Information processing method and computer program

Abstract

To provide a technology capable of effectively reducing the leakage of confidential information contained in text and appropriately disclosing the information to a reader.SOLUTION: A confidential information masking unit 31 acquires a processing target text Y that includes a target concept to be concealed. The confidential information masking unit 31 acquires a reader attribute R indicating the attribute of a reader of the processing target text Y. The confidential information masking unit 31 abstracts the target concepts contained in the processing target text Y according to the reader attributes R, using a concept information tree T that defines hierarchical relations between concepts.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The subject matter disclosed herein relates to information processing methods and computer programs. [Background technology]

[0002] BACKGROUND ART Conventionally, a system is known that outputs an answer to a question input by a user using a machine learning model that uses a collection of document data as training data (for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2024-049674 Summary of the Invention [Problem to be solved by the invention]

[0004] When confidential information is included in training data, the machine learning model may output text that includes the confidential information. In this case, there is a risk that confidential information may be leaked if a reader who should not disclose the confidential information views the text. On the other hand, if the confidential information is completely masked, there is a risk that the reader will not be able to effectively understand the information contained in the text.

[0005] An object of the present invention is to provide a technique that can appropriately disclose information to a reader while effectively reducing the leakage of confidential information contained in text. [Means for solving the problem]

[0006] In order to solve the above problem, a first aspect is an information processing method executed by a computer, comprising: a) acquiring a target text to be processed that includes a target concept to be concealed; b) acquiring reader attributes that indicate the attributes of a reader of the target text to be processed; and c) abstracting the target concept included in the target text to be processed according to the reader attributes using concept hierarchy information that defines the hierarchical relationships between multiple concepts.

[0007] A second aspect is an information processing method of the first aspect, wherein the concept hierarchy information defines a concept hierarchy for each concept, the reader attributes are information indicating a specific concept hierarchy, and step c) includes a step of abstracting the target concept of the text to be processed if the specific concept hierarchy indicated by the reader attributes is higher than the concept hierarchy of the target concept.

[0008] A third aspect is the information processing method of the second aspect, wherein step c) includes a step of replacing the target concept of the text to be processed with a concept of the specific conceptual hierarchy when the conceptual hierarchy of the target concept is lower than the specific conceptual hierarchy indicated by the reader attributes.

[0009] A fourth aspect is an information processing method according to any one of the first to third aspects, wherein the concept hierarchy information defines a disclosure range for each concept, the reader attributes are information indicating whether or not they are included in the disclosure range, and step c) includes a step of abstracting the target concept of the text to be processed if the reader attributes are not included in the disclosure range of the target concept.

[0010] A fifth aspect is an information processing method of the fourth aspect, wherein step c) includes a step of replacing the target concept of the text to be processed with a concept that is higher in level than the target concept and is in a public range that includes the reader attribute, if the reader attribute is not included in the public range of the target concept.

[0011] A sixth aspect is the information processing method of the fifth aspect, wherein the higher the relationship between concepts in the concept system information, the wider the range of disclosure.

[0012] A seventh aspect is a computer program executable by a computer, causing the computer to execute the information processing method of any one of the first to sixth aspects. [Effects of the Invention]

[0013] According to the first to seventh aspects, the text to be processed is abstracted according to the reader's attributes. This makes it possible to restrict the information disclosed to each reader according to the reader. Therefore, it is possible to appropriately disclose information to each reader while effectively reducing the leakage of confidential information.

[0014] According to the information processing method of the second aspect, the text to be processed can be abstracted based on the hierarchical attribute.

[0015] According to the information processing method of the fourth aspect, the text to be processed can be abstracted based on the disclosure range. [Brief explanation of the drawings]

