Data acquisition method, program, and data acquisition device

Abstract

This invention provides a data acquisition method, program, and data acquisition apparatus that enable the acquisition of data in which masking of confidential information has been performed with higher accuracy when a large-scale language model is used to perform masking of confidential information. [Solution] The data acquisition method includes the steps of: acquiring first data containing confidential information; acquiring second data in which all or part of the confidential information has been masked by transmitting all or part of the first data to a large-scale language model; receiving an instruction to change the target of the masking process in the second data; and acquiring third data in which the target of the masking process has been changed from the second data according to the change instruction.

Description

Technical Field

[0001] The present invention relates to a data acquisition method, a program, and a data acquisition device.

Background Art

[0002] Japanese Patent No. 6083101 (Patent Document 1) discloses an information processing apparatus. In this information processing apparatus, secondary data with sufficient anonymity is generated based on primary data in response to the input of primary data including personal information.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] An object of the present invention is to provide a data acquisition method, a program, and a data acquisition device capable of acquiring data in which masking processing for confidential information that requires masking is performed with higher accuracy when performing masking processing on confidential information for a large language model.

Means for Solving the Problems

[0005] A data acquisition method according to an aspect of the present invention includes: a step of acquiring first data including confidential information; a step of acquiring second data in which all or part of the confidential information is masked by transmitting all or part of the first data to a large language model; a step of receiving an instruction to change the target of the masking processing in the second data; and a step of acquiring third data in which the target of the masking processing is changed from the second data according to the change instruction.

[0006] In large-scale language models, masking is not always applied to sensitive information that requires masking. In this data acquisition method, a change instruction is received to modify the target of masking in the second data, and a third data is acquired in which the target of masking has been changed from the second data according to the change instruction. Therefore, with this data acquisition method, by giving appropriate change instructions, it is possible to acquire a third data in which sensitive information that requires masking has been masked with higher accuracy.

[0007] The above data acquisition method may further include the step of displaying a review screen that includes all or part of the content shown in the second data, and the above change instructions may be received through the review screen.

[0008] In this data acquisition method, instructions to change the target of the masking process in the second data are received through the review screen. Therefore, this data acquisition method allows users to easily give change instructions while the content shown in the second data is visible to the user.

[0009] In the data acquisition method described above, the targets of the masking process in the second data may be listed as selection items on the review screen, and the above change instruction may be performed by changing the selection status of the selection items.

[0010] In this data acquisition method, instructions to change the target of the masking process in the second data are made by changing the selection state of the selected item. Therefore, with this data acquisition method, the user can easily make change instructions simply by changing the selection state of the selected item.

[0011] In the data acquisition method described above, the instruction to make changes may be made by selecting confidential information included in the content described above.

[0012] In this data acquisition method, the instruction to change the target of the masking process in the second data is made by selecting the confidential information contained in the content shown in the second data. Therefore, with this data acquisition method, the user can easily make a change instruction simply by selecting the confidential information contained in the content shown in the second data.

[0013] In the above data acquisition method, the third data may be acquired by sending target data indicating the target of the masking process newly selected through the above change instruction to the large-scale language model.

[0014] According to this data acquisition method, the third data is generated in a large-scale language model based on the target data, making it possible to obtain third data with more accurate masking applied to sensitive information that requires masking.

[0015] The above data acquisition method may further include a step of receiving a switching instruction to include all or part of the content shown in the second data, or all or part of the content shown in the first data, in the review screen.

[0016] In this data acquisition method, content before and after masking is selectively included in the review screen. Therefore, this data acquisition method allows users to efficiently review the masking results while viewing the content before and after masking.

[0017] In the data acquisition method described above, a first prompt and a second prompt may be sent to the large-scale language model in stages. The first prompt may include an instruction to extract candidates for masking from all or part of the first data, and the second prompt may include an instruction to determine which candidates for masking are to be masked from the extracted candidates, taking into account the content of all or part of the content represented by the first data.

[0018] In this data acquisition method, masking is performed by sending prompts to a large-scale language model in two stages. Therefore, this data acquisition method allows for the acquisition of a third set of data in which sensitive information requiring masking has been masked with higher accuracy compared to a case where masking is performed by sending prompts to a large-scale language model in one stage.

[0019] The above data acquisition method may further include a step of receiving input of target information indicating the subject of the masking process before transmitting all or part of the first data to the large-scale language model, and the second data may be acquired by transmitting all or part of the first data and the target information to the large-scale language model.

