Confidentiality determination apparatus, confidentiality determination method, and program

Abstract

To provide a novel technique for concealing parts of a text.SOLUTION: A confidentiality determination apparatus according to the present disclosure is configured to: acquire a target text; extract confidentiality candidate words to be concealed from the target text; acquire answer data regarding answers to questions related to the confidentiality candidate words from an answer unit configured to generate the answers to the questions; and determine whether to conceal the confidentiality candidate words based on the answer data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a confidentiality determination device, a confidentiality determination method, and a program. [Background technology]

[0002] Techniques for concealing parts of text have been developed. For example, Patent Document 1 discloses a technique for extracting named entities such as people's names from input information and converting the extracted named entities into obscured character strings. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-331329 Summary of the Invention [Problem to be solved by the invention]

[0004] The device in Patent Document 1 converts all named entities, such as people's names, into obscured character strings. The present disclosure has been made in light of this problem, and one of its purposes is to provide a new technology for concealing part of a sentence. [Means for solving the problem]

[0005] The confidentiality determination device according to the present disclosure includes an acquisition unit that acquires a target text, an extraction means that extracts confidentiality candidate words, which are candidates for words to be confidential, from the target text, and a determination means that acquires answer data related to answers to questions related to the confidentiality candidate words from an answer means configured to generate answers to questions, and determines whether or not to conceal the confidentiality candidate words based on the answer data.

[0006] A confidentiality determination method according to the present disclosure is executed by a computer and includes an acquisition step of acquiring a target text, an extraction step of extracting confidentiality candidate words, which are candidates for words to be confidential, from the target text, and a determination step of acquiring answer data related to answers to questions related to the confidentiality candidate words from an answering means configured to generate answers to questions, and determining whether to conceal the confidentiality candidate words based on the answer data.

[0007] The program according to the present disclosure causes a computer to execute an acquisition step of acquiring a target text, an extraction step of extracting, from the target text, candidate words for concealment that are candidates for words to be concealed, and a determination step of acquiring answer data regarding answers to questions regarding the candidate words for concealment from an answering means configured to generate answers to questions, and determining whether or not to conceal the candidate words for concealment based on the answer data. [Effects of the Invention]

[0008] According to the present disclosure, a new technique for concealing a portion of a text is provided. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 10 is a diagram illustrating an example of an outline of the operation of the confidentiality determination device. [Figure 2] FIG. 2 is a block diagram illustrating a functional configuration of a confidentiality determination device. [Figure 3] FIG. 2 is a block diagram illustrating a hardware configuration of a computer that realizes the confidentiality determination device. [Figure 4] 10 is a flowchart illustrating a flow of processing executed by the confidentiality determination device. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In each drawing, the same or corresponding elements are designated by the same reference numerals, and duplicate explanations will be omitted as necessary for clarity. Furthermore, unless otherwise specified, predetermined values ​​such as predetermined values ​​and threshold values ​​are stored in advance in a storage device accessible from a device that uses the values. Furthermore, unless otherwise specified, the storage unit is composed of one or any number of storage devices.

[0011] <Summary> Fig. 1 is a diagram illustrating an example of an outline of the operation of the confidentiality determination device 2000. Here, Fig. 1 is a diagram for facilitating understanding of the outline of the confidentiality determination device 2000, and the operation of the confidentiality determination device 2000 is not limited to that shown in Fig. 1.

[0012] The confidentiality determination device 2000 identifies confidential words 60, which are words that should be kept confidential, from among words included in the target text 10. The target text 10 is text data representing an arbitrary sentence.

[0013] First, the confidentiality determination device 2000 extracts confidentiality candidate words 20, which are candidates for words to be confidential (i.e., candidates for confidential words 60), from the target text 10. The confidentiality candidate words 20 are, for example, words that represent information that can identify an individual or a group. Here, information that can identify a specific individual or a specific group is also called PII (Personal Identifiable Information).

[0014] The confidentiality determination device 2000 determines whether or not the confidentiality candidate words 20 should be concealed. The confidentiality candidate words 20 that are determined to be confidential are treated as confidential words 60. On the other hand, the confidentiality candidate words 20 that are determined not to be confidential are not treated as confidential words 60.

[0015] The answering unit 50 is used to determine whether or not the concealment candidate word 20 should be concealed. The answering unit 50 is configured to output data regarding an answer to a question (hereinafter, answer data) in response to input of text data representing the question (hereinafter, question text).