[0016] [Figure 1] 1 is a diagram illustrating a configuration of an information processing device according to a first embodiment. [Figure 2] 1 is a block diagram illustrating an outline of information processing executed by an information processing device according to a first embodiment. [Figure 3] FIG. 2 is a diagram showing a concept information tree which is concept system information in the first embodiment. [Figure 4] FIG. 10 is a diagram conceptually illustrating how an instance having a reader attribute is generated. [Figure 5] 10 is a diagram for explaining the correspondence between reader attributes and hierarchical attributes in a conceptual information tree. FIG. [Figure 6] FIG. 10 is a diagram showing a flow of processing a target text together with a specific example. [Figure 7] FIG. 10 is a diagram showing a concept information tree Ta, which is concept system information in the second embodiment. [Figure 8] FIG. 10 is a diagram conceptually illustrating how an instance having a reader attribute R is generated. [Figure 9] FIG. 8 is a diagram showing concepts that can be viewed by a reader with reader attribute R-5 in the concept information tree Ta shown in FIG. 7. [Figure 10] FIG. 10 is a diagram showing a viewable range table. [Figure 11] FIG. 11 is a diagram showing a concept information tree Tb which is concept system information in the third embodiment. [Figure 12] This is a diagram showing the concept of what can be viewed by readers to whom reader attribute D-3 is set. [Figure 13] This is a diagram showing the concept that readers who have reader attribute R-2 set corresponding to the disclosure range can view the content. DETAILED DESCRIPTION OF THE INVENTION

[0017] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings. Note that the components described in the embodiment are merely examples and are not intended to limit the scope of the present invention. In the drawings, the dimensions and numbers of each part may be exaggerated or simplified as necessary to facilitate understanding.

[0018] 1. First embodiment FIG. 1 is a diagram showing the configuration of an information processing device 1 according to the first embodiment. The information processing device 1 is a computer and includes a processor 11 and a memory 13. The processor 11 includes, for example, a CPU (Central Processing Unit). The memory 13 includes, for example, a ROM (Read-Only Memory) or a RAM (Random-Access Memory). The memory 13 may also include an auxiliary storage device such as a hard disk drive (HDD) or a solid-state drive (SSD). The memory 13 is connected to the processor 11 via a system bus.

[0019] The memory 13 stores a computer program P. The computer program P is executable by the processor 11 of the information processing device 1. When the processor 11 executes the computer program P, information processing, described below, is performed in the information processing device 1. The computer program P may be recorded on a non-transitory recording medium. The recording medium is, for example, optical media or a semiconductor memory such as a USB memory. The computer program P recorded on the recording medium can be read by a reading device (not shown). The computer program P may be stored in the memory 13 via a network line (not shown).

[0020] The information processing device 1 has a display 15 and an input device 17. The display 15 and the input device 17 are connected to the processor via a system bus. The display 15 is a device that visually displays the output of the information processing device 1, and is specifically a liquid crystal display. The input device 17 is a device that allows a user to input data and commands to the information processing device 1, and is specifically a keyboard or a mouse. Note that the display 15 may function as the input device 17 by being equipped with, for example, a touch panel.

[0021] 2 is a block diagram schematically illustrating information processing executed by the information processing device 1 of the first embodiment. The confidential information masking unit 31 shown in FIG. 2 is a function realized by the processor 11 executing a computer program P. The confidential information masking unit 31 abstracts confidential information included in processing target text Y, thereby outputting abstracted text A in which the confidential information is masked. The processing target text Y is a text such as an email message entered by a user via an input device, a text stored in the memory 13 of the information processing device 1, or a text obtained from a database.

[0022] The processing target text Y may be a sentence generated by a sentence generation model. Specifically, the sentence generation model is a large language model (LLM). The LLM is, for example, a deep neural network based on a self-attention mechanism called a Transformer. Transfer is capable of capturing the relationship of the entire input sequence by using the self-attention mechanism.

[0023] The confidential information masking unit 31 uses the concept information tree T to abstract confidential information (target concepts) to be concealed that is included in the processing target text Y, according to the pre-acquired reader attribute R. The concept information tree T is concept system information that includes multiple concepts and defines the hierarchical relationships between the multiple concepts.

[0024] FIG. 3 is a diagram showing a concept information tree T, which is the concept system information of the first embodiment. In this concept information tree T, each concept is assigned one concept number (identifier) ​​and one corresponding concept name (a word that is a specific representation of the concept). For example, the concept name of a concept with concept number C-1 is "medicinal solution." In addition, in the concept information tree T, the hierarchical relationships between concepts are described in a tree structure. For example, in FIG. 3, "medicinal solution" is the top-level concept (root concept), and "acidic chemical solution" (C-2) and "basic chemical solution" (C-3) are defined as its immediate subordinate concepts. Furthermore, "acidic chemical solution A" (C-4) and "acidic chemical solution B" (C-5) are defined as its immediate subordinate concepts. Note that "acidic chemical solution A" and the like are notation examples for convenience of explanation, and in reality, specific chemical solution names may be assigned.

