Masking system and masking program

Abstract

This invention provides a masking system and program that can mask a wide variety of information and allow for easy verification of the masking results. [Solution] The masking system 100 includes an input unit 210 that receives input of a document to be masked, a mask creation unit 220 that identifies one or more mask locations and one or more masking categories from the document and creates one or more sets that associate each of the one or more mask locations with each of the one or more masking categories, and an output unit 230 that outputs the masking results including one or more sets to a mask confirmation screen 240.

Description

Technical Field

[0001] This disclosure relates to masking technology.

Background Art

[0002] Documents such as reports, minutes of meetings, and contracts may contain information that should be concealed from third parties, such as personal information and amounts. In such cases, masking may be performed on the documents. Masking is an act or process of completely removing the desired information on the document, such as by blotting it out. Generally, since masking is performed manually, it requires a huge amount of working time. Also, when masking is performed manually, human errors such as overlooking the masking target may occur. Therefore, there is a need for a technology to reduce the burden of the masking work of documents.

[0003] Regarding masking technology, for example, Japanese Unexamined Patent Application Publication No. 2024-96560 (Patent Document 1) discloses a computer system. According to Patent Document 1, the computer system acquires, as a character string determination unit, related character strings associated with personal information from a personal information related character string DB. The computer system determines, as a character string determination unit, whether or not the related character string is included in the target character string to be determined. The computer system masks, as a masking processing unit, the character strings in the vicinity of the related character string when it is determined by the character string determination unit that the related character string is included in the target character string ([see the summary]).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] According to the technology disclosed in Patent Document 1, only pre-registered characters can be masked. However, the information to be masked can include a wide variety of information such as characters, tabular information, and images. Furthermore, the technology disclosed in Patent Document 1 does not provide a means to confirm whether or not the masking has been performed properly. Therefore, there is a need for a technology that can mask a wide variety of information and easily confirm the masking results.

[0006] This disclosure is made in light of the above-mentioned background, and in one aspect, its purpose is to provide a technology for masking a wide variety of information and for easily verifying the masking results. [Means for solving the problem]

[0007] According to one embodiment, a masking system is provided. The masking system comprises an input unit that accepts input of a document to be masked, a mask creation unit that identifies one or more mask locations and one or more masking categories from the document and creates one or more sets that associate each of the one or more mask locations with each of the one or more masking categories, and an output unit that outputs the masking results, including one or more sets, to a mask confirmation screen. Identifying one or more sets includes inputting the document into each of multiple AI models that analyze data in different formats, obtaining a portion of one or more sets from each of the multiple AI models, and aggregating the output data from each of the multiple AI models to create one or more sets. The mask confirmation screen includes a UI (User Interface) for editing one or more sets.

[0008] In a given scenario, the mask confirmation screen is configured to display temporary mask information by overlaying the document with one or more masked areas included in one or more sets, and to display one or more masking categories included in one or more sets. The temporary mask information is configured to allow viewing of the information that is to be masked.

[0009] In a given scenario, the masking system further includes a learning execution unit that, based on the fact that one or more sets have been edited on the mask confirmation screen, extracts the difference between the data before and after editing as training data for at least some of multiple AI models.

[0010] In a given scenario, the learning execution unit is configured to select an AI model to retrain from among multiple AI models based on the masking category associated with the difference.

[0011] In certain situations, the UI is configured to allow adjustment and deletion of one or more masked areas, and to add new masked areas to the temporary mask information.

[0012] In certain situations, the UI is configured to allow deletion, editing, and addition of new masking categories to one or more existing categories.

[0013] In certain scenarios, the output unit is configured to output the masked document after editing is complete on the mask confirmation screen.

[0014] In other embodiments, a masking program is provided that is executed by a masking system. The masking program causes the masking system to: accept input of a document to be masked; identify one or more masked areas and one or more masking categories from the document; create one or more sets that associate each of the one or more masked areas with each of the one or more masking categories; and output the masking results, including one or more sets, to a mask confirmation screen that includes a UI for editing one or more sets. Identifying one or more sets includes inputting the document into each of several AI models that analyze data in different formats; obtaining a portion of one or more sets from each of the several AI models; and aggregating the output data from each of the several AI models to create one or more sets. [Effects of the Invention]

[0015] According to an embodiment, various kinds of information can be masked and the masking result can be easily confirmed.

