JP2011221894A - Secure document detection method, secure document detection program, and optical character reader - Google Patents

Abstract

PROBLEM TO BE SOLVED: To highly accurately detect a secure document by a simple definition.SOLUTION: A secure document detection method is executed by a secure document detector, and the secure document detector includes an arithmetic unit and a storage device for holding a dictionary. In the dictionary, a plurality of keyword pairs each including at least two keywords and information indicating the position relation in a document of the two keywords included in each keyword pair are registered. The secure document detection method includes a first procedure of extracting the keyword pair registered in the dictionary from inputted document data, and a second procedure for determining whether or not the inputted document data is the secure document on the basis of the position relation in the inputted document data of the two keywords included in the extracted keyword pair.

Description

  The present invention relates to a technique for managing information security, and more particularly to a technique for detecting a secure document from a document stored in a storage device or a printed document.

  With increasing social interest in information security, it is possible to automatically detect whether or not secure information exists in a large amount of electronic documents stored in a server or a personal computer (PC) owned by an individual with high accuracy. Technology is required. Here, the secure information is information that needs to be kept confidential, such as private information of the company, confidential information of other companies, or personal information. For example, Patent Document 1 is disclosed as such an automatic detection technique.

  The confidential document detection system described in Patent Literature 1 divides an input document into a plurality of areas, detects feature elements in each area by referring to a dictionary corresponding to each area, and detects the detected feature elements Based on, the confidential information category to which each document belongs is determined.

JP 2006-209649 A

  When a conventional text search such as a personal PC check tool is used, there are many false detections of secure documents, which requires a lot of labor for human inspection. In the conventional secure document detection, a user can specify a keyword to be detected, but it is difficult to deal with documents in various formats.

  For example, the confidential document detection system described in Patent Document 1 detects a feature element for each area, but cannot detect a feature element that appears in an undefined area. Further, the confidential document detection system determines the confidential information category based on the correspondence between the keyword and the area in which the keyword is detected, but cannot determine the confidential information category based on the relationship between a plurality of keywords.

  A typical example of the present invention is as follows. That is, a secure document detection method executed by a secure document detection device, wherein the secure document detection device includes an arithmetic device and a storage device that holds a dictionary, each of which includes at least two keywords. And information indicating the positional relationship of the two keywords included in each keyword pair in the document are registered, and the secure document detection method is registered in the dictionary from input document data. Whether the input document data is a secure document based on the first procedure for extracting the keyword pair and the positional relationship of the two keywords included in the extracted keyword pair in the input document data And a second procedure for determining whether or not.

  According to an embodiment of the present invention, a secure document can be detected with high accuracy with a simple definition even when the types of input documents increase.

It is a block diagram which shows the outline | summary of embodiment of this invention. It is a block diagram which shows the hardware constitutions of the secure electronic document management system of the 1st Embodiment of this invention. It is explanatory drawing which shows the whole process which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is a flowchart explaining the detailed procedure of the document element extraction process and secure document determination process of the 1st Embodiment of this invention. It is explanatory drawing of the specific example of the secure document input into the secure document detection apparatus of the 1st Embodiment of this invention. It is explanatory drawing of the secure document detection based on the combination of the several keyword which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is explanatory drawing of the block identified by the secure document detection apparatus of the 1st Embodiment of this invention. It is explanatory drawing which shows the 1st specific example of the keyword extraction and the secure document determination which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is explanatory drawing which shows the 2nd specific example of the keyword extraction and secure document determination which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is explanatory drawing which shows the 3rd specific example of the keyword extraction and the secure document determination which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is explanatory drawing which shows the 4th specific example of the keyword extraction and the secure document determination which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is explanatory drawing which shows the 5th specific example of the keyword extraction and secure document determination which the secure document detection apparatus of the 1st Embodiment of this invention performs. It is explanatory drawing of the arrangement | positioning cost table contained in the secure document dictionary of the 1st Embodiment of this invention. It is explanatory drawing of the secure document dictionary of the 1st Embodiment of this invention. It is a block diagram which shows the hardware constitutions of the OCR integrated secure document detection apparatus of the 2nd Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing an outline of an embodiment of the present invention.

  First, conventional secure paper document management will be described.

  The optical character reading device (OCR device) 0302 reads the input paper document 0301 and creates a document file 0303. This document file 0303 includes image, text, or both data described in the paper document 0301. This document file may be, for example, a PDF (Portable Document Format) (registered trademark) file. The user refers to the document file 0303 displayed by the computer 0304, determines whether or not the document file 0303 is secure, and inputs the result to the computer 0304. When it is determined that the document file 0303 is secure, the user may input an instruction to lock the document file 0303 to the computer 0304. The computer 0304 creates a locked document file 0305 by locking the document file 0303, and outputs this document file 0305. The lock means a process for restricting browsing of the document file 0305, and a typical example is encryption.

  Next, OCR integrated secure paper document management, which is one embodiment of the present invention, will be described.

  The input paper document 0306 may be the same as the paper document 0301 already described. The OCR device 0307 reads the input paper document 0306 and extracts image information and text information included in the paper document 0306. The computer 0308 determines whether or not the extracted information includes secure information. When it is determined that secure information is included in the extracted information, the computer 0308 creates and outputs a locked document file 0309 including the extracted information. Since these determinations and creation are automatically performed without user intervention, the computer 0308 does not need to display the document file before being locked. Therefore, neither the OCR device 0307 nor the computer 0308 needs to create an unlocked document file before the above determination. Further, the user does not need to input an instruction to lock the document file to the computer 0308.