[0025] Furthermore, in the concept information tree T of this embodiment, a hierarchical attribute D indicating the concept hierarchy is defined for each concept. The hierarchical attribute D is information indicating the depth from a reference concept (here, the top-level root concept). In the example shown in Fig. 3, the hierarchical attribute of the root concept "medicinal solution" is "D-1", and the number of the hierarchical attribute increases by one each time the conceptual hierarchy becomes deeper, such as "D-2", "D-3", and so on.

[0026] 3 has only one root concept, it may have multiple types of root concepts. The concept information tree T may have a tree structure for each root concept.

[0027] The conceptual information tree T is prepared in advance by a user or the like and stored in the memory 13 together with the computer program P. The conceptual information tree T may be stored on a server or the like separate from the information processing device 1. In this case, a system may be constructed in which the information processing device 1 accesses the conceptual information tree T via a network line such as the Internet.

[0028] FIG. 4 is a conceptual diagram illustrating how an instance having a reader attribute R is generated. In this embodiment, in order to prevent leakage of confidential information contained in the processing target text Y, a reader attribute R corresponding to a reader is defined in advance. The reader attribute R is information indicating the abstraction level (authority) of confidential information that can be viewed. In the example shown in FIG. 4, reader attributes R-1, R-2, etc. are defined for each reader hierarchy. When a reader hierarchy is specified by the user, the information processing device 1 generates an instance having a reader attribute R corresponding to the reader hierarchy. Note that it is not essential that the reader hierarchy be set based on a specified input by the user. For example, when the confidential information masking unit 31 processes the body of an email, the reader hierarchy may be set based on the email destination.

[0029] 5 is a diagram for explaining the correspondence between a reader attribute R and a hierarchical attribute D in a conceptual information tree T. The reader attribute R in this embodiment is information indicating a specific conceptual hierarchy. That is, the reader attribute R corresponds to a specific hierarchical attribute D. More specifically, the reader attributes R-1, R-2, R-3, etc. correspond to the hierarchical attributes D-1, D-2, D-3, etc., respectively.

[0030] For example, the hierarchical attribute corresponding to hierarchical attribute R-3 is D-3. Therefore, as shown in Figure 5, a reader with reader attribute R-3 is permitted to view the hierarchical attribute D-3 corresponding to reader attribute R-3 and the concepts of hierarchical attributes D-1 and D-2 above it (concepts shown in solid-line boxes in Figure 4). In addition, they are prohibited from viewing the concepts of hierarchical attributes D-4... below hierarchical attribute D-3 corresponding to reader attribute R-3 (concepts shown in dashed-line boxes in Figure 4).

[0031] If the processing target text Y contains a word with a concept of hierarchical attribute D that is prohibited from being viewed (i.e., a hierarchical attribute lower than the hierarchical attribute indicated by the reader attribute R), the confidential information masking unit 31 abstracts the word into a concept of hierarchical attribute D that is permitted to be viewed. For example, if the reader attribute is D-3 and the processing target text Y contains a word with a concept of hierarchical attribute D-4, the confidential information masking unit 31 replaces the word with a word with a concept of hierarchical attribute D-3 that corresponds to the reader attribute R-3. Note that the confidential information masking unit 31 may also replace the word with a concept of a hierarchical attribute D (i.e., D-2, D-1) that is higher than the hierarchical attribute D-3 that corresponds to the reader attribute R-3.

[0032] 6 is a diagram showing a flow of processing the target text Y, along with a specific example. The confidential information masking unit 31 first analyzes the target text Y, performs word division, and acquires the reader attribute R of the instance. In the example shown in FIG. 6, the target text Y is a sentence that reads, "After dripping the acidic chemical solution A, use the basic chemical solution C1." and the reader attribute R is "R-3."

[0033] The confidential information masking unit 31 identifies words contained in the analyzed processing target text Y that are not viewable by the reader by comparing the words in the analyzed processing target text Y with the conceptual information tree T. The confidential information masking unit 31 then generates abstracted text A in which those words have been replaced with viewable words. In this example, the processing target text Y contains "acidic chemical solution A" of the hierarchical attribute D-3 and "basic chemical solution C1" of the hierarchical attribute D-4, so "basic chemical solution C1" is identified as an unviewable word. Then, abstracted text A is generated in which "basic chemical solution C1" has been replaced with "basic chemical solution C," which is a viewable word.