[0020] In this data acquisition method, target information indicating the items to be masked is accepted in advance, and the masking process is performed based on the pre-entered target information. Therefore, with this data acquisition method, since the masking process is performed in accordance with the user's intentions, it is possible to acquire third-party data in which confidential information requiring masking has been masked with higher accuracy.

[0021] A program according to another aspect of the present invention causes a computer to perform the following processes: acquiring first data containing confidential information; acquiring second data in which all or part of the confidential information has been masked by transmitting all or part of the first data to a large-scale language model; receiving instructions to change the target of the masking process in the second data; and acquiring third data in which the target of the masking process has been changed from the second data according to the change instructions.

[0022] When this program is executed, an instruction to change the target of the masking process in the second data is received, and third data in which the target of the masking process is changed from the second data in accordance with the change instruction is acquired. Therefore, according to this program, by giving an appropriate change instruction, it is possible to acquire third data in which the masking process for confidential information that requires masking is performed with higher precision.

[0023] A data acquisition device according to another aspect of the present invention includes a memory that stores a program and a processor that executes the program. The program causes the processor to execute a process of acquiring first data including confidential information, a process of acquiring second data in which all or part of the first data is masked by transmitting all or part of the first data to a large language model, a process of receiving an instruction to change the target of the masking process in the second data, and a process of acquiring third data in which the target of the masking process is changed from the second data in accordance with the change instruction.

[0024] In this data acquisition device, an instruction to change the target of the masking process in the second data is received, and third data in which the target of the masking process is changed from the second data in accordance with the change instruction is acquired. Therefore, according to this data acquisition device, by giving an appropriate change instruction, it is possible to acquire third data in which the masking process for confidential information that requires masking is performed with higher precision.

Advantages of the Invention

[0025] According to the present invention, there can be provided a data acquisition method, a program, and a data acquisition device capable of acquiring data in which the masking process for confidential information that requires masking is performed with higher precision when causing a large language model to perform the masking process on the confidential information.

Brief Description of the Drawings

[0026] [Figure 1] It is a diagram schematically showing the configuration of a WF management system. [Figure 2] This is a block diagram schematically showing the server's hardware configuration. [Figure 3] This is a block diagram schematically showing the hardware configuration of a user terminal. [Figure 4] This diagram schematically shows an example of a wireframe creation screen. [Figure 5] This is the first flowchart illustrating some of the processes defined in the masking component. [Figure 6] This diagram schematically shows an example of a review screen. [Figure 7] This flowchart shows the procedure for switching between displaying or not displaying masks on content shown on the review screen. [Figure 8] This diagram schematically shows an example of a review screen that displays the content of the target file before masking. [Figure 9] This flowchart shows the procedure for changing the target of the masking process in the target file. [Figure 10] This diagram illustrates an example of how to add objects to be masked. [Figure 11] This is a second flowchart illustrating some of the processes defined in the masking component. [Figure 12] This diagram schematically shows an example of a wireframe creation screen where the target of the masking process can be specified. [Modes for carrying out the invention]

[0027] Hereinafter, an embodiment relating to one aspect of the present invention (hereinafter also referred to as "this embodiment") will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals, and their descriptions will not be repeated. Furthermore, each drawing is schematically depicted with parts omitted or exaggerated as appropriate for ease of understanding.

[0028] [1. Overview] Figure 1 is a schematic diagram showing the configuration of a Workflow (WF) management system 10 that uses a data acquisition method according to this embodiment. As shown in Figure 1, the WF management system 10 includes a server 100, a plurality of user terminals 200, and an AI server 300. In the WF management system 10, the server 100, the plurality of user terminals 200, and the AI ​​server 300 are configured to communicate with each other via the Internet N1.

[0029] Server 100 provides a screen for creating workflows (hereinafter also referred to as the "WF creation screen"). Users can create workflows through a user terminal 200 with the WF creation screen displayed. As will be described in detail later, the WF management system 10 has several types of components prepared in advance, and various workflows can be created by combining multiple components. Each component defines, for example, the processing to be performed on the input file (hereinafter also referred to as the "input file"). In a workflow composed of multiple types of components, various processes are applied sequentially to the input file, and files after various processing (hereinafter also referred to as "output files") are generated. Workflows created by users are managed by Server 100.