  Note that that the created document file is secure means that the original paper document 0306 is also secure. For this reason, the computer 0308 may control the discharge method of the paper document 0306 by the OCR device 0307 based on the determination result of whether or not the document file is secure.

  Details of the OCR-integrated secure paper document management will be described later as a second embodiment of the present invention.

  Next, secure electronic document management, which is another embodiment of the present invention, will be described.

  When a document file 0310 is input, the computer 0311 determines whether the document file 0310 is secure. The document file 0310 may be the same as the document file 0303, for example. When it is determined that the document file 0310 is secure, the computer 0311 creates and outputs a locked document file 0312 by locking it. As in the case of the OCR-integrated secure paper document management described above, these determinations and creations are automatically executed without user intervention.

  Details of the secure electronic document management will be described later as the first embodiment of the present invention.

  In the above description, a PDF file is exemplified as an example of a document file including data such as text. However, these document files may be a document file of a format other than PDF or a drawing file.

<First Embodiment>
FIG. 2 is a block diagram illustrating a hardware configuration of the secure document detection device 0100 according to the first embodiment of this invention.

  The secure document detection apparatus 0100 is an example of an apparatus that realizes the secure electronic document management of the present invention shown in FIG.

  The secure document detection device 0100 of this embodiment includes an operation terminal device 0101, a display terminal device 0102, an external storage device 0103, a memory 0104, a central processing unit 0105, a communication device 0107, and a communication line 0106 that interconnects them. The secure document detection device 0100 may be a general personal computer, for example.

  The operation terminal device 0101 is, for example, a keyboard or a mouse, and is used by the user to input an instruction or data to the secure document detection device 0100.

  The display terminal device 0102 is a device that displays text, images, and the like, such as a liquid crystal display device.

  The external storage device 0103 is a storage device such as a hard disk device or a flash memory, and stores input document data (for example, a document file 0310) and output document data (for example, a locked document file 0312). Furthermore, a program executed by the central processing unit 0105 to realize this embodiment may be stored.

  The memory 0104 is, for example, a semiconductor memory, and stores a program executed by the central processing unit 0105, data to be referred to, and the like. At least a part of programs, data, and the like stored in the external storage device 0103 may be copied to the memory 0104 as necessary.

  The central processing unit 0105 executes a program stored in the memory 0104 and controls the operation terminal device 0101, the display terminal device 0102, the external storage device 0103, and the communication device 0107 as necessary. In the following description, the processing executed by the secure document detection device 0100 is actually executed by the central processing unit 0105.

  The communication device 0107 is an interface that is connected to a network (not shown) and communicates with other devices (not shown) connected to the network. For example, the communication device 0107 may receive the document file 0310 as input data and transmit the locked document file 0312.

  FIG. 3 is an explanatory diagram illustrating the entire processing executed by the secure document detection device 0100 according to the first embodiment of this invention.

  The processing executed by the secure document detection device 0100 is divided into a learning phase 0500 and a usage phase 0510.

  In the learning phase 0500, the secure document detection device 0100 creates a secure document dictionary 0504 based on the input information.

  Specifically, for example, the user inputs the secure document example 0501 and the secure term definition 0502 to the secure document detection apparatus 0100.

  The secure document example 0501 is an actual document file that the user thinks should be detected as a secure document. The secure document example 0501 may be input to the secure document detection device 0100 via the communication device 0107, for example.

  The secure term definition 0502 is a list of keywords used for secure document detection. The user can input a keyword list including character strings included in a document to be detected as a secure document to the secure document detection apparatus 0100 as the secure term definition 0502. In particular, for example, a document including a combination of a character string indicating the creator or owner of a document such as “XX Manufacturing” and a character string indicating a document type such as “design document” is detected as a secure document. When necessary, the user can input a combination of keywords composed of such character strings (hereinafter also referred to as keyword pairs) to the secure document detection apparatus 0100 as the secure term definition 0502. The secure term definition 0502 may be input via the communication device 0107, for example, or may be input via the operation terminal device 0101.

  The secure document detection apparatus 0100 executes a secure dictionary learning process 0503 based on the input secure document example 0501 and the secure term definition 0502. As a result, a secure document dictionary 0504 is created. As will be described later (see FIG. 10), the secure document dictionary 0504 includes information indicating a combination of character strings registered as keywords, a positional relationship between two keywords included in each keyword pair, and the like. . Note that the information indicating the positional relationship between the two keywords is, for example, a vector representing the direction and distance in which the keywords are arranged. Hereinafter, the positional relationship between two keywords included in such a keyword pair is also referred to as “the positional relationship between keyword pairs”.

  Next, the usage phase 0510 will be described. The user inputs the unmanaged document 0511 to the secure document detection apparatus 0100. The unmanaged document 0511 is a document that the user intends to manage from now on. In other words, it is a document that needs to be determined whether or not it includes secure information. In accordance with the determination result, the document management method (for example, whether to lock the document or the like) is determined. The unmanaged document 0511 corresponds to, for example, the document file 0310 in FIG.

  The secure document detection device 0100 executes document element extraction processing 0512 for the input non-management document 0511. As a result, document elements, that is, text, keyword (KW), ruled line, information indicating the position of the keyword, information indicating the arrangement of blocks, and the like are extracted from the unmanaged document 0511. In order to extract a keyword and its position, the keyword information 0513 included in the secure document dictionary 0504 is referred to.

  If the file format of the input unmanaged document 0511 and the file format that can be processed by the document element extraction process 0512 are different, the secure document detection apparatus 0100 executes the document conversion process 0517 to input the input unmanaged document 0511. Convert the file format. For example, when the document element extraction process 0512 can process only a PDF file, but a file in another format (for example, a document file created by general document creation software) is input as the unmanaged document 0511, A document conversion process 0517 converts the file format of the unmanaged document 0511 to PDF.