[0016] The above and other objects, features, aspects and advantages of this disclosure will become apparent from the following detailed description of the disclosure understood in connection with the accompanying drawings.

Brief Description of the Drawings

[0017] [Figure 1] It is a diagram showing an example of the operation of the masking system 100 according to this embodiment. [Figure 2] It is a diagram showing an example of the functional configuration of the masking system 100. [Figure 3] It is a diagram showing an example of the hardware configuration of the masking system 100. [Figure 4] It is a diagram showing a first example of the screen of the masking system 100. [Figure 5] It is a diagram showing a second example of the screen of the masking system 100. [Figure 6] It is a diagram showing a third example of the screen of the masking system 100. [Figure 7] It is a diagram showing a fourth example of the screen of the masking system 100. [Figure 8] It is a diagram showing a fifth example of the screen of the masking system 100. [Figure 9] It is a diagram showing a sixth example of the screen of the masking system 100. [Figure 10] It is a diagram showing an example of the processing procedure of the masking system 100.

Modes for Carrying Out the Invention

[0018] Hereinafter, embodiments of the technical idea according to the present disclosure will be described while referring to the drawings. In the following description, the same parts are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed descriptions thereof will not be repeated. Further, each embodiment, each modification example, each software or program configuration, each hardware configuration, each function, and each process, etc. may be selectively combined as appropriate.

[0019] <Operation Example of Masking System> FIG. 1 is a diagram showing an example of the operation of a masking system 100 according to the present embodiment. The masking system 100 provides a function of masking various types of information and a function for a user to efficiently confirm and correct the masking result compared to the prior art.

[0020] The masking system 100 receives the input of a document 10 that may include various types of information. The document 10 is realized as digital data and may include text, tables, graphics, images, information in any other format, and combined information thereof.

[0021] The masking system 100 uses a plurality of AI (Artificial Intelligence) models to mask the input document 10. More specifically, each AI model extracts a location to be masked (hereinafter referred to as a "mask location") from the document 10. In other words, each AI model outputs region information of the location to be masked on the document 10. As an example, the region information may include coordinate information indicating a rectangle. In the example of FIG. 1, mask locations 21, 22, 23, and 24 are extracted from the document 10.

[0022] Furthermore, each AI model outputs a masking category corresponding to the extracted masked areas. In the example in Figure 1, masking categories 31, 32, 33, and 34 are output, corresponding to masked areas 21, 22, 23, and 24, respectively. "Each masking category" indicates the type of each masked area. As an example, masking categories may include names, faces, seals, planning basis information, planned traffic volume, monetary information, dates of birth, addresses, telephone numbers, email addresses, company names, vehicle numbers, job titles, landowner names, underground utility location information, habitats of rare wild animals and plants, and interview information.

[0023] The masking system 100 aggregates the sets of masked areas and masking categories obtained from each AI model to create one or more sets that associate one or more masked areas with one or more masking categories. In the example in Figure 1, the masking system 100 creates four sets: "masked area 21, masking category 31", "masked area 22, masking category 32", "masked area 23, masking category 33", and "masked area 24, masking category 34".

[0024] The masking system 100 outputs a masking result 20 containing one or more sets. The masking system 100 outputs the masking result 20 on a UI with editing capabilities. The user can review and modify the masking result 20 through this UI. More specifically, the user can understand why each masked area is masked by referring to the masking category corresponding to each masked area. The user can determine whether each masked area is necessary by referring to the masking category. Furthermore, the user can determine whether the area of ​​the masked area is appropriate by referring to the masking category. For example, suppose masking category 31 is "address". In this case, the user can understand that by referring to masking category 31, they should check whether masked area 21 is an address and whether the entire address is properly masked. Conversely, if there is a contradiction between the masking category and the masked area, the user can understand that the masking category may be incorrect. For example, suppose masked area 22 is "telephone number" and masking category 32 is "name". In this case, the user can understand that an incorrect masking category 32 may have been associated with masked area 22 by referring to the set of masked area 22 and masking category 32. Thus, masking categories can indicate not only the classification of masked areas but also the reason why those areas are masked.