[0030] For example, among multiple users belonging to the same company, workflows created by each user are shared. By sharing high-quality workflows among multiple users, the quality of output files is stabilized, and the quality of work of each user is standardized to a high level.

[0031] In a company's sales department, for example, sales materials used by top salespeople may be shared with other salespeople in order to share the know-how of top salespeople. On the other hand, in a software vendor, for example, two competing companies may both be clients. To prevent confidential information from leaking to the other, the sales department may be divided into multiple teams, with one team handling sales to one company and the other handling the other. In such a case, sales materials used by salespeople in one team may contain confidential information that should not be known to salespeople in the other team. If these sales materials are shared within the company as is, the confidential information will be revealed to salespeople who should not have access to it.

[0032] In the WF management system 10, a component for masking confidential information (hereinafter also referred to as the "masking component") is provided as a component used to create workflows. For example, when the processing defined in the masking component is applied to a sales material file, confidential information is removed from the sales material. By sharing the sales material after masking, it is possible to achieve both the sharing of know-how and the prevention of leakage of confidential information.

[0033] The AI ​​server 300 provides a generative AI model 310. The generative AI model 310 is an example of a large-scale language model and is generated, for example, through learning using information periodically acquired from various sites on the Internet N1. The generative AI model 310 is configured to accept natural language input and generate various responses corresponding to that natural language based on common sense. Examples of generative AI models 310 include GPT-4o and Gemini 2.0.

[0034] As will be explained in more detail later, the masking process in the masking component is mainly performed in the generating AI model 310. When the masking process is performed in the generating AI model 310, if no special measures are taken, it is possible that the masking process may not be sufficiently performed on confidential information that needs to be masked. In other words, it is possible that confidential information that needs to be masked may not be masked, while information that does not need to be masked may be masked. The WF management system 10 has been designed to suppress the occurrence of such a situation. The details of the WF management system 10 will be explained below.

[0035] [2. Structure] <2-1. Server Configuration> Figure 2 is a schematic block diagram showing the hardware configuration of server 100. Server 100 is implemented, for example, by a general-purpose computer. As shown in Figure 2, server 100 includes a control unit 110, a communication interface 130, and a storage unit 120. Each component is electrically connected via a bus.

[0036] The control unit 110 includes a CPU (Central Processing Unit) 112, RAM (Random Access Memory) 114, and ROM (Read Only Memory) 116, and is configured to control each component according to information processing.

[0037] The communication interface 130 is configured to communicate with the user terminal 200 and the AI ​​server 300 (Figure 1) via the internet N1. The communication interface 130 consists of, for example, a wired LAN (Local Area Network) module or a wireless LAN module.

[0038] The storage unit 120 is composed of, for example, an auxiliary storage device such as a hard disk drive or a solid-state drive. The storage unit 120 stores, for example, a control program 122, a component database 123, a workflow database 124, a shared file database 125, and a group of prompt template data 126. Various functions of the server 100 are realized when the control program 122 is executed by the CPU 112 (processor).

[0039] The component DB123 manages multiple types of components, each used to create workflows. The workflow DB124 manages one or more workflows created by users. The shared file DB125 manages one or more masked files that can be shared among multiple users, for example. The prompt template data group 126 contains multiple prompt template data. Each of the multiple prompt template data consists of instructions to the generating AI model 310. The multiple prompt template data includes template data for the first prompt, second prompt, and third prompt, as described below.

[0040] <2-2. User Terminal Configuration> Figure 3 is a schematic block diagram showing the hardware configuration of the user terminal 200. The user terminal 200 can be implemented, for example, by a PC (Personal Computer), a tablet, or a smartphone. As shown in Figure 3, the user terminal 200 includes a control unit 210, a communication interface 230, an operation unit 240, a display 250, and a storage unit 220. In the user terminal 200, each component is electrically connected via a bus.

[0041] The control unit 210 includes a CPU, RAM, ROM, etc., and is configured to control each component according to information processing. The communication interface 230 is configured to communicate with the server 100 and the AI ​​server 300 via the internet N1. The communication interface 230 is composed of, for example, a wired LAN module or a wireless LAN module.

[0042] The control unit 240 is configured to receive input from the user. The control unit 240 consists of, for example, some or all of a touch panel, keyboard, mouse, and microphone. Input on the wireframe creation screen, etc., is performed, for example, through the control unit 240.