  Next, the secure document detection device 0100 executes secure document determination processing 0515 for the document information 0518 extracted by the document element extraction processing 0512. Specifically, the secure document detection apparatus 0100 refers to the document information 0518 and the pattern information / placement likelihood 0514 included in the secure document dictionary 0504, and the secure information likelihood of the input unmanaged document 0511. And whether or not the non-managed document 0511 is a secure document (that is, whether or not it includes secure information) is determined based on this.

  Then, the secure document detection device 0100 outputs a result 0516 of the secure document determination process 0515. This result includes information indicating whether or not the unmanaged document 0511 is a secure document, and may further include information indicating a secure likelihood or a risk based on the secure likelihood.

  If the secure document dictionary 0504 is held in advance, the secure document detection device 0100 can execute only the use phase 0510 without executing the learning phase 0500. For example, the user may acquire the secure document dictionary 0504 created by the manufacturer of the secure document detection device 0100, or may acquire the secure document dictionary 0504 created by another user executing the learning phase 0500. Good.

  FIG. 4 is a flowchart illustrating detailed procedures of the document element extraction process 0512 and the secure document determination process 0515 according to the first embodiment of this invention.

  The secure document detection device 0100 extracts document elements from the input electronic document file 0411 (step 0401). Specifically, the secure document detection device 0100 extracts from the electronic document file 0411 character information of text included in the electronic document, a position on the paper where the character is written, a ruled line position, and the like. Thereby, each character is extracted, and further, a character string corresponding to each line is specified from the position of each character and the position of the ruled line. The electronic document file 0411 corresponds to the unmanaged document 0511 in FIG.

  Next, the secure document detection device 0100 analyzes the document structure using the extracted document element (step 0402). Specifically, the secure document detection device 0100 divides characters on the document into blocks based on the extracted characters and the positions of ruled lines. For example, when a document includes a header, a footer, and a body, the header, footer, and body are each identified as one block. If the text is in columns, each column is identified as one block. If the document contains a table, the table is identified as a block. The secure document detection device 0100 may refer to a document structure dictionary (not shown) in step 0402. As a result, the block to which each row extracted in step 0401 belongs is specified. Such document structure analysis can be performed by a known method. For example, there are methods such as an XY recursive analysis method and a character string moving distance minimum method.

  Next, the secure document detection device 0100 rearranges the blocks and lines so as to match the reading order of the text (in other words, the order in which they appear in the document) (step 0403). Thereby, the lines in each block are rearranged in the text reading order, and the blocks are also rearranged in the text reading order. For example, when the body is composed of a plurality of blocks, these blocks are rearranged in the reading order of the text. Similar to step 0402, this rearrangement can also be performed by a known method.

  Next, the secure document detection device 0100 extracts document elements (step 0404). Specifically, the secure document detection device 0100 extracts a ruled line and a layout formed by the ruled line and the like. Further, the secure document detection device 0100 extracts a keyword from the character string extracted in step 0401. Specifically, the secure document detection device 0100 searches for the character string extracted in step 0401 using the keyword registered in the keyword information 0412 of the secure document dictionary as a search key. The keyword information 0412 corresponds to the keyword information 0513 in FIG.

  Next, the secure document detection device 0100 calculates the secure information likelihood using the pattern information 0413 included in the secure document dictionary (step 0405). The secure information likelihood is an index indicating the secure document quality of the input document (details will be described later). The pattern information 0413 corresponds to a part of the pattern information / placement likelihood 0514 in FIG.

  Next, the secure document detection device 0100 calculates secure information likelihood using the placement likelihood information 0414 included in the secure document dictionary (step 0406). The placement likelihood information 0414 corresponds to a part of the pattern information / placement likelihood 0514 of FIG.

  The calculation of the secure information likelihood will be described later (see FIGS. 8 to 9 and equations (1) to (3)).

  The secure document detection device 0100 determines whether or not the input electronic document is a secure document based on the secure information likelihood calculated in steps 0405 and 0406 (step 0407). For example, the secure document detection device 0100 may determine that the input electronic document is a secure document when the calculated secure information likelihood is greater than a predetermined threshold. The user may set this threshold value.

  If it is determined in step 0407 that the input electronic document is a secure document (ie, “Yes”), the secure document detection device 0100 locks the input electronic document file 0411 (step 0408). On the other hand, when it is determined that the input electronic document is not a secure document (that is, “No”), the secure document detection device 0100 does not execute Step 0408.

  Next, the secure document detection device 0100 outputs an electronic document (step 0409). Specifically, the secure document detection device 0100 outputs a locked electronic document if “Yes” in step 0407, and if it is “No”, the secure document detection device 0100 does not lock the electronic document (that is, the input electronic document file). 0411 itself) is output. The output electronic document 0415 (corresponding to the document file 0312 in FIG. 1) is stored in the external storage device 0103. Further, the secure document detection device 0100 may output the secure information likelihood itself, or may output the risk (or required protection level) determined based on the secure information likelihood.

  In the above, an example in which it is determined in step 0407 whether or not the document is a secure document has been described. However, in step 0407, multivalue determination may be performed instead of such binary determination. For example, the secure document detection device 0100 may determine the rank of the secure information likelihood by comparing the calculated secure information likelihood with a plurality of threshold values. In this case, an electronic document output method (for example, the strength of encryption used) may be selected according to the determined rank. For example, the secure document detection device 0100 may use a longer encryption key to encrypt the higher-ranked electronic document file 0411.