[0025] The user can view the masking result 20 on the UI and modify at least one of the masked areas and masking categories. Such modifications may include adding, deleting, or resizing masked areas. The user can effectively move masked areas by deleting and adding them. Such modifications may also include adding, deleting, or editing masking categories, adding associations with masked areas, or removing associations with masked areas.

[0026] The masking system 100 can use the edited result of the masking result 20 as training data for at least one of multiple AI models. The masking system 100 can acquire high-precision masking capabilities by repeating a series of cycles of file masking, accepting edits, and performing feedback.

[0027] As explained with reference to Figure 1, the masking system 100 can mask a document 10 containing information in any format by combining multiple AI models. The masking system 100 also outputs a masking result 20 which includes a set of masked areas and masking categories. The user can easily review and modify the masking result 20 by referring to each set included in the masking result 20. Furthermore, the masking system 100 can retrain at least one of the multiple AI models based on the editing results of the masking result 20. In other words, the masking system 100 has the function of masking a document 10 containing a wide variety of information, the function of easily modifying the masking result 20, and the function of feeding back the modified information to each AI model.

[0028] <Masking System Configuration> Figure 2 shows an example of the functional configuration of the masking system 100. The masking system 100 comprises an input unit 210, a mask creation unit 220, an output unit 230, a mask confirmation screen 240, and a learning execution unit 250. The mask creation unit 220 is also configured to communicate with multiple AI models. Each functional configuration shown in Figure 2 may be implemented as a program. In this case, each function of the masking system 100 can be realized by executing a program on the hardware shown in Figure 3.

[0029] The input unit 210 accepts input of documents to be masked. The input unit 210 may accept input of documents to be masked stored in the storage 303 (see Figure 3) of the masking system 100. Alternatively, the input unit 210 may accept input of documents to be masked from other devices. As an example, the input unit 210 may provide a screen 500 (see Figure 5) for inputting documents.

[0030] The mask creation unit 220 uses multiple AI models to create the masking result. More specifically, the mask creation unit 220 inputs the document to be masked into each of the multiple AI models. Next, the mask creation unit 220 obtains output information from each AI model. The output information includes a set of information about the masked area and information about the masking category associated with that masked area. The mask creation unit 220 aggregates the output information obtained from each AI model to create the masking result. The mask creation unit 220 outputs the created masking result to the output unit 230.

[0031] In the example shown in Figure 2, the mask creation unit 220 is configured to communicate with three AI models 260, 262, and 264. This is just one example, and the mask creation unit 220 can communicate with any number of AI models. The masking system 100 may also include multiple AI models. Furthermore, the masking system 100 may include multiple AI models and be configured to communicate with multiple external AI models.

[0032] The AI ​​model that should be used to extract the masked areas differs depending on the data to be masked. For example, suppose the data to be masked is text. In this case, a Natural Language Processing (NLP) AI model is suitable for extracting the masked areas. Another example is suppose the data to be masked is people and cars in an image. In this case, a YOLO AI model is suitable for extracting the masked areas. Yet another example is suppose the data to be masked is tabular data. In this case, a TableNet AI model is suitable for extracting the masked areas. Furthermore, suppose the document to be masked contains different types of data to be masked, such as text, images, and tabular data. In this case, it is desirable to extract the masked areas by combining multiple AI models. The masking system 100 may be configured to appropriately change the AI ​​model used depending on the data to be masked. The masking system 100 may also be configured to allow the addition, deletion, and modification of AI models.

[0033] The output unit 230 outputs the masking results. More specifically, the output unit 230 includes a mask confirmation screen 240. The mask confirmation screen 240 displays the masking results and also functions as a UI for editing the masking results. That is, the mask confirmation screen 240 includes a UI for editing one or more sets. The user edits the masking results via the mask confirmation screen 240. The output unit 230 outputs the edited masking results. The output unit 230 also outputs the edited masking results to the learning execution unit 250.

[0034] The learning execution unit 250 may use the edited masking result as training data for at least one AI model. For example, the learning execution unit 250 may perform the training process for at least one AI model based on the edited result. As another example, the learning execution unit 250 may send the edited result to another device as training data for the training process of at least one AI model. In this case, the other device may perform the training process for at least one AI model based on the edited result. The learning execution unit 250 may process the edited result to make it easier to use as training data.