[0016] For example, the answering unit 50 is realized by a language model called a large language model (LLM). However, the answering unit 50 may be realized by a language model that is not classified as an LLM. Furthermore, the answering unit 50 may be realized using something other than a language model.

[0017] The answering unit 50 may be provided inside the confidentiality determination device 2000 or outside the confidentiality determination device 2000. In the latter case, for example, the answering unit 50 is realized in another device communicably connected to the confidentiality determination device 2000 via a network.

[0018] The confidentiality determination device 2000 generates a question text 30 representing a question regarding the confidentiality candidate word 20. For example, if the confidentiality candidate word 20 is the word "ABC," the question text 30 is text data representing a question inquiring about the meaning of the confidentiality candidate word 20, such as "What is ABC?"

[0019] The confidentiality determination device 2000 acquires answer data 40 using the question text 30. The answer data 40 is answer data obtained from the answer section 50 by inputting the question text 30 to the answer section 50. For example, the answer data 40 includes text data representing an answer (hereinafter, answer text).

[0020] The confidentiality determination device 2000 determines whether or not the confidentiality candidate word 20 should be concealed, using the answer data 40. In other words, the confidentiality determination device 2000 determines whether or not the confidentiality candidate word 20 should be treated as a confidential word 60, using the answer data 40.

[0021] Here, a plurality of concealment candidate words 20 may be extracted from the target text 10. The concealment determination device 2000, for example, generates a question text 30 for each of the plurality of concealment candidate words 20, and determines whether or not each of the concealment candidate words 20 should be concealed. However, as will be described later, the concealment determination device 2000 may use one question text 30 that includes a question regarding a plurality of concealment candidate words 20.

[0022] <Examples of effects> According to the present disclosure, concealment candidate words 20 are extracted from a target text 10, and whether or not to conceal the concealment candidate words 20 is determined using answer data 40 obtained from an answering unit 50 in response to a question regarding the concealment candidate words 20. Therefore, not all of the concealment candidate words 20 extracted from the target text 10 are necessarily concealed. In this way, the concealment determination device 2000 provides a new technique for concealing a part of a sentence.

[0023] For example, suppose that PII such as a person's name is extracted as a confidentiality candidate word 20. The PII contained in the target text 10 may include public information that does not need to be confidential. Examples of public information that does not need to be confidential include names of historical figures and famous people. Since such public information is already publicly available on the Internet, it is considered that there is no problem with including it in the target text 10.

[0024] Here, if a large amount of information is concealed in the target text 10, it may become difficult to understand the content of the target text 10. Therefore, it is preferable to conceal information that should be concealed in the target text 10, while not concealing information that does not need to be concealed.

[0025] According to the concealment determination device 2000, it is determined whether or not to conceal the concealment candidate words 20 using the response data 40 obtained for the concealment candidate words 20. Therefore, it is possible to accurately distinguish between the concealment candidate words 20 that should be concealed and the concealment candidate words 20 that do not need to be concealed for the target text 10. Therefore, by using the concealment determination device 2000, it is possible to conceal information that should be concealed for the target text 10 while not concealing information that does not need to be concealed. Therefore, by using the concealment determination device 2000, it is possible to conceal information that should be concealed in the target text 10 while preventing the content of the target text 10 from becoming difficult to understand.

[0026] <Example of use of the confidentiality determination device 2000> For example, the confidentiality determination device 2000 can be used to find information that should be kept secret (hereinafter, confidential information) from a question before inputting the question to a language model that is publicly available on the Internet, etc. If a question containing confidential information is input to a public language model, there is a risk that the confidential information will be leaked to the outside (e.g., outside the company). Therefore, it is preferable to conceal the confidential information in the question.

[0027] On the other hand, if a large part of the question is concealed, it becomes difficult for the language model to understand the content of the question, and as a result, the language model may not be able to provide an accurate answer to the question.

[0028] By using the confidentiality determination device 2000, it is possible to prevent information that does not need to be confidential from being concealed while keeping information that should be confidential in the question confidential. Therefore, by using the confidentiality determination device 2000, it is possible to prevent confidential information from leaking to the outside and obtain an accurate answer to the question from the language model.

[0029] The confidentiality determination device 2000 of this embodiment will be described in more detail below.