[0043] The display 250 is configured to display various screens. The display 250 consists of a monitor such as an LCD monitor or an OLED (Electro-Luminescence) monitor. The display 250 displays a screen such as a WF creation screen provided by the server 100.

[0044] The storage unit 220 is, for example, an auxiliary storage device such as a hard disk drive or a solid-state drive. The storage unit 220 stores, for example, a control program 222. Various functions in the user terminal 200 are realized when the control program 222 is executed by the CPU (processor) of the control unit 210.

[0045] [3. Workflow creation function] Figure 4 is a schematic diagram showing an example of a WF creation screen. Referring to Figure 4, the user terminal 200 has, for example, an application (hereinafter also referred to as the "WF management app") installed for using various functions provided by the WF management system 10. When the WF management app is launched and a command to create a new workflow is received from the user, for example, the WF creation screen IM1 is displayed on the display 250 of the user terminal 200.

[0046] On the WF creation screen IM1, for example, button B1 is displayed. When button B1 is pressed by the user, for example, multiple types of component CP1 are listed in a selectable state. From the listed multiple types of component CP1, the icon representing the component CP1 selected by the user is displayed in area TR1. The user can change the position of the icon in area TR1 by, for example, dragging and dropping the icon. The user can also change the connection order of the multiple icons displayed in area TR1 as appropriate. When the process shown in the workflow is started, for example, the processing defined in the component CP1 corresponding to each icon is executed in the order in which the icons are connected.

[0047] Furthermore, the WF creation screen IM1 displays, for example, a text box TB1 and a button B2. The text box TB1 is used to enter, for example, the title of the workflow created in area TR1. When button B2 is pressed by the user, the workflow created in area TR1 is saved. The saved workflow is stored, for example, in workflow DB124 (see Figure 2).

[0048] When the workflow shown in Figure 4 is executed, for example, the following processes are executed in this order: (1) accepting file input, (2) summarizing the contents of the input file, (3) applying masking to the summarized input file and the input file itself, and (4) outputting the masked summary and the masked input file itself. In this way, the WF management system 10 creates workflows through the WF creation screen IM1.

[0049] [4. Operation] Figure 5 is a first flowchart showing some of the processing defined in the masking component. The processing shown in this flowchart is executed, for example, by the control unit 110 of the server 100 when the processing of the component executed immediately before the masking component is completed during the execution of a workflow that includes the masking component (hereinafter also referred to as the "predetermined workflow").

[0050] Referring to Figure 5, the control unit 110 of the server 100 determines whether or not it has received a file to be masked (hereinafter also referred to as the "target file") from the user terminal 200 (step S100). The target file contains, for example, confidential information. For example, a predetermined workflow includes a component that accepts input of a target file, and when the processing defined in this component is executed, the user is requested to select a target file. Once the user selects a target file, the selected target file is sent to the server 100.

[0051] If it is determined that the target file has not been received from the user terminal 200 (NO in step S100), the control unit 110 waits until the target file is received. On the other hand, if it is determined that the target file has been received from the user terminal 200 (YES in step S100), the control unit 110 controls the communication interface 130 to send all or part of the target file (hereinafter also referred to as "target data") and the first prompt to the generating AI model 310 (AI server 300) (step S110).

[0052] The first prompt is an instruction to the generating AI model 310, which includes, for example, an instruction to extract candidates for the masking process from the target data. The first prompt may also include an instruction to extract candidates for the pre-specified target from the target data if the target for the masking process is pre-specified. For example, if "company name" is pre-specified as the target for the masking process, the first prompt may include an instruction to extract candidates for the string corresponding to "company name" from the target data. The first prompt also includes an instruction to generate a list of candidates for the masking process (hereinafter also referred to as the "candidate list"). The generated candidate list is sent directly or indirectly from the AI ​​server 300 to the server 100. In the first stage, the extraction of candidates for the masking process is performed in the generating AI model 310.

[0053] The control unit 110 determines whether or not it has received a candidate list from the generating AI model 310 (step S120). If it is determined that the candidate list has not been received from the generating AI model 310 (NO in step S120), the control unit 110 waits until it receives the candidate list.

[0054] On the other hand, when it is determined that a candidate list has been received from the generating AI model 310 (YES in step S120), the control unit 110 controls the communication interface 130 to send the candidate list and the second prompt to the generating AI model 310 (step S130).