[0035] The masking system 100 may accept input from multiple documents 10 and mask multiple documents 10 at once. In this case, the output unit 230 outputs multiple masking results 20. The user can edit some or all of the multiple masking results 20 at any time via the mask confirmation screen 240.

[0036] The masking system 100 may store various logs in the storage 303. These logs may include information about the process, the number of times the user edited the masking result 20, the edited file name, the start time of the masking process, the end time of the masking process, and information about the person responsible for masking the document 10. The information about the process may include one or more pieces of information about the masking process and the masking editing process. As an example, the masking system 100 may generate a log based on the fact that it has performed a masking process. As another example, the masking system 100 may generate a log based on the fact that it has received editing input for the masking result 20 from the user.

[0037] Furthermore, the masking system 100 may be configured to notify users of logs via email or other means at any time. The logs may be attached to the email or included in the email itself. Alternatively, the email may include a URL (Uniform Resource Locator) for a webpage to view the logs. The masking system 100 may also have a UI for registering log notification destinations and a UI for viewing logs. For example, by checking the logs, users can understand the time taken for masking work and use the logs to reduce the workload of masking work.

[0038] Furthermore, the masking system 100 may include information in the masking result 20 indicating areas where detection may be missed. Document 10 may contain areas where the judgment probability by each AI model is slightly below a threshold. In this case, the masking system 100 may include information in the masking result 20 indicating areas where the judgment probability by the AI ​​model is slightly below a threshold as "areas where detection may be missed." By referring to this information, the user can become aware of the existence of "areas where detection may be missed." The user can also choose whether or not to include "areas where detection may be missed" in the masked areas on the mask confirmation screen 240. The masking system 100 may also determine areas where the judgment probability by the AI ​​model is below a threshold within a predetermined range as "areas where detection may be missed."

[0039] Furthermore, the masking system 100 may be provided as a cloud service. In this case, the masking system 100 can mask one or more uploaded documents 10. The masking system 100 is also configured to allow the documents 10 to be uploaded encrypted via encrypted communication.

[0040] As explained with reference to Figure 2, the masking system 100 includes an input unit 210 that receives input of a document 10 to be masked, a mask creation unit 220 that identifies one or more mask locations and one or more masking categories from the document 10 and creates one or more sets that associate each of the one or more mask locations with each of the one or more masking categories, and an output unit 230 that outputs the masking results, including one or more sets, to a mask confirmation screen 240. Identifying one or more sets includes inputting the document 10 into each of multiple AI models that analyze data in different formats, obtaining a portion of one or more sets from each of the multiple AI models, and aggregating the output data from each of the multiple AI models to create one or more sets. The mask confirmation screen 240 includes a UI for editing one or more sets.

[0041] Figure 3 shows an example of the hardware configuration of the masking system 100. The masking system 100 does not necessarily have to include some of the configurations shown in Figure 3. Furthermore, the masking system 100 may include configurations not shown in Figure 3. In addition, the masking system 100 may include two or more of each configuration. The various functions of the masking system 100 described herein can be realized by executing a program on the hardware shown in Figure 3. Since the masking system 100 may consist of one or more information processing devices, it encompasses both systems and devices.

[0042] The masking system 100 includes a processor 301, memory 302, storage 303, an external device interface (IF) 304, an input interface 305, an output interface 306, and a communication interface 307. The masking system 100 also includes a bus 308 that connects these components.

[0043] The processor 301 can execute programs to implement various functions of the masking system 100. The processor 301 is composed of, for example, at least one integrated circuit. According to one embodiment, the integrated circuit may include at least one CPU (Central Processing Unit), at least one GPU (Graphics Processing Unit), at least one FPGA (Field Programmable Gate Array), at least one ASIC (Application Specific Integrated Circuit), at least one AI (Artificial Intelligence) chip, or a combination thereof.

[0044] Memory 302 functions as a workspace for processor 301. Memory 302 stores programs executed by processor 301 and data referenced by processor 301. Memory 302 can be implemented using DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory), etc.