  Hereinafter, details of the processing of FIG. 4 will be described.

  FIG. 5 is an explanatory diagram of a specific example of a secure document input to the secure document detection apparatus 0100 according to the first embodiment of this invention.

  The present invention can be applied to secure documents in which access needs to be restricted. A typical example of such a secure document is a document including the confidential information of the company created by the company, a document including confidential information of the other company acquired from another company, or a document including personal information of a customer or the like. Such a typical example will be described.

  FIG. 5A to FIG. 5C are examples of a secure document in which the document title and the name of a specific company are displayed. For example, the title of the cover page of a document includes a specific character string such as “design document”, “specification document”, or “alliance”, and the cover page (for example, as the creator of the document or the distribution destination of the document) ) The specific company name “XX” or “XX Manufacturing” is included. Note that the underline 0601 displayed in FIG. 5 indicates the specific character string, company name, and the like displayed in each document to explain the present embodiment, and the underline 0601 itself is It is not displayed in the document.

  FIG. 5D to FIG. 5F are examples of secure documents that include a specific character string (for example, a company name) in the header and the like, and further include a specific prefix or suffix next to the character string. . In the example of FIG. 5D, a specific suffix “created” is displayed next to a specific character string “(stock) xx”. In the example of FIG. 5 (e), a specific suffix “confidental” is displayed next to the specific character string “XX”. In the example of FIG. 5 (e), a specific suffix “Prepeard” is displayed next to the specific character string “XX”.

  FIG. 5G and FIG. 5H are examples of design drawings and product specifications, respectively. This type of document does not necessarily include a specific character string, but often has a specific format using ruled lines.

  FIG. 5I is an example of a document in which a character string or a graphic (for example, an imprint including a character such as “secret”) indicating that confidential information is included is displayed.

  FIG. 5 (j) and FIG. 5 (k) show an example in which a specific character string and a specific prefix or suffix are mixed in a document.

  In the example of FIG. 5 (j), “Hokkaido” and “Sapporo City” are displayed in the text, and “Hokkaido” and “Manufacturing” are displayed in the footer. In this case, “Hokkaido” in the text is merely a place name, but “Hokkaido” in the footer is a specific company name (or part thereof).

  In the example of FIG. 5 (k), a specific character string “XXΔΔ” indicating a person's name is displayed in the text, and character strings “attendance” and “like” are displayed adjacently before and after the character string. .

  The secure document detection apparatus 0100 according to the present embodiment detects these secure documents based on a keyword included in the input document, a position where the keyword is described, a format of the document, and the like.

  The document shown in FIG. 5 may be input to the secure document detection apparatus 0100 as the secure document example 0501, for example, or may be input to the secure document detection apparatus 0100 as the unmanaged document 0511 (that is, the electronic document file 0411).

  For example, when the document shown in FIG. 5A is input as the secure document example 0501 and the character strings “design document” and “XX manufactory” are input as the secure term definition 0502, based on those inputs. A secure dictionary learning process 0503 is executed. As a result, the character strings “design document” and “xx factory” are registered in the secure document dictionary 0504 as keywords. Further, the positional relationship of these keywords (for example, a vector representing the distance between them and the direction in which they are arranged) is also extracted from the secure document example 0501 and registered in the secure document dictionary 0504. At this time, for example, “XX factory” is the main keyword, and “design document” is the auxiliary keyword, and those sets (keyword pairs) may be registered.

  In this embodiment, an example is shown in which a proper noun such as a company name “XX Manufacturing” is mainly treated as a main keyword, and a common noun such as “design document” is treated as an auxiliary keyword. Can be registered as a main keyword and an auxiliary keyword. For example, a keyword pair made up of an auxiliary keyword “XX factory” and a main keyword “design document” may be registered.

  When the document shown in FIG. 5A is input as an unmanaged document 0511 (that is, an electronic document file 0411), a plurality of keywords extracted from the document and their positional relationships, a plurality of registered keywords, These positional relationships are referred to, and it is determined whether or not the document is a secure document.

  The documents shown in FIGS. 5B to 5F, FIG. 5J, and FIG. 5K are the same as described above. That is, character strings such as company names, document titles, prefixes and suffixes included in those documents are registered as main keywords or auxiliary keywords, and whether or not a document input based on these keywords is a secure document. Is determined.

  The document shown in FIG. 5 is merely a typical example, and the present invention can be applied to all types of secure documents.

  FIG. 6 is an explanatory diagram of secure document detection based on a combination of a plurality of keywords executed by the secure document detection apparatus 0100 according to the first embodiment of this invention.

  When the document 0701 shown in FIG. 6 is input as the unmanaged document 0511, the secure document detection apparatus 0100, from the input document, the keyword including the main keyword “xx” indicating the company name and the auxiliary keyword “confidental”. A keyword pair consisting of a pair and a keyword “Hokkaido” indicating the name of another company and an auxiliary keyword “creation” is extracted. Then, the secure document detection device 0100 calculates the secure information likelihood by comparing the extracted keyword pair and the positional relationship of each keyword pair with the information registered in the secure document dictionary 0504.

  FIG. 7 is an explanatory diagram of blocks identified by the secure document detection device 0100 according to the first embodiment of this invention.

  Specifically, FIG. 7 shows a specific example of blocks extracted in step 0402 of FIG. 4 and rearranged in step 0403.

  A document 0820 shown in FIG. 7A is composed of a title 0851, an author name 0852, and a text 0853. When this document 0820 is input as the electronic document file 0411, the secure document detection apparatus 0100 extracts the block B1_0801, the block B2_0802, and the block B3_0803 (step 0402). Block B1_0801 corresponds to the area where the title 0851 is displayed, block B2_0802 corresponds to the area where the author name 0852 is displayed, and block B3_0803 corresponds to the area where the body text 0853 is displayed.