[0055] The second prompt is an instruction to the generating AI model 310, which includes, for example, an instruction to determine the target of masking from the candidate list, taking into account the content of all or part of the content represented by the target data. The second prompt may also include an instruction to determine the target of masking from the candidate list, taking into account the context of all or part of the content represented by the target data. Furthermore, for example, if the target file is presentation material including figures and tables, in order to take into account the context of all or part of the content represented by the target data, an image file representing the content of the target file may be sent to the generating AI model 310 in step S130, in addition to the candidate list and the second prompt.

[0056] The control unit 110 determines whether or not it has acquired the first processed data (step S140). The first processed data is a target file that has been masked. In the first processed data, the target of the masking process determined in the generating AI model 310 (for example, a string) is masked. The first processed data may be generated in the generating AI model 310 and received from the generating AI model 310, or it may be generated in the server 100 based on a list of targets for masking (hereinafter also referred to as the "target list") received from the generating AI model 310. If the first processed data is generated in the generating AI model 310, the target list may be sent from the generating AI model 310 to the server 100 along with the first processed data. That is, the second prompt may include an instruction to generate the target list and send the target list to the server 100, or it may include an instruction to generate the target list and the first processed data and send the target list and the first processed data to the server 100.

[0057] If it is determined that the first processed data has not been acquired (NO in step S140), the control unit 110 waits until the first processed data is acquired. On the other hand, if it is determined that the first processed data has been acquired (YES in step S140), the control unit 110 controls the communication interface 130 to send image data showing the review screen to the user terminal 200 (step S150). As a result, the review screen is displayed on the display 250 of the user terminal 200. The review screen is a screen for the user to review the results of the masking process on the target file.

[0058] Thus, in the WF management system 10, masking is performed by sending prompts to the generating AI model 310 in two stages. In large-scale language models, the quality of the output tends to improve by reviewing its own output. Therefore, with this WF management system 10, compared to the case where masking is performed by sending prompts to the generating AI model 310 in one stage, it is possible to obtain first processed data in which the masking of confidential information requiring masking has been performed with higher accuracy.

[0059] Figure 6 is a schematic diagram illustrating an example of a review screen. Referring to Figure 6, the user terminal 200's display 250 shows, for example, the review screen IM2. In the review screen IM2, the masked content of the target file is displayed in area TR2. In the figure, [C1] and [ADR1] represent the parts of the target file that have been masked. The file name of the target file is displayed in area TR3.

[0060] The review screen IM2 also displays a slider bar SL1, multiple checkboxes CB1, and buttons B3 and B4. The slider bar SL1 can be slid left and right. Moving the position of the slider bar SL1 switches the content displayed in area TR2. Specifically, when the slider bar SL1 is on the right, the content of the target file that has been masked is displayed in area TR2, and when the slider bar SL1 is on the left, the content of the target file before masking is displayed in area TR2.

[0061] Thus, in the WF management system 10, the content before and after masking is selectively displayed on the review screen IM2. Therefore, with the WF management system 10, users can efficiently review the masking results while checking the content before and after masking.

[0062] Figure 7 is a flowchart showing the procedure for switching between having a mask on or off in the content displayed on the review screen IM2. The process shown in this flowchart is repeatedly executed at predetermined intervals by the control unit 210 of the user terminal 200 while the review screen IM2 is being displayed.

[0063] Referring to Figure 7, the control unit 210 of the user terminal 200 determines whether or not it has received an instruction to change the presence or absence of a mask in the content displayed on the review screen IM2 (step S200). More specifically, the control unit 210 determines whether or not it has received a slide operation of the slider bar SL1 from the user. If it is determined that a slide operation of the slider bar SL1 has not been received (NO in step S200), the control unit 210 waits until it receives a slide operation of the slider bar SL1.

[0064] On the other hand, if it is determined that a slide operation of the slide bar SL1 has been accepted (YES in step S200), the control unit 210 controls the display 250 to change whether or not a mask is present in the content displayed on the review screen IM2 (step S210).

[0065] Figure 8 schematically shows an example of the review screen IM2 that displays the content of the target file before masking. Referring to Figure 8, in this example, the slider bar SL1 is located on the left side, and the content of the target file before masking is displayed in area TR2. The part that was "[C1] Corporation" in Figure 6 is now "ABCD Assembly Corporation", and the part that was "[ADR1]" in Figure 6 is now "Roppongi, Minato-ku, Tokyo...".