[0030] <Example of functional configuration> 2 is a block diagram illustrating an example of the functional configuration of the confidentiality determination device 2000. In the example of FIG. 2, the confidentiality determination device 2000 includes an acquisition unit 2020, an extraction unit 2040, and a determination unit 2060. The acquisition unit 2020 acquires a target text 10. The extraction unit 2040 extracts confidentiality candidate words 20 from the target text 10. The determination unit 2060 generates a question text 30 expressing a question regarding the confidentiality candidate words 20. Furthermore, the determination unit 2060 acquires answer data 40 to the question text 30 from the answer unit 50. Then, the determination unit 2060 uses the answer data 40 to determine whether or not the confidentiality candidate words 20 should be confidential.

[0031] <Example of hardware configuration> Each functional component of the confidentiality determination device 2000 may be realized by hardware that realizes each functional component (e.g., a hardwired electronic circuit, etc.), or may be realized by a combination of hardware and software (e.g., a combination of an electronic circuit and a program that controls it, etc.). Below, a case where each functional component of the confidentiality determination device 2000 is realized by a combination of hardware and software will be further described.

[0032] 3 is a block diagram illustrating an example of a hardware configuration of a computer 1000 that realizes the confidentiality determination device 2000. The computer 1000 is any computer. For example, the computer 1000 is a stationary computer such as a PC (Personal Computer) or a server machine. Alternatively, the computer 1000 may be a portable computer such as a smartphone or a tablet terminal. The computer 1000 may be a dedicated computer designed to realize the confidentiality determination device 2000, or may be a general-purpose computer.

[0033] For example, by installing a predetermined application in the computer 1000, each function of the confidentiality determination device 2000 is realized by the computer 1000. The application is configured by a program for realizing each functional component of the confidentiality determination device 2000.

[0034] When the answering unit 50 is realized inside the confidentiality determination device 2000, the application may also include a program for realizing the answering unit 50. However, the application for realizing the answering unit 50 and the application for realizing each functional configuration unit of the confidentiality determination device 2000 may be different applications.

[0035] The program may be acquired by any method. For example, the program may be acquired from a storage medium (such as a DVD (Digital Versatile Disc) or a USB (Universal Serial Bus) memory) on which the program is stored. Alternatively, the program may be acquired by downloading the program from a server device that manages the storage device on which the program is stored.

[0036] The computer 1000 has a bus 1020, a processor 1040, a memory 1060, a storage device 1080, an input / output interface 1100, and a network interface 1120. The bus 1020 is a data transmission path for the processor 1040, the memory 1060, the storage device 1080, the input / output interface 1100, and the network interface 1120 to transmit and receive data to and from each other. However, the method for connecting the processor 1040 and the like to each other is not limited to a bus connection.

[0037] The processor 1040 is a variety of processors, such as a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA). The memory 1060 is a main storage device realized using a random access memory (RAM) or the like. The storage device 1080 is an auxiliary storage device realized using a hard disk, a solid state drive (SSD), a memory card, a read only memory (ROM), or the like.

[0038] The input / output interface 1100 is an interface for connecting the computer 1000 to an input / output device. For example, the input / output interface 1100 is connected to an input device such as a keyboard and an output device such as a display device.

[0039] The network interface 1120 is an interface for connecting the computer 1000 to a network. This network may be a LAN (Local Area Network) or a WAN (Wide Area Network).

[0040] The storage device 1080 stores a program (a program that realizes the above-mentioned application) that realizes each functional component of the confidentiality determination device 2000. The processor 1040 reads this program into the memory 1060 and executes it, thereby realizing each functional component of the confidentiality determination device 2000.

[0041] The confidentiality determination device 2000 may be realized by one computer 1000 or by multiple computers 1000. In the latter case, the configurations of the computers 1000 do not need to be the same, and can be different from each other.

[0042] Note that when the answering unit 50 is realized outside the confidentiality determination device 2000, any hardware configuration may be used for the device that realizes the answering unit 50. For example, the hardware configuration of the device that realizes the answering unit 50 may be represented in FIG. 3, similar to the hardware configuration of the confidentiality determination device 2000.

[0043] <Processing flow> 4 is a flowchart illustrating the flow of processing executed by the concealment determination device 2000. The acquisition unit 2020 acquires the target text 10 (S102). The extraction unit 2040 extracts concealment candidate words 20 from the target text 10 (S104). The determination unit 2060 generates a question text 30 (S106). The determination unit 2060 acquires answer data 40 corresponding to the question text 30 (S108). The determination unit 2060 uses the answer data 40 to determine whether the concealment candidate words 20 should be concealed (S110).