  A document 0830 shown in FIG. 7B includes a text 0855 and a text 0856. In this example, the text is in a column, and the text 0855 and the text 0856 correspond to each level, and the text 0856 should be read after the text 0855. When this document 0830 is input as the electronic document file 0411, the secure document detection device 0100 extracts the block B5_0805 corresponding to the area where the text 0855 is displayed and the block B6_0806 corresponding to the area where the text 0856 is displayed. (Step 0402). Further, the secure document detection device 0100 rearranges these blocks so that the block B6_0806 follows the block B5_0805 in the same manner as the reading order of the text (step 0403).

  A document 0840 shown in FIG. 7C includes a text 0857, a text 0858, a footnote 0859, a header 0860, and a footer 0861. In this example, the text is in a column, and the text 0857 and the text 0858 correspond to each level, and the text 0858 should be read after the text 0857.

  When this document 0840 is input as the electronic document file 0411, the secure document detection apparatus 0100 extracts block B7_0807, block B8_0808, block B9_0809, block B10_0810, and block B11_0811 (step 0402). Block B7_0807 and block B8_0808 correspond to the area where the text 0857 and text 0858 are displayed, block B9_0809 corresponds to the area where the footnote 0859 is displayed, block B10_0810 and block B11_0811 correspond to the area where the header 0860 and footer 0861 are displayed, respectively To do.

  Further, the secure document detection device 0100 rearranges these blocks so that the block B8_0808 follows the block B7_0807, similarly to the reading order of the text (step 0403).

  8A to 8E are explanatory diagrams illustrating specific examples of keyword extraction and secure document determination executed by the secure document detection device 0100 according to the first embodiment of this invention.

  In the example of FIG. 8A, a document 0610 is input as the electronic document file 0411. The document 0610 is the same as that shown in FIG. The document 0610 includes a character string “design document” 0611 and “XX Manufacturing” 0612. For example, when the combination (keyword pair) of the main keyword “XX factory” corresponding to the company name and the auxiliary keyword “design document” is registered in the keyword information 0412, the secure document detection apparatus 0100 performs step 0404. The character strings “design document” 0611 and “xxx factory” 0612 are extracted as the auxiliary keyword 0613 and the main keyword 0614, respectively, by the keyword extraction process.

  Note that a document 0610 on the left side of FIG. 8A indicates characters or the like actually displayed in the input document. On the other hand, the center and right side documents 0610 are diagrams for explaining the keyword extraction processing. That is, figures such as double-line ellipses and double-line rectangles, and characters such as “company name” and “auxiliary KW” are not actually displayed in the document 0610, and keyword extraction processing is performed. It is given for convenience of explanation. The same applies to FIGS. 8B to 8E.

  Further, the secure document detection device 0100 calculates the strength of the relationship between the extracted main keyword 0614 and auxiliary keyword 0613 based on the positional relationship between the extracted keyword and the extracted keyword based on the strength of the relationship. It is determined whether the pair is a linked keyword pair. In the present embodiment, based on the Euclidean distance between two keywords and the contextual distance of each keyword, the strength of the relationship between the two keywords is calculated. The significance of the cooperative keyword pair will be described later with reference to FIG. 8C and the like, and the criterion for determining the cooperative keyword pair will be described later with reference to Equation (1) and the like.

  When the extracted main keyword 0614 and auxiliary keyword 0613 are linked keyword pairs, a positional relationship between them, specifically, a vector 0615 representing the positional relationship is extracted. This vector 0615 represents the direction from the main keyword 0614 to the auxiliary keyword 0613 and the distance between them. When the similarity between the vector 0615 and the vector indicating the positional relationship between the main keyword “xx factory” and the auxiliary keyword “design document” registered in the secure document dictionary 0504 is higher than a predetermined threshold, the document 0610 is It is determined that the document is a secure document.

  In the learning phase 0500, the user inputs the document 0610 in advance to the secure document detection apparatus 0100 as the secure document example 0501. Further, the secure keyword definition 0502 is a combination of the main keyword “XX factory” and the auxiliary keyword “design document”. As the keyword information 0412, the combination of the main keyword “XX factory” and the auxiliary keyword “design document” and the positional relationship of these keywords in the document 0610 are registered in the secure document dictionary 0504. . Thereafter, when the document 0610 (or a document including the character strings “design document” and “xx manufacturing” as in the document 0610) is input, the secure document detection device 0100 refers to the keyword information 0412 and inputs it. As described above, the main keyword 0614 and the auxiliary keyword 0613 are extracted from the obtained document, and it can be determined based on them whether the document 0610 is a secure document. The same applies to FIGS. 8B and 8C described later.

  In the example of FIG. 8B, a document 0620 is input as the electronic document file 0411. The document 0620 is the same as that shown in FIG. This document 0620 includes a character string “(share) xx” 0621 and “creation” 0622. For example, when the combination of the main keyword “(stock) XX” corresponding to the company name and the auxiliary keyword “creation” is registered as the keyword information 0412, the character string “(stock)” is obtained by the keyword extraction process in step 0404. “XX” 0621 and “Create” 0622 are extracted as the main keyword 0623 and the auxiliary keyword 0624, respectively. Also in this case, as in the case of FIG. 8A, the vector 0625 indicating the positional relationship between the extracted keywords is specified, and based on this, it is determined whether or not the document 0620 is a secure document.