[0045] Storage 303 is non-volatile memory that stores programs executed by processor 301 and data referenced by processor 301. Processor 301 executes programs read from storage 303 to memory 302 and references data read from storage 303 to memory 302. Storage 303 can be implemented by HDD (Hard Disk Drive), SSD (Solid State Drive), EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), or flash memory, etc.

[0046] The external device IF304 can be connected to any external device such as a printer, scanner, and external HDD. The external device IF304 can be implemented using a USB (Universal Serial Bus) terminal or the like.

[0047] The input interface 305 can be connected to any input device such as a keyboard, mouse, touchpad, or gamepad. The input interface 305 can be implemented using a USB terminal, PS / 2 terminal, or Bluetooth® module.

[0048] Output IF306 can be connected to any output device such as a CRT display, LCD display, or OLED display. Output IF306 can be implemented using a USB terminal, D-sub terminal, DVI (Digital Visual Interface) terminal, HDMI® (High-Definition Multimedia Interface) terminal, and DisplayPort terminal, etc.

[0049] The communication IF307 is connected to other devices via a wired or wireless network. The communication IF307 can be implemented using a wired LAN (Local Area Network) port and a Wi-Fi® (Wireless Fidelity) module, etc. The communication IF307 can send and receive data using communication protocols such as TCP / IP (Transmission Control Protocol / Internet Protocol) and UDP (User Datagram Protocol).

[0050] <Screenshot of the masking system> The masking system 100 provides screens 400 to 900. Each screen may be a web application screen or an installed application screen. The mask confirmation screen 240 encompasses screens 400 to 900. Screens 400 to 900 may be implemented as separate screens that can be transitioned to from one another. Alternatively, screens 400 to 900 may be implemented as a single screen. In this case, each of screens 400 to 900 represents a screen for a different mode, such as editing or confirmation.

[0051] Figure 4 shows a first example of the screen of the masking system 100. Screen 400 may be provided by the input unit 210. The user can select the masking categories to be included in the masking target from the masking category list 410 via screen 400. The masking categories displayed in the masking category list 410 correspond to the masking categories described with reference to Figure 1. In addition, each masking category is associated with at least one of several AI models. For example, "name" is often included as text in documents and is therefore associated with a natural language processing AI model. As another example, "face" is often included as an image in documents and is therefore associated with an AI model suitable for object detection.

[0052] The masking system 100 obtains information on the selected masking category via the screen 400. The masking system 100 may pre-store correspondence information between masking categories and AI models in the storage 303. In this case, the masking system 100 may change the AI ​​model used according to the information on the selected masking category. Alternatively, the masking system 100 may use all available AI models each time. The user can transition from screen 400 to screen 500 by pressing button 420.

[0053] Figure 5 shows a second example of the screen of the masking system 100. Screen 500 is the document upload screen. The user may drag and drop a document in the upload item 510, or specify the name or path of the document. The user can transition from screen 500 to screen 600 by pressing button 520. Based on the pressing of button 520, the masking system 100 retrieves the document to be masked via screen 500 and performs the masking process. The masking system 100 then outputs the masking result to screen 600.

[0054] Figure 6 shows a third example of the screen of the masking system 100. Screen 600 is the display screen for the masking results output by the masking system 100. In other words, screen 600 displays the masking results before receiving correction input from the user.

[0055] The masking results include temporary mask information 610 and masking category information 615. Temporary mask information 610 displays a document with one or more masked areas marked. Each masked area on screen 600 is not yet filled in, allowing the user to see what is written in each masked area. Masking category information 615 displays one or more masking categories associated with any of the masked areas.

[0056] In the example in Figure 6, the masking result includes masked areas 620A, 620B, 620C, and 620D. The masking result also includes masking categories 630A, 630B, 630C, and 630D. Each of the masked areas 620A, 620B, 620C, and 620D is associated with each of the masking categories 630A, 630B, 630C, and 630D. Each masked area may be associated with two or more masking categories. For example, suppose multiple AI models determine that masked area 620A corresponds to multiple masking categories. In this case, masked area 620A may correspond to multiple masking categories (e.g., "Name", "Company Name", etc.).