[0044] <About answer section 50> The answering section 50 is configured to output answer data in response to input of question text. As described above, the answer data includes, for example, the answer text.

[0045] For example, the answering unit 50 is realized by a general-purpose language model that is pre-trained to output an answer to an arbitrary question. Alternatively, for example, the answering unit 50 may be realized by a language model that is trained specifically for the confidentiality determination device 2000. In the latter case, for example, the language model is trained using a plurality of training data in which arbitrary question texts are associated with ground truth answer data for the questions. Furthermore, the answering unit 50 may be realized by a language model that is further trained on a general-purpose language model using the above-mentioned training data.

[0046] The answer data may further include a value indicating the accuracy of the answer (hereinafter, "answer accuracy") in addition to the answer text or instead of the answer text. The answer accuracy indicates the probability that the answer is correct. It can also be said that the answer accuracy indicates the degree of confidence that the answering unit 50 has in the correctness of the answer.

[0047] The answer accuracy can be expressed, for example, by the following formula (1).

number

[0048] P(w_i|V_i) represents the conditional probability of generating word w_i when words other than the i-th word are given. Here, P_i(w_i|V_i) can also be said to represent the probability that w_i is located at the i-th position in the answer text. Therefore, in formula (1), the answer probability, which is the probability of the entire answer text, is expressed as the total product of P(w_i|V_i).

[0049] The answer certainty may be calculated using the following formula (2).

number

[0050] <Getting target text 10: S102> The acquisition unit 2020 acquires the target text 10 (S102). There are various methods for acquiring the target text 10. For example, the target text 10 is transmitted from another arbitrary device to the concealment determination device 2000. In this case, the acquisition unit 2020 acquires the target text 10 by receiving the target text 10 transmitted from the other device.

[0051] The device that transmits the target text 10 to the confidentiality determination device 2000 is, for example, a terminal (hereinafter referred to as a user terminal) used by a user of the confidentiality determination device 2000. The user terminal is any computer such as a smartphone or a PC.

[0052] For example, the confidentiality determination device 2000 provides a user terminal with an input screen for inputting the target text 10. In this case, the user inputs the target text 10 into the input screen, thereby providing the target text 10 to the confidentiality determination device 2000. The input screen is provided as, for example, a web page.

[0053] The target text 10 may be directly input to the confidentiality determination device 2000. In this case, for example, the above-mentioned input screen is displayed on a display device connected to the confidentiality determination device 2000. A user inputs the target text 10 to the confidentiality determination device 2000 using an input device such as a keyboard or a touch panel connected to the confidentiality determination device 2000. Then, the acquisition unit 2020 acquires the input target text 10.

[0054] Alternatively, for example, the acquiring unit 2020 may acquire the target text 10 by acquiring any document file including the target text 10. For example, the confidentiality determination device 2000 receives a document file transmitted from another device such as a user terminal, and acquires the target text 10 from the document file. Alternatively, for example, the confidentiality determination device 2000 may provide an input screen on which a document file can be specified to another device such as a user terminal. In this case, the acquiring unit 2020 acquires the document file specified on the input screen from the storage unit, and acquires the target text 10 from the document file.

[0055] <Extraction of confidentiality candidate words 20: S104> The extraction unit 2040 extracts (S104) concealment candidate words 20 from the target text 10. As described above, the concealment candidate words 20 are candidates for words to be concealed.

[0056] For example, the anonymization candidate words 20 are words that represent information that can identify an individual. The information that can identify an individual is, for example, the name of the individual, the identification information of the individual (such as an employee number, a membership number, or an account identifier), or the email address of the individual. In addition, for example, the anonymization candidate words 20 are words that represent information that can identify a group. The information that can identify a group is, for example, the name of the group or the identification information of the group.

[0057] There are various methods for extracting the concealment candidate words 20 from the target text 10. For example, the extraction unit 2040 extracts one or more PIIs from the target text 10 by applying a PII extraction process to the target text 10. Then, the extraction unit 2040 treats each PII extracted from the target text 10 as a concealment candidate word 20.

[0058] Here, the PII extraction process can identify the type of PII for each extracted PII. The type of PII can be, for example, a person's name, a group name, identification information, or an email address. Therefore, by using the PII extraction process, the extraction unit 2040 can extract one or more anonymization candidate words 20 from the target text 10 and identify the type of each anonymization candidate word 20.