[0066] Referring again to Figure 6, each of the multiple checkboxes CB1 is associated with a target to be masked. In this example, two checkboxes CB1 are displayed on the review screen IM2. The upper checkbox CB1 is associated with "ABCD Assembly Co., Ltd.", and the lower checkbox CB1 is associated with "Roppongi, Minato-ku, Tokyo...". The association between checkboxes CB1 and targets to be masked is performed, for example, based on a target list received from the generating AI model 310. On the review screen IM2, for example, as many checkboxes CB1 as there are targets included in the target list are displayed.

[0067] When checkbox CB1 is checked, it indicates that the object associated with checkbox CB1 will be subject to the masking process. In this example, for instance, when checkbox CB1 is unchecked, the object associated with checkbox CB1 will be removed from the masking process.

[0068] Figure 9 is a flowchart showing the procedure for changing the target of the masking process in the target file. The process shown in this flowchart is repeatedly executed at predetermined intervals by the control unit 210 of the user terminal 200 while the review screen IM2 is displayed.

[0069] Referring to Figure 9, the control unit 210 of the user terminal 200 determines whether or not it has received an instruction regarding the addition or reduction of the targets for masking in the target file (step S300). The reduction of targets for masking is performed, for example, by unchecking checkbox CB1.

[0070] Figure 10 illustrates an example of how to add items to be masked. Referring to the upper part of Figure 10, the user selects an item to be added to the masking process from the content displayed in area TR2, for example. In this example, the string "XYZ Investment Company" is selected. With this selected, the user then performs an operation to display area TR4, for example. In area TR4, button B5 is displayed, for example. When the user presses button B5, "XYZ Investment Company" is added to the masking process.

[0071] Referring to the lower diagram in Figure 10, one checkbox CB1 has been added to the review screen IM2, and three checkboxes CB1 are now displayed. The bottommost checkbox CB1 is associated with "XYZ Investment Company". For example, items to be masked are added using this procedure.

[0072] Referring again to Figure 9, if it is determined in step S300 that no instruction has been received regarding the addition or reduction of targets for masking in the target file (NO in step S300), the control unit 210 waits until an instruction is received regarding the addition or reduction of targets for masking in the target file. On the other hand, if it is determined in step S300 that an instruction has been received regarding the addition or reduction of targets for masking in the target file (YES in step S300), the control unit 210 performs processing according to the received instruction (step S310). That is, if an instruction to reduce the targets for masking is received, the control unit 210 performs the process of unchecking the checkbox CB1 associated with the targets to be removed. Also, if an instruction to add targets for masking is received, the control unit 210 performs the process of adding the checkbox CB1 associated with the added targets.

[0073] Thus, in the WF management system 10, instructions to change the target of the masking process in the target file are received through the review screen IM2. Therefore, with the WF management system 10, the user can easily give change instructions while the content shown in the masked target file (first processed data) is visible to the user.

[0074] Furthermore, in the WF management system 10, instructions to reduce the amount of data subject to masking in the first processed data are given by changing the checked state of checkbox CB1. Therefore, with the WF management system 10, users can easily give reduction instructions simply by changing the checked state of checkbox CB1.

[0075] Furthermore, in the WF management system 10, additional instructions for masking the first processed data are made by selecting the confidential information contained in the content represented by the first processed data. Therefore, with the WF management system 10, users can easily make additional instructions simply by selecting the confidential information contained in the content represented by the first processed data.

[0076] Referring again to Figure 6, button B3 is pressed by the user, for example, when it is determined that there are no problems with the masking state of the content displayed in area TR2. In the WF management system 10, for example, the approval flow for each workflow is predetermined. When button B3 is pressed by the user, an approval request is sent to a predetermined approver, and when approved by the approver, an approval notification is received and the approval flow is completed. Button B4 is pressed, for example, when it is determined that there is a problem with the masking state of the content displayed in area TR2, after the instruction to change the target of the masking process has been completed. When button B4 is pressed by the user, the masking process for the target file is performed again.

[0077] Figure 11 is a second flowchart showing part of the processing defined in the masking component. The processing shown in this flowchart is repeatedly executed at predetermined intervals by the control unit 110 of the server 100 when the review screen IM2 is displayed on the user terminal 200.

[0078] Referring to Figure 11, the control unit 110 of the server 100 determines whether it has received a signal indicating an instruction to re-execute the masking process and the latest target list from the user terminal 200 (step S400). When button B4 is pressed on the user terminal 200 (see Figure 6), a signal indicating an instruction to re-execute the masking process and the latest target list are sent from the user terminal 200 to the server 100. The latest target list includes each target associated with the checkbox CB1 that was checked on the review screen IM2.

