the large-scale language model and instructing the large-scale language model to determine one or more data categories from the plurality of data categories that need to be obscured to protect data privacy, wherein each data category includes a plurality of data values; For each data category of one or more data categories that need to be obscured: inputting criteria for hiding the data categories into the large-scale language model; instructing the large-scale language model to provide a plurality of candidate methods for concealing the data category according to the criteria; the large-scale language model selecting one of the plurality of candidate methods as a concealment method according to the criterion; The criteria include at least: The data values after concealment of the data category correspond one-to-one to the original data values; and the concealed data values of the data category and the original data values are consistent in data meaning and data distribution. (Appendix 2) When receiving a data value of one of the one or more data categories that need to be concealed from a user, concealing the data value of the data category according to a concealment method for the data category; inputting the concealed data values into the large-scale language model and instructing the large-scale language model to process the concealed data values; 2. The method of claim 1, further comprising the steps of: obtaining a processing result from the large-scale language model; and restoring the concealed data values included in the processing result to their original data values according to a restoration method. (Appendix 3) the concealment method includes establishing a mapping from original data values to concealed data values; 3. The method of claim 2, wherein the restoration method is based on an inverse mapping of the mapping. (Appendix 4) the plurality of data categories are data categories related to a specific field; 2. The method of claim 1, further comprising instructing the large-scale language model, as an expert in the particular domain, to determine one or more data categories that need to be concealed from the plurality of data categories. (Appendix 5) instructing the large-scale language model to generate a concealment program according to the concealment method; or 2. The method of claim 1, further comprising a step of manually instructing the user to create a concealment program according to the concealment method. (Appendix 6) The large-scale language model further includes a step of instructing the large-scale language model to provide the plurality of candidate methods and select the concealment method according to a rule; The said rule is: The data value after the hiding of the data category and the original data value match in data type, and the data type includes numbers, alphabets, symbols, character strings, dates and times, or a combination of two or more thereof; the data values of the data categories after concealment deviate from the original data values by a fixed value; and 2. The method of claim 1, including one or more of: dividing the original data values of the data category into a plurality of groups, and in each group, there is a one-to-one correspondence between the concealed data values and the original data values. (Appendix 7) (a) executing the concealment program on sample data values to generate concealed sample data values; (b) recovering the original sample data values based on the concealed sample data values; (c) determining whether the reconstructed sample data values are identical to the original sample data values; (d) if the restored sample data values are different from the original sample data values, determining that the hiding program is unusable and regenerating the hiding program according to the hiding method; 6. The method of claim 5, wherein steps (a) to (d) are repeatedly performed. (Appendix 8) if step (c) determines that the reconstructed sample data values are identical to the original sample data values, then determining that the hiding program is available and terminating the method; 8. The method of claim 7, wherein the method is terminated when steps (a) to (d) have been repeated a predetermined number of times. (Appendix 9) 1. An apparatus for processing data in communication with a large scale language model, comprising: a memory in which a program is stored; one or more processors; The processor executes the program, inputting a plurality of data categories into the large-scale language model and instructing the large-scale language model to determine one or more data categories from the plurality of data categories that need to be obscured to protect data privacy, wherein each data category includes a plurality of data values; For each data category of one or more data categories that need to be obscured: inputting criteria for hiding the data categories into the large-scale language model; instructing the large-scale language model to provide a plurality of candidate methods for concealing the data category according to the criteria; the large-scale language model is configured to perform a method including selecting one of the plurality of candidate methods as a concealment method according to the criterion; The criteria include at least: The data values after concealment of the data category correspond one-to-one to the original data values; and The apparatus includes a data value after concealment of the data category and an original data value, the data value being consistent in data meaning and data distribution. (Appendix 10) The method comprises: When receiving a data value of one of the one or more data categories that need to be concealed from a user, concealing the data value of the data category according to a concealment method for the data category; inputting the concealed data values into the large-scale language model and instructing the large-scale language model to process the concealed data values; 10. The apparatus of claim 9, further comprising: obtaining a processing result from the large-scale language model; and restoring the concealed data values included in the processing result to their original data values according to a restoration method. (Appendix 11) the concealment method includes establishing a mapping from original data values to concealed data values; 11. The apparatus of claim 10, wherein the restoration method is based on an inverse mapping of the mapping. (Appendix 12) the plurality of data categories are data categories related to a specific field; The method comprises: 10. The apparatus of claim 9, further comprising instructing the large-scale language model, as an expert in the particular domain, to determine one or more data categories that need to be concealed from the plurality of data categories. (Appendix 13) The method comprises: instructing the large-scale language model to generate a concealment program according to the concealment method; or 10. The apparatus of claim 9, further comprising a step of manually instructing the apparatus to create a concealment program according to the concealment method. (Appendix 14) The method comprises: The large-scale language model further includes a step of instructing the large-scale language model to provide the plurality of candidate methods and select the concealment method according to a rule; The said rule is: The data value after the hiding of the data category and the original data value match in data type, and the data type includes numbers, alphabets, symbols, character strings, dates and times, or a combination of two or more thereof; the data values of the data categories after concealment deviate from the original data values by a fixed value; and 10. The apparatus of claim 9, including one or more of: original data values of the data category are divided into a plurality of groups, and in each group, there is a one-to-one correspondence between the concealed data values and the original data values. (Appendix 15) The method comprises: (a) executing the concealment program on sample data values to generate concealed sample data values; (b) recovering the original sample data values based on the concealed sample data values; (c) determining whether the reconstructed sample data values are identical to the original sample data values; (d) if the restored sample data values are different from the original sample data values, determining that the hiding program is unusable and regenerating the hiding program according to the hiding method; 14. The apparatus of claim 13, wherein steps (a) to (d) are repeatedly performed. (Appendix 16) if step (c) determines that the reconstructed sample data values are identical to the original sample data values, then determining that the hiding program is available and terminating the method; 16. The apparatus of claim 15, wherein the method is terminated when steps (a) through (d) have been repeated a predetermined number of times. (Appendix 17) A non-transitory computer-readable storage medium having stored thereon a program that, when executed by a computer, causes the computer to perform a method for processing data according to any one of claims 1 to 8.