[0059] The method for extracting the concealment candidate words 20 is not limited to the method for executing the PII extraction process. For example, the extraction unit 2040 extracts nouns from the target text 10 by applying a part-of-speech decomposition process to the target text 10. Furthermore, the extraction unit 2040 determines whether each noun extracted from the target text 10 is a noun of a predetermined type that should be extracted as a concealment candidate word 20. The extraction unit 2040 treats each noun determined to fall into the predetermined type that should be extracted as a concealment candidate word 20 as a concealment candidate word 20. Note that the predetermined type here is, for example, the same as the type of PII described above.

[0060] <Generation of question text 30: S106> The determination unit 2060 generates a question text 30 representing a question regarding the concealment candidate word 20 (S106). For example, the question text 30 represents a question inquiring about the meaning of the concealment candidate word 20. Such a question text 30 represents a question such as, "Please tell me the meaning of ABC." In this example, ABC is the concealment candidate word 20.

[0061] Alternatively, the question text 30 may express a question inquiring about information related to the concealment candidate word 20, regardless of its meaning, such as "Please tell me about ABC."

[0062] For example, the determination unit 2060 generates a question text 30 by applying the concealment candidate words 20 to a predetermined template text. The template text, for example, represents a question about a specific word and includes text (hereinafter, a position mark) indicating a position where the concealment candidate words 20 are to be embedded.

[0063] For example, suppose the template text is "Please tell me the meaning of X." In this case, X is a position mark. The determination unit 2060 generates a question text 30 by replacing X in the template text with a concealment candidate word 20.

[0064] A template text may be prepared for each type of concealment candidate word 20. In this way, a question text more suitable for each type of concealment candidate word 20 can be generated.

[0065] For example, for a type called a person's name, a template text such as "What kind of person is X?" is prepared. Also, for a type called a group name, a template text such as "What kind of group is X?" is prepared.

[0066] When a template text is prepared for each type of concealment candidate word 20, the determination unit 2060 acquires the template text corresponding to the type of concealment candidate word 20. Then, the determination unit 2060 generates the question text 30 using the acquired template text and the concealment candidate word 20.

[0067] <<About the case where multiple confidentiality candidate words 20 are extracted>> A method for generating a question text 30 will be illustrated for a case where multiple concealment candidate words 20 are extracted from the target text 10. For example, the determination unit 2060 generates one question text 30 for each concealment candidate word 20. The method for generating one question text 30 for one concealment candidate word 20 is as described above.

[0068] Alternatively, for example, the extraction unit 2040 may generate one question text 30 representing a question regarding a plurality of concealment candidate words 20. For example, the extraction unit 2040 generates the question text 30 by replacing the position marks in the template text described above with text in which all of the concealment candidate words 20 are listed.

[0069] As an example, suppose that the concealment candidate words 20, ABC, DEF, and GHI, are extracted from the target text 10. Also, suppose that the template text is "Please tell me the meaning of X." In this case, the extraction unit 2040 replaces the position marks in the template text with the text "ABC, DEF, and GHI," in which all the concealment candidate words 20 are listed. As a result, a question text 30, "Please tell me the meaning of ABC, DEF, and GHI," is generated.

[0070] <Acquisition of response data 40: S108> The determination unit 2060 obtains the answer data 40 using the question text 30 (S108). When the answering unit 50 is provided inside the confidentiality determination device 2000, the determination unit 2060 inputs the question text 30 to the answering unit 50, thereby obtaining the answer data 40 output from the answering unit 50.

[0071] On the other hand, when the answering unit 50 is provided outside the confidentiality determination device 2000, for example, the determination unit 2060 transmits the question text 30 to a device in which the answering unit 50 is realized (hereinafter referred to as the answering device). The answering device receives the question text 30 by the determination unit 2060 and inputs the received question text 30 to the answering unit 50. As a result, the answering device obtains answer data 40 from the answering unit 50. The answering device transmits the answer data 40 to the confidentiality determination device 2000. The determination unit 2060 acquires the answer data 40 by receiving the answer data 40 transmitted by the answering device.

[0072] <Decision on whether to keep it secret: S110> The determination unit 2060 determines whether the concealment candidate words 20 should be concealed or not by using the answer data 40 (S110). A specific method for determining whether the concealment candidate words 20 should be concealed or not may differ depending on the type of information represented by the answer data 40. Below, examples of methods for determining whether the concealment candidate words 20 should be concealed or not for each type of information represented by the answer data 40 will be described.