[0079] When it is determined that a signal indicating an instruction to re-execute the masking process and the latest target list have been received from the user terminal 200 (YES in step S400), the control unit 110 controls the communication I / F 130 to send the latest target list and a third prompt to the generating AI model 310 (step S410). The third prompt is an instruction to the generating AI model 310, which includes, for example, an instruction to determine the target of the masking process from the latest target list, taking into account the content of all or part of the content indicated by the target data.

[0080] The control unit 110 determines whether or not it has acquired the second processed data (step S420). The second processed data is a masked target file in which the target of the masking process has been changed from the first processed data. In the second processed data, the target of the masking process (e.g., a string) determined in the generating AI model 310 is masked. The second processed data may be generated in the generating AI model 310 and received from the generating AI model 310, or it may be generated in the server 100 based on the updated latest target list received from the generating AI model 310. If the second processed data is generated in the generating AI model 310, the updated latest target list may be sent from the generating AI model 310 to the server 100 along with the second processed data. That is, the third prompt may include an instruction to generate the updated latest target list and send the updated latest target list to the server 100, or it may include an instruction to generate the updated latest target list and the second processed data and send the updated latest target list and the second processed data to the server 100.

[0081] If it is determined that the second processed data has not been acquired (NO in step S420), the control unit 110 waits until the second processed data is acquired. On the other hand, if it is determined that the second processed data has been acquired (YES in step S420), the control unit 110 controls the communication interface 130 to send image data showing the review screen to the user terminal 200 (step S430). As a result, the review screen is displayed on the display 250 of the user terminal 200.

[0082] In step S400, if it is determined that a signal indicating an instruction to re-execute the masking process and the latest target list have not been received from the user terminal 200 (NO in step S400), the control unit 110 determines whether or not it has received an approval request for the masking process (step S440). For example, when button B3 (see Figure 6) is pressed on the user terminal 200, an approval request signal is sent from the user terminal 200 to the server 100. Specifically, in step S440, it is determined whether or not an approval request signal has been received from the user terminal 200.

[0083] If it is determined that the request for approval of the masking process has not been received (NO in step S440), the control unit 110 executes the process in step S400 again. On the other hand, if it is determined that the request for approval of the masking process has been received (YES in step S440), the control unit 110 executes the process for the approval request (step S450) and terminates the process shown in this flowchart. Once the contents of the masking process are approved by the approver, for example, the target file after the masking process is stored in the shared file DB125 and shared among multiple users.

[0084] [5. Features] As described above, the WF management system 10 receives an instruction to change the target of the masking process in the first processed data, and obtains second processed data in which the target of the masking process has been changed from the first processed data according to the change instruction. Therefore, with the WF management system 10, by giving an appropriate change instruction, it is possible to obtain second processed data in which the masking process for confidential information requiring masking has been performed with higher accuracy.

[0085] [6. Other Embodiments] The concept of the above embodiments is not limited to those described above. Examples of other embodiments to which the concept of the above embodiments can be applied will be described below.

[0086] <6-1> In the above embodiment, the confidential information subject to masking is not necessarily limited to strings. The confidential information subject to masking may be, for example, a design such as a company logo. In this case, for example, the first prompt may include an instruction to extract candidate designs corresponding to "company logos" from the target data.

[0087] <6-2> In the above embodiment, the masking process does not necessarily have to be provided in the form of a masking component. The masking process may be included as a process defined in another component. In this case, the processes shown in the flowcharts of Figures 5, 7, 9, and 11 may be defined in the other component.

[0088] <6-3> In the above embodiment, the target of the masking process may be specified by the user during the workflow creation stage. During the workflow creation stage, for example, the user may specify a specific target (e.g., "company name", "address", "interest rate", "contract amount") as the target of the masking process.

[0089] Figure 12 schematically shows an example of a WF creation screen in which the target of masking can be specified. Referring to Figure 12, the WF creation screen IM1A is displayed on the display 250 of the user terminal 200. In the WF creation screen IM1A, component CP1 corresponding to the masking process is selected. In the WF creation screen IM1A, area TR5 is displayed for making various settings related to the selected component CP1. Area TR5 contains multiple text boxes TB2. For example, target information indicating the target of the masking process can be entered into each text box TB2. For example, the target information entered into each text box TB2 may be reflected in at least one of the first prompt and the second prompt, and the target information may be sent to the generating AI model 310. As a result, first processed data may be generated.