  In the example of FIG. 8C, a document 0630 is input as the electronic document file 0411. The document 0630 is the same as that shown in FIG. This document 0630 includes character strings “Hokkaido” 0631, “Manufacturing” 0632, and “Hokkaido” 0633. For example, if a combination of the main keyword “Hokkaido” corresponding to the first half of the company name “Hokkaido Manufacturing” and the auxiliary keyword “manufacturing” corresponding to the second half is registered as the keyword information 0412, the keyword extraction in step 0404 is performed. Through the processing, the character strings “Hokkaido” 0631 and “Manufacturer” 0632 are extracted as the main keyword 0635 and the auxiliary keyword 0636 constituting the company name 0634, respectively. Also in this case, as in the case of FIG. 8A, the vector 0637 indicating the positional relationship between the extracted keywords is specified, and based on this, it is determined whether or not the document 0630 is a secure document.

  The document 0630 also includes a character string “Hokkaido” 0633 that is the same as the first half of the company name. In this case, a combination of “Hokkaido” 0633 and “Manufacturer” 0632 is also extracted as a keyword pair. However, as can be seen from the character string “Hokkaido” 0633 followed by the character string “Sapporo City”, this character string “Hokkaido” 0633 is not a part of the company name, but merely a place name.

  For example, in the footer portion of the material created by the company “Hokkaido Seisakusho”, for example, a character string “Hokkaido Seisakusho” is printed as shown in FIG. The user can register a keyword pair including the main keyword “Hokkaido” and the auxiliary keyword “manufacturer” and a vector (for example, a vector equivalent to the vector 0637) representing the positional relationship thereof in the secure document dictionary 0504. .

  However, after that, when the document 0630 is input as the electronic document file 0411, not only the combination of “Hokkaido” 0631 and “Manufacturer” 0632 as described above, but also the combination of “Hokkaido” 0633 and “Manufacturer” 0632 Extracted as keyword pairs. In this example, “Hokkaido” 0631 and “Manufacturer” 0632 are part of the company name “Hokkaido Manufactory”, so the relationship between them is strong, but “Hokkaido” 0633 and “Manufacturer” 0632 are completely different from each other. There is no relationship between them because they belong to the context. In such a case, even if the vector representing the positional relationship is specified for the combination of “Hokkaido” 0633 and “Manufacturer” 0632 and the vector is compared with the vector registered in the secure document dictionary 0504, the comparison is not Does not contribute to secure document detection. For this reason, it is desirable to exclude the combination of “Hokkaido” 0633 and “Manufacturing” 0632 from the comparison target of vectors.

  The secure document detection device 0100 of this embodiment further extracts a keyword pair (hereinafter referred to as a cooperative keyword pair) to be compared with a vector registered in the secure document dictionary 0504 from the extracted keyword pair. Whether or not the extracted keyword pair is a linked keyword pair is determined based on the strength of the relationship between the two keywords included in the keyword pair and the importance level determined in advance for the keywords. . The positional relationship between the linked keyword pairs extracted in this way is compared with a vector registered in the secure document dictionary 0504.

  For example, the secure document detection device 0100 calculates the strength of the relationship between two keywords included in all keyword pairs extracted from one document, and the ranking of the values is higher than a predetermined threshold value. You may extract as a cooperation keyword pair. Alternatively, the secure document detection device 0100 may extract a link keyword pair whose association strength calculated as described above exceeds a predetermined threshold. If the relationship between “Hokkaido” 0633 and “Manufacturer” 0632 is sufficiently low, the combination of “Hokkaido” 0633 and “Manufacturer” 0632 is not extracted as a linked keyword pair.

  Furthermore, the secure document detection apparatus 0100 of this embodiment can also positively exclude keywords that should not be extracted as linked keyword pairs.

  For example, in the learning phase 0500, the user may register a combination of the character string “Hokkaido” and the character string “Sapporo City” in the keyword information 0412 as a keyword pair that should not be extracted as a cooperative keyword pair. If such information is registered, when the document 0630 is input, it is determined that the character string “Hokkaido” 0633 is strongly related to the character string “Sapporo City”. Not extracted.

  In the example of FIG. 8D, a document 0640 is input as the electronic document file 0411. The document 0640 is the same as that shown in FIG. Document 0640 includes drawing 0641. The drawing 0641 includes, for example, drawings of parts and the like (not shown), a drawing title 0642, a drawing creation date 0643, and the like, and the layout of these elements is defined by ruled lines 0644. The layout of the document 0640 is extracted as the specific format 0645 by the ruled line / layout extraction process in step 0404. The secure information likelihood of the document 0640 can be calculated by comparing the specific format 0645 with the pattern information 0413 (step 0405).

  It should be noted that the specific format 0645 can be registered in the secure document dictionary 0504 as the pattern information 0413 when the user inputs the document 0640 as the secure document example 0501 to the secure document detection device 0100 in advance in the learning phase 0500. Thereafter, when the document 0640 (or a document having a layout similar to that of the document 0640) is input, the secure document detection device 0100 refers to the pattern information 0413 and specifies the specific format 0645 as described above from the input document. Can be extracted. The same applies to FIG. 8E described later.

  In the example of FIG. 8E, a document 0650 is input as the electronic document file 0411. The document 0650 is the same as that shown in FIG. The document 0650 includes an imprint 0651. The seal impression 0651 includes a character “secret” which means that the document on which it is displayed contains confidential information. The imprint 0651 is extracted as the specific style 0652 by the ruled line / layout extraction process in step 0404. The secure information likelihood of the document 0650 can be calculated by comparing the specific format 0652 and the pattern information 0413 (step 0405).