[0034] The abstract text A is displayed on the display 15, allowing the reader to view the abstract text A. Confidential information contained in the processing target text Y (i.e., words registered in the conceptual information tree T) is abstracted according to the reader's reader attribute R. This allows the information disclosed to the reader to be limited depending on the reader, and allows the reader to grasp the meaning of the abstract text A, compared to when confidential information is completely masked. Therefore, even if the processing target text Y contains confidential information, the information can be appropriately disclosed to the reader while reducing the leakage of confidential information.

[0035] 2. Second Embodiment Next, a second embodiment will be described. In the following description, elements having the same functions as elements already described will be given the same reference numerals or reference numerals with an additional alphabetical character, and detailed description thereof may be omitted.

[0036] 7 is a diagram showing a concept information tree Ta, which is the concept system information of the second embodiment. In the concept information tree Ta, each concept has a concept number and a concept name, and a disclosure range is defined. The disclosure range is information indicating the range of reader attributes that are permitted to disclose the concept. In other words, the reader information is information indicating whether or not the reader is included in the disclosure range.

[0037] For example, the disclosure range of "acidic chemical solution" (C-2) is "R-2, + lower." This means that the disclosure range includes the reader attribute R-2 and all reader attributes included in the disclosure range of concepts lower than "acidic chemical solution" (specifically, "acidic chemical solution A," "acidic chemical solution B," "acidic chemical solution A1," "basic chemical solution A2," and "acidic chemical solution A2"). In the concept information tree Ta, the higher the relationship between concepts, the wider the disclosure range. For example, the disclosure range of "acidic chemical solution" (C-2) is wider than the disclosure range of "acidic chemical solution A" (C-4), which is a lower concept of that concept, and the disclosure range of "acidic chemical solution B" (C-5).

[0038] FIG. 8 is a conceptual diagram showing how an instance having a reader attribute R is generated. In this embodiment, the reader attribute R is defined in advance for each reader. The reader attribute for each reader is managed, for example, in a database. The information processing device 1 generates an instance having the reader attribute R corresponding to a reader specified by a user or the like. Note that, in the second embodiment as well, an instance of the reader attribute R corresponding to the reader hierarchy may be generated, as described in FIG. 4 of the first embodiment.

[0039] Fig. 9 is a diagram showing concepts that can be viewed by a reader with a reader attribute R-5 in the concept information tree Ta shown in Fig. 7. When the reader attribute R of an instance is R-5, viewing is permitted for concepts whose disclosure range includes R-5 (concepts shown in solid line boxes in Fig. 9). Also, viewing is prohibited for concepts whose disclosure range does not include R-5 (concepts shown in dashed line boxes in Fig. 9).

[0040] In this embodiment, the confidential information masking unit 31 also analyzes the processing target text Y, performs word segmentation, and acquires the reader attribute R of the instance. Then, the confidential information masking unit 31 identifies words contained in the processing target text Y that are not accessible to the reader by referencing the analyzed words in the concept information tree Ta. Specifically, words of concepts registered in the concept information tree Ta, whose reader attribute R of the instance is not included in the disclosure range, are identified as unaccessible words. Then, the confidential information masking unit 31 replaces the identified word (target concept) with a word of a concept that is higher in level than the identified word and whose reader attribute R is included in the disclosure range.

[0041] Here, assume that the reader attribute R of the instance is "R-5" and the processing target text Y includes "basic chemical liquid C1" (C-13). In this case, the disclosure range of "basic chemical liquid C1" is only R-8, and does not include R-5. Therefore, the confidential information masking unit 31 replaces the word of the concept in the processing target text Y with the superordinate concept "basic chemical liquid" (C-3), which includes R-5 in its disclosure range.

[0042] In this embodiment, the disclosure range is set for each concept (word), so the abstraction level of confidential information for readers can be set more finely than in the first embodiment, where a hierarchical attribute D is defined according to the depth of the concept.

[0043] FIG. 10 is a diagram showing a viewable range table 41. In the viewable range table 41, row headers (row labels) indicate concept numbers, and column headers (column labels) indicate reader attribute R. A circle (○) in the table indicates that the concept number in the row header is viewable for the reader attribute R of the column header. The concept numbers written in the table indicate replacement destinations for the concept numbers in row headers that are not viewable for the reader attribute R of the column header. For example, if the reader attribute R is R-5, the replacement destination for C-13 (basic chemical solution C1) is C-3 (basic chemical solution). A viewable range table 41 that defines such replacement destinations may be prepared and stored in memory 13 in advance. The confidential information mask unit 31 then refers to the viewable range table 41 to perform a replacement process for the not viewable concepts, thereby improving processing speed.