[0090] In this WF management system 10, target information indicating the items to be masked is accepted in advance, and the masking process is performed based on the pre-entered target information. Therefore, with this WF management system 10, since the masking process is performed in accordance with the user's intentions, it is possible to obtain second-processed data in which the masking process for confidential information requiring masking has been performed with higher accuracy.

[0091] Embodiments of the present invention have been described illustratively above. That is, a detailed description and accompanying drawings have been disclosed for illustrative purposes. Therefore, some of the components described in the detailed description and accompanying drawings may not be essential for solving the problem. Consequently, the mere fact that these non-essential components are described in the detailed description and accompanying drawings does not mean that they should be immediately assumed to be essential.

[0092] Furthermore, the above embodiments are merely illustrative in every respect of the present invention. The above embodiments can be improved or modified in various ways within the scope of the present invention. For example, at least a part of the configuration of one embodiment may be combined with at least a part of the configuration of any other embodiment. In other words, in carrying out the present invention, specific configurations can be appropriately adopted depending on the embodiment. [Explanation of Symbols]

[0093] 10 WF management system, 100 server, 110,210 control unit, 112 CPU, 114 RAM, 116 ROM, 120,220 storage unit, 122,222 control program, 123 component DB, 124 workflow DB, 125 shared file DB, 126 prompt template data set, 130,230 communication I / F, 200 user terminal, 240 operation unit, 250 display, 300 AI server, 310 generated AI model, B1-B5 buttons, CB1 checkbox, CP1 component, IM1,IM1A WF creation screen, IM2 review screen, SL1 slider bar, TB1,TB2 text box, TR1-TR5 area.

Claims

1. The steps include obtaining first data containing confidential information, The steps include: obtaining second data in which all or part of the confidential information has been masked by transmitting all or part of the first data to a large-scale language model; A step of receiving an instruction to change the target of the masking process in the second data, A data acquisition method comprising the step of acquiring third data in which the target of the masking process has been changed from the second data in accordance with the change instruction.

2. The step further includes displaying a review screen that includes all or part of the content shown in the second data, The data acquisition method according to claim 1, wherein the change instruction is received through the review screen.

3. The targets of the masking process in the second data are listed as selection items on the review screen. The data acquisition method according to claim 2, wherein the change instruction is performed by changing the selection state of the selected item.

4. The data acquisition method according to claim 2 or 3, wherein the change instruction is performed by selecting the confidential information included in the content.

5. The data acquisition method according to any one of claims 1 to 3, wherein the third data is acquired by transmitting target data indicating the target of the masking process newly selected through the change instruction to the large-scale language model.

6. The data acquisition method according to claim 2 or 3, further comprising the step of receiving a switching instruction to include in the review screen all or part of the content indicated by the second data, and all or part of the content indicated by the first data.

7. The first and second prompts are sent sequentially to the aforementioned large-scale language model. The first prompt includes an instruction to extract candidates for the masking process from all or part of the first data, The data acquisition method according to any one of claims 1 to 3, wherein the second prompt includes an instruction to determine the target of the masking process from the extracted candidates, based on the content of all or part of the content indicated by the first data.

8. The process further includes receiving input of target information indicating the target of the masking process before transmitting all or part of the first data to the large-scale language model, The data acquisition method according to any one of claims 1 to 3, wherein the second data is acquired by transmitting all or part of the first data and the target information to the large-scale language model.

9. The process of obtaining the first data containing confidential information, A process of obtaining second data in which all or part of the confidential information has been masked by transmitting all or part of the first data to a large-scale language model, A process for receiving an instruction to change the target of the masking process in the second data, A program that causes a computer to perform the following: a process of obtaining a third data which is the target of the masking process that has been changed from the second data in accordance with the aforementioned change instruction.

10. Memory for storing programs, The system comprises a processor that executes the aforementioned program, The aforementioned program, The process of obtaining the first data containing confidential information, A process of obtaining second data in which all or part of the confidential information has been masked by transmitting all or part of the first data to a large-scale language model, A process for receiving an instruction to change the target of the masking process in the second data, A data acquisition device that causes the processor to perform the following: a process of acquiring a third data in which the target of the masking process has been changed from the second data in accordance with the change instruction.

Images