  Next, secure information likelihood calculation executed in step 0406 of FIG. 4 will be described.

An index L _pair (mw _i , hw _j ) indicating the degree of cooperation of the extracted combination (pair) of the main keyword mw _i and the auxiliary keyword hw _j is calculated by the following equation (1).

Here, D _BLK (mw _i , hw _j ) is a document type block distance between the main keyword and the auxiliary keyword. In other words, the document type block distance is a distance in the context of two keywords, and is an index indicating the strength of the relationship between the contexts of the two keywords. In general, if the order in which two keywords are read is close, their contextual association is strong. For example, comparing the document-type block distances when two keywords belong to one block and when each belongs to another block, even if the Euclidean distance between the two keywords is the same, generally the latter document The type block distance is larger than the former document type block distance.

  Here, the block distance is an extension of the concept of distance used in the field of image processing for documents. In general, the block distance in image processing refers to a distance measure group in which the distance between two points is represented by the sum of the difference in the X direction and the difference in the Y direction | X | + | Y | or | X + Y | . The simple block distance is a uniform distance measure between any two points on the document that does not reflect the layout structure representing semantic information such as paragraphs and tables written on the document. The document type block distance changes the weight of the distance measure by reflecting the strength of the relation in the document context from the document structure.

  α is a weighting coefficient for calculating the degree of cooperation based on the document type block distance. The user can set an arbitrary value as α, but it is desirable to set an appropriate value according to the arrangement of the two keywords. An example of the value of α will be described later with reference to FIG.

D _EUC (mw _i , hw _j ) is the Euclidean distance between the main keyword and the auxiliary keyword, that is, the Euclidean distance between the position where the main keyword is displayed and the position where the auxiliary keyword is displayed in the document. Distance.

  β is a weighting coefficient for calculating the degree of cooperation based on the Euclidean distance. The user can set an arbitrary value as β.

  Note that “+1” in the denominator of the first term and the second term on the right side of Expression (1) is attached to prevent the value from diverging when the distance is zero.

L _word (mw _i , hw _j ) is an index representing the importance of keywords (importance of each keyword or importance of keyword pairs), and is determined in advance by the user. For example, the user may set a higher value for L _word (mw _i , hw _j ) relating to a combination of keywords including the names of important customers than for other combinations.

  γ is a weighting factor for calculating the degree of cooperation based on the importance of the keyword. The user can set an arbitrary value as γ.

As a result, the degree of cooperation between the main keyword mw _i and the auxiliary keyword hw _j is higher as the document type block distance D _BLK (mw _i , hw _j ) is smaller, and as the Euclidean distance D _EUC (mw _i , hw _j ) is smaller The higher the importance of a predetermined keyword is, the higher the value is.

The secure document detection device 0100 calculates L _word (mw _i , hw _j ) for all keyword pairs extracted from the input document, and determines that those having large values are linked keyword pairs. Good. Specifically, for example, when the value of L _word (mw _i , hw _j ) of a certain keyword pair is larger than a predetermined threshold, it may be determined that the keyword pair is a linked keyword pair. Alternatively, all of the _{L word (mw i, hw j} ) calculated for each document within the keyword pair size rank corresponds to greater than a predetermined threshold value may be determined that the cooperative keyword pairs .

The secure information likelihood (that is, an index indicating the likelihood of a secure document of the document) L _sequre (doc _i ) of the input document doc _i is calculated by the following equation (2).

Here, dic _j is the j-th document included in the secure document example 0501 (that is, the document case 0702) input to the secure document dictionary 0504.

L _format (doc _i , dic _j ) is an index indicating the secure document quality of the document doc _i based on a comparison between the format of the document doc _{i and} the format of the document dic _j . Specifically, the value of L _format (doc _i , dic _j ) increases as the similarity between the layout of document doc _i extracted in step 0404 in FIG. 4 and the layout of document dic _j increases. .

_{L keyword (doc i, dic j} ) is based on a comparison of the keywords included in the keyword and document dic _j included in the document doc _i, an index indicating the secure document likeness document doc _i. Specifically, the positional relationship between the keyword pairs extracted from the document doc _i in step 0404 of FIG. 4 and the keyword pairs included in the document dic _j (that is, the keyword combination extracted from the secure document example 0501 or the secure term definition). The degree of similarity with the positional relationship of the keyword combination input as 0502 is calculated, and the value of L _keyword (doc _i , dic _j ) increases as the degree of similarity increases. A method for calculating L _keyword (doc _i , dic _j ) will be described later (see Equation (3)).

The maximum value of L _format (doc _i , dic _j ) + L _keyword (doc _i , dic _j ) calculated for all documents dic _j is L _sequre (doc _i ).

Note that the secure information likelihood of the document doc _i may be calculated based only on the combination of keywords, regardless of the document layout. In that case, it is not necessary to calculate L _format (doc _i , dic _j ), and the maximum value of L _keyword (doc _i , dic _j ) is L _sequre (doc _i ).

L _keyword (doc _i , dic _j ) is calculated by Equation (3).

The distance between the vector representing the positional relationship between the keyword pairs extracted from the document doc _i and the vector representing the positional relationship between the keyword pairs included in the document dic _j is calculated by Equation (3), and secure based on the distance. Information likelihood is calculated. At this time, the mathematical formula (3) may be calculated not for all the keyword pairs extracted from the document doc _i but only for the linked keyword pairs. In this case, the total sum of the likelihoods calculated for all the linked keyword pairs extracted from the document doc _i is L _keyword (doc _i , dic _j ).