[0044] 3. Third Embodiment FIG. 11 is a diagram showing a concept information tree Tb, which is conceptual system information of the third embodiment. In this concept information tree Tb, a hierarchical attribute D is defined for each concept, as in the concept information tree T of the first embodiment. Furthermore, for some concepts, the disclosure range described in the second embodiment is defined. Here, the disclosure range is defined for concepts deep in the hierarchy (specifically, concepts below the hierarchical concept D4). For example, for "basic chemical solution C1" (C-13) of the hierarchical attribute D-4, R-4 is defined as the disclosure range. In this embodiment, at least one of a reader attribute indicating the hierarchical attribute D or a reader attribute R indicating whether or not the instance is included in the disclosure range is set as the reader attribute R of the instance.

[0045] Figure 12 is a diagram showing the concepts that can be viewed by a reader to whom reader attribute D-3 is set. Reader attribute D-3 is an attribute corresponding to hierarchical attribute D-3. As shown in Figure 12, a reader with reader attribute D-3 is permitted to view concepts at or above hierarchical attribute D-3.

[0046] Figure 13 is a diagram showing the concepts that can be viewed by a reader for whom reader attribute R-2 corresponding to the disclosure range is set. As shown in Figure 13, when reader attribute R-2 corresponding to the disclosure range is set, the reader is permitted to view concepts whose disclosure range includes R-2 ("acidic chemical solution A1" and "basic chemical solution A1"). Furthermore, concepts whose disclosure range includes R-2 appear at hierarchical attribute D-4 and below. Therefore, the reader is permitted to view concepts whose disclosure range includes hierarchical attribute D-3 and above.

[0047] In this way, by defining the hierarchical attribute D and the disclosure range in the concept information tree Tb, it is possible to increase the variety of substitutions. Also, by not specifying the disclosure range for concepts whose conceptual hierarchy is higher than a predetermined standard, and specifying the disclosure range only for concepts whose conceptual hierarchy is lower than a predetermined standard, it is possible to simplify the viewable range table 41.

[0048] Although the present invention has been described in detail, the above description is merely illustrative in all respects and does not limit the present invention. It is understood that countless variations not illustrated can be envisioned without departing from the scope of the present invention. The configurations described in the above embodiments and variations can be combined or omitted as appropriate as long as they are not mutually inconsistent. [Explanation of symbols]

[0049] 1. Information processing equipment (computer) Y Processing target text T,Ta,Tb Concept information tree (concept system information)

Claims

1. 1. A computer-implemented information processing method, comprising: a) obtaining a text to be processed that includes a target concept to be anonymized; b) acquiring a reader attribute indicating an attribute of a reader of the text to be processed; c) abstracting the target concepts included in the processing target text in accordance with the reader attributes using concept hierarchy information that defines the hierarchical relationships between a plurality of concepts; An information processing method, including:

2. 2. The information processing method according to claim 1, the concept hierarchy information defines a concept hierarchy for each concept, The reader attribute is information indicating a specific conceptual hierarchy, The information processing method includes a step of abstracting the target concept of the text to be processed when the specific conceptual hierarchy indicated by the reader attribute is higher than the conceptual hierarchy of the target concept.

3. 3. The information processing method according to claim 2, The information processing method includes a step of replacing the target concept in the processing target text with a concept in the specific conceptual hierarchy when the conceptual hierarchy of the target concept is lower than the specific conceptual hierarchy indicated by the reader attributes.

4. 3. The information processing method according to claim 1, further comprising: the concept system information defines a disclosure range for each concept, The reader attribute is information indicating whether or not the reader is included in the disclosure range, The information processing method, wherein step c) includes a step of abstracting the target concept of the processing target text when the reader attribute is not included in the disclosure range of the target concept.

5. 5. The information processing method according to claim 4, The information processing method includes a step of replacing the target concept of the text to be processed with a concept that is higher in level than the target concept and is in a public range that includes the reader attribute, if the reader attribute is not included in the public range of the target concept.

6. 6. The information processing method according to claim 5, In the concept system information, the higher the relationship between concepts, the wider the scope of disclosure.

7. A computer-executable computer program, A computer program causing the computer to execute the information processing method according to claim 1 or 2.

Images