[0073] <<When response data 40 indicates the accuracy of the response>> As described above, the answer data 40 may indicate answer certainty. Here, suppose that the concealment candidate word 20 represents information that does not need to be kept secret (for example, the name of a public person or group). In this case, it can be predicted that the answer certainty of the answer generated by the answering unit 50 for a question regarding the concealment candidate word 20 will be relatively high. On the other hand, suppose that the concealment candidate word 20 represents information that should be kept secret (for example, the name of a public individual or group). In this case, it can be predicted that the answer certainty of the answer generated by the answering unit 50 for a question regarding the concealment candidate word 20 will be relatively low. Therefore, if the answer certainty is low, it is considered that the concealment candidate word 20 should be kept secret.

[0074] Therefore, the determination unit 2060 determines whether the answer accuracy included in the answer data 40 is equal to or less than a threshold value. This threshold value is a predetermined threshold value for determining whether the answer accuracy of the answer text corresponding to the question text 30 is sufficiently small.

[0075] If the answer certainty included in the answer data 40 is equal to or less than the threshold, the determination unit 2060 determines to conceal the concealment candidate word 20. On the other hand, if the answer certainty included in the answer data 40 is greater than the threshold, the determination unit 2060 determines not to conceal the concealment candidate word 20.

[0076] <<When answer data 40 indicates answer text>> As described above, the answer data 40 may represent answer text. Here, it is considered that answer text generated by the answering unit 50 in response to a question about information that should be kept secret is more likely to be unnatural than answer text generated by the answering unit 50 in response to a question about information that does not need to be kept secret. This is because it is more difficult to obtain knowledge about information that should be kept secret than about information that does not need to be kept secret.

[0077] Therefore, the determination unit 2060 determines whether the sentence represented by the answer text included in the answer data 40 is unnatural. If the sentence represented by the answer text included in the answer data 40 is unnatural, the determination unit 2060 determines to conceal the concealment candidate words 20. On the other hand, if the sentence represented by the answer text included in the answer data 40 is not unnatural, the determination unit 2060 determines not to conceal the concealment candidate words 20.

[0078] Whether a sentence represented by an answer text is unnatural or not is determined, for example, using an arbitrary machine learning model (e.g., a neural network). This model is called a determination model. The determination model is trained in advance so that, in response to input of text data representing a sentence, it outputs a label indicating whether the input sentence is unnatural or not.

[0079] The determination unit 2060 inputs the response text into the determination model. If a label indicating that the sentence is unnatural is output from the determination model, the determination unit 2060 determines that the concealment candidate word 20 should be concealed. On the other hand, if a label indicating that the sentence is not unnatural is output from the determination model, the determination unit 2060 determines that the concealment candidate word 20 should not be concealed.

[0080] <<When multiple confidentiality candidate words 20 are extracted>> When a plurality of concealment candidate words 20 are extracted, a question text 30 is generated for each of the concealment candidate words 20. In this case, the determination unit 2060 obtains answer data 40 corresponding to the question text 30 for each of the concealment candidate words 20. Then, for each of the concealment candidate words 20, the determination unit 2060 determines whether or not to conceal the concealment candidate word 20 using the answer data 40 corresponding to the concealment candidate word 20.

[0081] On the other hand, suppose that one question text 30 is generated that summarizes questions regarding all the concealment candidate words 20. In this case, answer data 40 including answers regarding all the concealment candidate words 20 is obtained from the answer section 50.

[0082] For example, the determination unit 2060 acquires information about each concealment candidate word 20 from the response data 40. Then, the determination unit 2060 uses the information obtained about each concealment candidate word 20 to determine whether or not to conceal the concealment candidate word 20.

[0083] For example, assume that the answer data 40 indicates an answer text. In this case, the determination unit 2060 extracts, for each concealment candidate word 20, text (hereinafter referred to as partial answer text) that represents information related to the concealment candidate word 20 from the answer text. Then, the determination unit 2060 determines whether or not to conceal the concealment candidate word 20 corresponding to the partial answer text, depending on the unnaturalness of the partial answer text.

[0084] For example, suppose the question text 30 is "Please tell me the meanings of ABC, DEF, and GHI." In this case, the answer text is considered to indicate the answer to each concealment candidate word 20, such as "ABC means... DEF means... GHI means...."