Claims
1. 1. A computer-implemented method for processing data using a large-scale language model, comprising: inputting a plurality of data categories into the large-scale language model and instructing the large-scale language model to determine one or more data categories from the plurality of data categories that need to be obscured to protect data privacy, wherein each data category includes a plurality of data values; For each data category of the one or more data categories that need to be obscured: inputting criteria for hiding the data categories into the large-scale language model; instructing the large-scale language model to provide a plurality of candidate methods for concealing the data category according to the criteria; the large-scale language model selecting one of the plurality of candidate methods as a concealment method according to the criterion; The criteria include at least: The data values after concealment of the data category correspond one-to-one to the original data values; and the concealed data values of the data category and the original data values are consistent in data meaning and data distribution.
2. When receiving a data value of one of the one or more data categories that need to be concealed from a user, concealing the data value of the data category according to a concealment method for the data category; inputting the concealed data values into the large-scale language model and instructing the large-scale language model to process the concealed data values; The method of claim 1 , further comprising the steps of: obtaining a processing result of the large-scale language model; and restoring the concealed data values included in the processing result to their original data values according to a restoration method.
3. the concealment method includes establishing a mapping from original data values to concealed data values; The method of claim 2 , wherein the reconstruction method is based on an inverse mapping of the mapping.
4. the plurality of data categories are data categories related to a specific field; 10. The method of claim 1, further comprising instructing the large-scale language model as the domain expert to determine one or more data categories that need to be concealed from the plurality of data categories.
5. instructing the large-scale language model to generate a concealment program according to the concealment method; or The method of claim 1 , further comprising the step of manually instructing the user to create a hidden program according to the hiding method.
6. The large-scale language model further includes a step of instructing the large-scale language model to provide the plurality of candidate methods and select the concealment method according to a rule; The said rule is: The data value after the hiding of the data category and the original data value match in data type, and the data type includes numbers, alphabets, symbols, character strings, dates and times, or a combination of two or more thereof; the data values of the data categories after concealment deviate from the original data values by a fixed value; and 2. The method of claim 1, comprising one or more of: dividing the original data values of the data category into a plurality of groups, and in each group, there is a one-to-one correspondence between the concealed data values and the original data values.
7. (a) executing the concealment program on sample data values to generate concealed sample data values; (b) recovering the original sample data values based on the concealed sample data values; (c) determining whether the reconstructed sample data values are identical to the original sample data values; (d) if the restored sample data values are different from the original sample data values, determining that the hiding program is unusable and regenerating the hiding program according to the hiding method; The method of claim 5 , wherein steps (a) through (d) are repeatedly performed.
8. if step (c) determines that the reconstructed sample data values are identical to the original sample data values, then determining that the hidden program is available and terminating the method; The method of claim 7 , wherein the method is terminated when steps (a) through (d) have been repeated a predetermined number of times.
9. 1. An apparatus for processing data in communication with a large scale language model, comprising: a memory in which a program is stored; one or more processors; The processor executes the program, inputting a plurality of data categories into the large-scale language model and instructing the large-scale language model to determine one or more data categories from the plurality of data categories that need to be obscured to protect data privacy, wherein each data category includes a plurality of data values; For each data category of the one or more data categories that need to be obscured: inputting criteria for hiding the data categories into the large-scale language model; instructing the large-scale language model to provide a plurality of candidate methods for concealing the data category according to the criteria; the large-scale language model is configured to perform a method including selecting one of the plurality of candidate methods as a concealment method according to the criterion; The criteria include at least: The data values after concealment of the data category correspond one-to-one to the original data values; and The apparatus includes a data value after concealment of the data category and an original data value, the data value being consistent in data meaning and data distribution.
10. A non-transitory computer-readable storage medium having a program stored thereon, the program causing the computer to execute the method for processing data according to any one of claims 1 to 8 when the program is executed by the computer.