[0057] The user can confirm the location and extent of each masked area by referring to the temporary mask information 610. Furthermore, the user can confirm which masking category each masked area corresponds to by referring to the masking category information 615. In other words, the user can confirm the reason why each masked area was masked by referring to the masking category information 615.

[0058] In screen 600, when a masked area is selected using the mouse cursor or similar, the masking category corresponding to the selected masked area is highlighted. Conversely, in screen 600, when a masking category is selected using the mouse cursor or similar, the masked area corresponding to the selected masking category is highlighted. In the example in Figure 6, masked area 620D is selected, and masking category 630D is highlighted.

[0059] Screen 600 is configured to output a masking list 650 that includes page number information, mask location information, and masking category information. The page number information is the page number of the document to be masked. The mask location information indicates the coordinates and range of each mask location. The masking category information includes the masking category associated with any of the mask locations. The masking list 650, together with the masking list 850 described later, may be used as training data for at least one of multiple AI models. The masking system 100 may automatically create the masking list 650 as training data and save it to storage 303 or the like when it has created a masking result. Alternatively, screen 600 may output the masking list 650 based on the press of a save button 640 or the like. In this case, the user can obtain the masking list 650 for use as training data.

[0060] Screen 600 can transition to screen 700 for editing the masking result by pressing the edit button 645 or the next button 660. Alternatively, screen 600 switches to edit mode by pressing the edit button 645 or the next button 660.

[0061] As explained with reference to Figure 6, the mask confirmation screen 240 displays temporary mask information by overlaying document 10 with one or more masked areas included in one or more sets. The mask confirmation screen 240 is also configured to display one or more masking categories included in one or more sets. The temporary mask information is configured to allow viewing of the information that is to be masked.

[0062] Figure 7 shows a fourth example of the screen of the masking system 100. Screen 700 is a screen for editing the masking results output by the masking system 100. On screen 700, the user can delete one or more masked areas. The user can also adjust the range of each masked area on screen 700. The user can also add one or more masked areas on screen 700. The user can also change each masking category on screen 700. The user can also add and delete masking categories on screen 700. Furthermore, the user can associate desired masked areas with desired masking categories on screen 700.

[0063] Screen 700 includes temporary mask information 710, masking category information 715, and a masking category list 740. In the temporary mask information 710, the user can adjust the position and range of the desired mask area using a mouse cursor or the like. Comparing Figure 6 and Figure 7, it can be seen that the position and range of mask areas 620A, 620B, 620C, and 620D have been adjusted to become mask areas 720A, 720B, 720C, and 720D. In the masking category information 715, the user can change the desired masking category using a mouse cursor or the like. Comparing Figure 6 and Figure 7, it can be seen that masking categories 630A, 630B, 630C, and 630D have been changed to masking categories 730A, 730B, 730C, and 730D. More specifically, the user can select the desired masking category from the masking category list 740 and change one or more masking categories in the masking category information 715. For example, suppose a user selects "Company Name" from the masking category list 740 and selects masking category 730B. In this case, masking category 730B will be changed to "Company Name". Masking category information 715 may display all masking categories shown on screen 400 as options. Alternatively, the user may select a desired masking category from masking category information 715 and manually overwrite that masking category. For example, a user can overwrite a masking category selected from masking category information 715 with an unregistered masking category.

[0064] Screen 700 can transition to screen 800, which allows confirmation of the masking results after editing, by pressing the Next button 750. Alternatively, screen 700 switches to confirmation mode by pressing the Next button 750.

[0065] As explained with reference to Figure 7, the UI of the mask confirmation screen 240 is configured to allow adjustment and deletion of one or more mask areas, and to add new mask areas to the temporary mask information.

[0066] Furthermore, the UI of the mask confirmation screen 240 is configured to allow users to delete, edit, and add new masking categories to each of the one or more masking categories.

[0067] Figure 8 shows a fifth example of the screen of the masking system 100. Screen 800 is a screen for the user to check the masking results that have been edited. On screen 800, the user can check the masking results after editing. On screen 800, the masked areas are not yet filled in, and the user can check what is written in each masked area.