[0085] The determination unit 2060 uses each partial answer text obtained from the answer text to determine whether or not to conceal each concealment candidate word 20. For example, the partial answer text corresponding to the concealment candidate word 20 ABC is "ABC means...." Therefore, the determination unit 2060 determines whether or not to conceal the concealment candidate word 20 ABC by determining whether or not the sentence represented by this partial answer text is unnatural.

[0086] Assume that the answer accuracy is used in a case where one question text 30 includes questions about a plurality of concealment candidate words 20. In this case, the answering unit 50 is configured to output the answer accuracy for each concealment candidate word 20. That is, for each concealment candidate word 20, the answer accuracy of the partial answer text corresponding to that concealment candidate word 20 is output from the answering unit 50.

[0087] Therefore, the determination unit 2060 determines whether to conceal the concealment candidate word 20 based on the answer accuracy of the partial answer text corresponding to the concealment candidate word 20. If the answer accuracy of the partial answer text corresponding to the concealment candidate word 20 is equal to or less than a threshold, the determination unit 2060 determines to conceal the concealment candidate word 20. On the other hand, if the answer accuracy of the partial answer text corresponding to the concealment candidate word 20 is greater than the threshold, the determination unit 2060 determines not to conceal the concealment candidate word 20.

[0088] <Output of processing results> The confidentiality determination device 2000 outputs information indicating the processing result (hereinafter, output information). For example, the output information indicates confidentiality candidate words 20 output from the target text 10 and the determination result as to whether or not to conceal the confidentiality candidate words 20. When multiple confidentiality candidate words 20 are extracted from the target text 10, the output information indicates pairs of (confidentiality candidate words 20, whether or not to conceal the confidentiality candidate words 20) for all the confidentiality candidate words 20.

[0089] The concealment determination device 2000 may perform a concealment process on the target text 10 so that the concealment candidate words 20 determined to be concealed (i.e., the concealed words 60) are concealed. The text generated from the target text 10 by the concealment process is called an "output text."

[0090] For example, the confidentiality determination device 2000 generates an output text by replacing the confidential words 60 with predetermined words (hereinafter, replacement words) in the target text 10. The replacement words are preferably words whose contents are unknown compared to the original confidential words 60. For example, the replacement words are composed of one or more symbols (such as asterisks).

[0091] A replacement word may be prepared for each type of secret word 60. The replacement word prepared for each type of secret word 60 is, for example, a general-purpose expression for the word of that type. For example, "Person P" can be used as a general-purpose expression for a person's name. Also, "Group G" can be used as a general-purpose expression for a group's name.

[0092] When replacement words are prepared for each type of secret word 60, the confidentiality determination device 2000 identifies a replacement word that corresponds to the type of secret word 60. Then, the confidentiality determination device 2000 generates an output text by replacing the secret word 60 with the identified replacement word in the target text 10. By generating an output text using replacement words according to the type of secret word 60 in this way, it is possible to generate a more natural output text compared to a case in which an output text is generated without considering the type of secret word 60. Therefore, it is possible to make the output text easy to read while concealing information that should be kept secret.

[0093] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0094] Each drawing is merely an example for describing one or more embodiments. Each drawing may relate not only to one particular embodiment, but also to one or more other embodiments. As will be understood by those skilled in the art, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings to create, for example, an embodiment not explicitly shown or described. Not all features or steps shown in any one drawing are necessary to describe an exemplary embodiment, and some features or steps may be omitted. The order of steps described in any drawing may be changed as appropriate.

[0095] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes. (Appendix 1) an acquisition unit for acquiring a target text; an extraction means for extracting candidate words to be concealed from the target text; a determination means for obtaining answer data relating to an answer to a question about the concealment candidate word from an answer means configured to generate an answer to the question, and determining whether or not to conceal the concealment candidate word based on the answer data. (Appendix 2) The determination means generating a question text representing a question regarding the concealment candidate word; 2. The confidentiality determination device according to claim 1, wherein the question text is input to the answering means, and the answer data output from the answering means is obtained. (Appendix 3) the answer data indicates an answer text that represents an answer to a question regarding the concealment candidate word; 3. The confidentiality determination device according to claim 1, wherein the determination means determines that the confidentiality candidate word is to be confidential when a sentence represented by the answer text is unnatural. (Appendix 4) The determination means a judgment model for judging whether a sentence represented by an input text is unnatural; 4. The confidentiality determination device according to claim 3, wherein the answer text is input to the determination model to determine whether a sentence represented by the answer text is unnatural. (Appendix 5) the answer data indicates the accuracy of an answer to a question regarding the concealment candidate word; 3. The confidentiality determination device according to claim 1, wherein the determination means determines that the confidentiality candidate word is to be confidential when the certainty is equal to or less than a threshold value. (Appendix 6) 6. The anonymity determination device according to claim 1, wherein the extraction means extracts words that can identify an individual or a group as the anonymity candidate words. (Appendix 7) 7. The confidentiality determination device according to any one of appendices 1 to 6, wherein the answering means is configured with a language model trained to output an answer corresponding to a question in response to the question being input. (Appendix 8) an acquisition step of acquiring a target text; an extraction step of extracting candidate words to be concealed from the target text; a determination step of obtaining answer data regarding an answer to a question regarding the concealment candidate word from an answering means configured to generate an answer to the question, and determining whether or not to conceal the concealment candidate word based on the answer data. (Appendix 9) In the determining step, generating a question text representing a question regarding the concealment candidate word; The confidentiality determination method according to claim 8, wherein the question text is input to the answering means, and the answer data output from the answering means is obtained. (Appendix 10) an acquisition step of acquiring a target text; an extraction step of extracting candidate words to be concealed from the target text; a determination step of acquiring answer data regarding an answer to a question regarding the concealment candidate word from an answering means configured to generate an answer to the question, and determining whether or not to conceal the concealment candidate word based on the answer data.

[0096] Some or all of the elements (e.g., configurations and functions) described in Supplementary Notes 3 to 7 that are dependent on Supplementary Note 1 may also be dependent on Supplementary Note 8 (method) in the same dependency relationship as Supplementary Note 3 to 7. Some or all of the elements (e.g., configurations and functions) described in Supplementary Note 2 to 7 that are dependent on Supplementary Note 1 may also be dependent on Supplementary Note 10 (program) in the same dependency relationship as Supplementary Note 2 to 7. [Explanation of symbols]

[0097] 10 Target text 20 Concealment candidate words 30 Question Text 40 Response data 50 Answer section 60 Secret Words 1000 computers 1020 Bus 1040 processor 1060 memory 1080 storage device 1100 Input / Output Interface 1120 Network Interface 2000 Confidentiality Determination Device 2020 Acquisition Department 2040 Extraction part 2060 Judgment section

Claims

1. an acquisition unit for acquiring a target text; an extraction means for extracting candidate words to be concealed from the target text; a determination means for obtaining answer data relating to an answer to a question about the concealment candidate word from an answer means configured to generate an answer to the question, and determining whether or not to conceal the concealment candidate word based on the answer data.

2. The determination means generating a question text representing a question regarding the concealment candidate word; The confidentiality determination device according to claim 1 , wherein the answer data output from the answering means is obtained by inputting the question text to the answering means.

3. the answer data indicates an answer text that represents an answer to a question regarding the concealment candidate word; The confidentiality determination device according to claim 1 , wherein the determination means determines that the confidentiality candidate word is to be concealed when a sentence represented by the answer text is unnatural.

4. The determination means a judgment model for judging whether a sentence represented by an input text is unnatural; The confidentiality determination device according to claim 3 , wherein the answer text is input to the determination model to determine whether a sentence represented by the answer text is unnatural.

5. the answer data indicates the accuracy of an answer to a question regarding the concealment candidate word; The confidentiality determination device according to claim 1 , wherein the determination means determines that the confidentiality candidate word is to be confidential when the certainty is equal to or less than a threshold value.

6. The anonymity determination device according to claim 1 , wherein the extraction unit extracts, as the anonymity candidate words, words that can identify an individual or a group.

7. 3. The confidentiality determination device according to claim 1, wherein the answering means comprises a language model trained to output an answer corresponding to an input question.

8. an acquisition step of acquiring a target text; an extraction step of extracting candidate words to be concealed from the target text; a determination step of obtaining answer data regarding an answer to a question regarding the concealment candidate word from an answering means configured to generate an answer to the question, and determining whether or not to conceal the concealment candidate word based on the answer data.

9. In the determining step, generating a question text representing a question regarding the concealment candidate word; The confidentiality determination method according to claim 8 , wherein the answer data output from the answering means is obtained by inputting the question text to the answering means.

10. an acquisition step of acquiring a target text; an extraction step of extracting candidate words to be concealed from the target text; a determination step of obtaining answer data regarding an answer to a question regarding the concealment candidate word from an answering means configured to generate an answer to the question, and determining whether or not to conceal the concealment candidate word based on the answer data.

Images