[0068] The learning execution unit 250 may automatically create a masking list 850 as training data and save it to storage 303 or the like when editing the masking results is complete. The masking list 850 may be used together with the aforementioned masking list 650 as training data for at least one of multiple AI models. The screen 800 may also output the masking list 850 based on the press of the save button 840 or the like. In this case, the user can obtain the masking list 850 to be used as training data. The items included in the masking list 850 are the same as the items included in the masking list 650.

[0069] The learning execution unit 250 may extract the difference between masking lists 650 and 850 and retrain at least one of multiple AI models based on that difference. Alternatively, the learning execution unit 250 may retrain at least one of multiple AI models using masking list 850 as the result of successful masking. Furthermore, the learning execution unit 250 may determine, based on the masking category, which AI model to apply the information of each record in masking lists 650 and 850 to training. For example, the first record of masking list 850 is "Company Name" and is likely to be text. In this case, the learning execution unit 250 may use the first record of masking list 850 as training data for a natural language processing AI model.

[0070] The learning execution unit 250 may train at least one of multiple AI models using the training data for each AI model obtained from the masking lists 650 and 850. The training data may be provided on a record-by-record basis in the masking lists 650 and 850. The learning execution unit 250 may also transmit the training data for each AI model obtained from the masking lists 650 and 850 to an external device. In this case, the external device trains at least one of the multiple AI models based on the received training data. The masking system 100 may receive and use multiple trained AI models from the external device at any time. The masking system 100 may also switch the AI ​​model it communicates with from the currently used AI model to a trained AI model at any time.

[0071] When the Next button 860 is pressed, screen 800 transitions to screen 900, which displays the document after masking is complete. Alternatively, when the Next button 860 is pressed, screen 800 switches to a mode that outputs the document after masking is complete.

[0072] As explained with reference to Figure 8, the masking system 100 further includes a learning execution unit 250 that, based on the fact that one or more sets have been edited on the mask confirmation screen 240, extracts the difference between the data before and after editing as training data for at least some of the AI ​​models.

[0073] Furthermore, the learning execution unit 250 is configured to select an AI model to be retrained from among multiple AI models based on the masking category associated with the difference.

[0074] Figure 9 shows a sixth example of the screen of the masking system 100. Screen 900 is a screen for displaying the document after masking is complete. In screen 900, each masked area is filled in. If the user confirms that there are no problems after checking screen 900, they can print the masked document by pressing the download button or the like. As explained with reference to Figure 8, the output unit 230 is configured to print the masked document after editing is completed on the mask confirmation screen 240.

[0075] <Processing procedure for the masking system> Figure 10 shows an example of the processing procedure of the masking system 100. The processor 301 may read a program for performing the processing in Figure 10 from the storage 303 into the memory 302 and execute the program. Part or all of the processing can also be implemented as a combination of circuit elements configured to perform the processing. The following steps may be performed in any order.

[0076] In step S1010, the masking system 100 accepts input of the document to be masked. For example, the masking system 100 may accept input of the document to be masked via the screen 500.

[0077] In step S1020, the masking system 100 inputs the acquired documents into each of the multiple AI models. For example, the masking system 100 inputs the documents into AI models 260, 262, and 264. The masking system 100 may also select which AI model to use from among the multiple AI models based on one or more categories selected on screen 400.

[0078] In step S1030, the masking system 100 acquires output information from each of the multiple AI models. The output information includes information on the set of masked areas and categories. In step S1040, the masking system 100 aggregates the acquired output information to create a masking result.

[0079] In step S1050, the masking system 100 outputs the masking result. The masking result is displayed on the mask confirmation screen 240. Furthermore, the masked area will not be filled in until the user has finished editing the masking result.

[0080] In step S1060, the masking system 100 accepts editing of the masked areas and categories. In step S1070, the masking system 100 feeds the editing results back to at least one of the multiple AI models. The learning execution unit 250 may use the editing results as training data to train at least one of the multiple AI models. The editing results may be processed to make them easier to use as training data.

[0081] As explained with reference to Figure 10, the masking system 100 can perform document masking by executing a masking program. The masking program causes the masking system to perform the following: accept input of a document to be masked; identify one or more masked areas and one or more masking categories from the document; create one or more sets by associating each of the one or more masked areas with each of the one or more masking categories; and output the masking results, including one or more sets, to a mask confirmation screen 240 which includes a UI for editing one or more sets. Identifying one or more sets includes inputting the document into each of multiple AI models that analyze data in different formats; obtaining a portion of one or more sets from each of the multiple AI models; and aggregating the output data from each of the multiple AI models to create one or more sets.

[0082] <Summary> As described above, the masking system 100 according to this embodiment can mask data in a wide variety of formats using multiple AI models. The masking system 100 can also output a set of one or more masked areas and one or more masking categories. The user can verify whether the masking is done properly by reviewing the set. Furthermore, the masking system 100 can use the user's edited masking results as training data for at least one of the multiple AI models.

[0083] The embodiments disclosed herein should be considered in all respects as illustrative and not restrictive. The scope of this disclosure is indicated by the claims rather than the foregoing description, and all modifications are intended to be equivalent to the claims. Furthermore, the disclosures described in the embodiments and each variation are intended to be implemented, as far as possible, individually or in combination. [Explanation of Symbols]

[0084] 10 Documents, 20 Masking Results, 21, 22, 23, 24, 620A, 620B, 620C, 620D, 720A, 720B, 720C, 720D Masked Areas, 31, 32, 33, 34, 630A, 630B, 630C, 630D, 730B, 730C, 730D Masking Categories, 100 Masking System, 210 Input Unit, 220 Mask Creation Unit, 230 Output Unit, 240 Mask Confirmation Screen, 250 Learning Execution Unit, 260, 262, 264 AI Model, 301 Processor, 302 Memory, 303 Storage, 304 External Device IF, 305 Input IF, 306 Output IF, 307 Communication IF, 308 Bus, 400, 500, 600, 700, 800, 900 Screen, 410, 740 Masking category list, 420, 520 Button, 510 Upload item, 610, 710 Temporary mask information, 615, 715 Masking category information, 645 Edit button, 650, 850 Masking list, 660, 750, 860 Next button.

Claims

1. An input section that accepts input from documents to be masked, A mask creation unit identifies one or more mask locations and one or more masking categories from the aforementioned documents, and creates one or more sets by associating each of the one or more mask locations with each of the one or more masking categories. The system includes an output unit that outputs the masking results, including one or more sets, to a mask confirmation screen. Identifying one or more sets means This involves inputting the aforementioned documents into each of several AI models that analyze data in different formats, From each of the aforementioned multiple AI models, obtain a portion of one or more sets, This includes aggregating the output data of each of the multiple AI models to create one or more sets, The mask confirmation screen is a masking system that includes a user interface (UI) for editing one or more sets.

2. The aforementioned mask confirmation screen is: The document is overlaid with one or more mask locations included in the set, and temporary mask information is displayed. It is configured to display the one or more masking categories included in the one or more sets, The masking system according to claim 1, wherein the temporary mask information is configured to allow viewing of information that is to be masked.

3. The masking system according to claim 2, further comprising a learning execution unit that, based on the fact that one or more sets have been edited on the mask confirmation screen, extracts the difference between the data before and after editing as training data for at least some of the multiple AI models.

4. The masking system according to claim 3, wherein the learning execution unit is configured to select an AI model to be retrained from among the plurality of AI models based on the masking category associated with the difference.

5. The masking system according to any one of claims 2 to 4, wherein the UI is configured to allow adjustment, deletion, and addition of new mask locations to the temporary mask information.

6. The masking system according to any one of claims 2 to 4, wherein the UI is configured to allow deletion, editing, and addition of each of the one or more masking categories.

7. The masking system according to claim 1, wherein the output unit is configured to output the masked document after editing is completed on the mask confirmation screen.

8. A masking program executed by a masking system, Accepting input from documents to be masked, From the aforementioned documents, identify one or more masked areas and one or more masking categories, and create one or more sets by associating each of the one or more masked areas with each of the one or more masking categories. The masking system is instructed to output the masking result, which includes one or more sets, to a mask confirmation screen that includes a UI for editing one or more sets. Identifying one or more sets means This involves inputting the aforementioned documents into each of several AI models that analyze data in different formats, From each of the aforementioned multiple AI models, obtain a portion of one or more sets, A masking program that includes aggregating the output data of each of the multiple AI models to create one or more sets.

Images