The purpose of L _format is to find a document registered as a dictionary and a document having similar keywords and in a similar arrangement. Equation (3) shows that the similarity between keyword pairs is derived by simply defining the Euclidean distance. That is, there is some distance measure between keywords (for example, auxiliary keywords such as “confidential” and “prepared” are regarded as similar, and the distance is 0, and an auxiliary keyword “ A distance scale can be defined between keywords so that the “middle”, “destination”, etc. are regarded as different types and the distance is large.

  For example, the distance scale is the distance scale for the same part of speech in the morphological analysis, and the distance scale is 1 for the other parts), and there is a distance scale between the keyword placement positions (for example, the document type block distance described above). If there is a distance measure between the similarity measures, the distance between the two keyword pairs can be calculated if the Euclidean distance is calculated by regarding these as vectors as in Equation (3).

It is also possible to introduce likelihood into this. If the relevance L _word of each of the two keyword pairs is high and the likelihood is high that the arrangement is similar, the likelihood obtained by converting the distance between 0 and 1 is the likelihood Can be seen. That is, for any calculation formula that uses a term having a likelihood as an input, the likelihood attached to it can be calculated.

  FIG. 9 is an explanatory diagram of the arrangement cost table 0900 included in the secure document dictionary 0504 according to the first embodiment of this invention.

  The arrangement cost table 0900 includes the positions of two keywords included in the keyword pair extracted from the document in the document (that is, from which area in the document the keywords are extracted), and the expression (1). It is a table which matches weighting coefficient (alpha). Specifically, the arrangement cost table 0900 includes rows 0901 to 0903 and columns 0911 to 0913.

  In row 0901, the value of the weighting coefficient α when the main keyword is extracted from the table in the document is registered. In row 0902, the value of the weighting coefficient α when the main keyword is extracted from the text or title in the document is registered. In row 0903, the value of the weighting factor α when the main keyword is extracted from the header or footer in the document is registered.

  In a column 0911, the value of the weighting factor α when the auxiliary keyword is extracted from the table in the document is registered. In a column 0912, the value of the weighting coefficient α when the auxiliary keyword is extracted from the text in the document is registered. In a column 0913, the value of the weighting factor α when the auxiliary keyword is extracted from the header or footer in the document is registered.

  In the example of FIG. 9, column 0912 corresponds to the text, but column 0912 may correspond to the case where the auxiliary keyword is extracted from the text or the title. Moreover, the classification of the areas as described above is merely an example. For example, when documents are arranged in columns, each column may be treated as an independent area. Alternatively, the header and footer may be treated as independent areas.

  In the example of FIG. 9, when both the main keyword and the auxiliary keyword are extracted from the table in the document, the value of the weighting factor α is “α11”. When the main keyword is extracted from the title or the text and the auxiliary keyword is extracted from the table, the value of the weight coefficient α is “α21”.

  The user can register an arbitrary value as the weighting factor α in the arrangement cost table 0900. However, in general, when the main keyword and the auxiliary keyword are extracted from the same area in the document (for example, title, text, table, header, footer, etc.), the main keyword and the auxiliary keyword are compared with the other cases. Estimated to be close in context. As will be described later, the closer the context distance between two keywords is, the higher the possibility that the relationship between the keywords is strong. For this reason, typically, the value of α when the main keyword and the auxiliary keyword are extracted from the same region in the document is set to be larger than that when it is not. For example, typically, the value of α11 is larger than the value of α21.

  When the main keyword and the auxiliary keyword are extracted from the same region, the value of the weighting factor α may be further determined according to the positional relationship between these keywords.

  Specifically, when the main keyword is extracted from the text or the title and the auxiliary keyword is extracted from the text, the weighting coefficient α is further determined according to the position of the main keyword and the auxiliary keyword in the title and the text. The

  For example, when the main keyword and the auxiliary keyword are words adjacent to each other, the value of the weighting coefficient α is “α22a”. When the main keyword and the auxiliary keyword are not adjacent but are included in the same row, the value of the weight coefficient α is “α22b”. When the main keyword and the auxiliary keyword are included in different lines, but are included in the same paragraph, the value of the weight coefficient α is “α22c”. When the main keyword and the auxiliary keyword are included in different paragraphs, the value of the weighting coefficient α is “α22d”.

  The user can arbitrarily set these values. However, if two keywords are included in the same paragraph rather than included in different paragraphs, the two keywords are included in the same line rather than included in different lines, rather than the two keywords are not adjacent. The context distance of those keywords is closer. The closer the context, the more likely the keywords are related.

  For example, when the main keyword “(share) xx” and the auxiliary keyword “creation” are extracted from the continuous character string “(share) xx creation”, these keywords are adjacent to each other. In this case, the contextual meanings of “(share) xx” and “creation” are usually related to each other. Specifically, the above character string means that the company “(share) XX” “created” something, and the document itself including those keywords as shown in the example of FIG. May be created by a company “(share) xx”.

  On the other hand, for example, even when the main keyword “(share) xx” and the auxiliary keyword “creation” are extracted from different paragraphs, the keywords may be related to each other. However, the auxiliary keyword “created” may be extracted from a character string “created by XX factory”, for example. In this case, the extracted keyword combination does not mean that “(stock) xx” “created” something. That is, there is no contextual relationship between “(share) xx” and “creation”. In this case, it is unlikely that the document itself including those keywords is created by a company “(share) xx”.

  When it is necessary to determine that a document created by the company “(share) xx” is a secure document, a keyword pair consisting of the main keyword “(share) xx” and the auxiliary keyword “create” is linked to the keyword. It is desirable to extract as a pair. Considering the above example, it is desirable to calculate the degree of cooperation when the main keyword “(share) xx” and the auxiliary keyword “creation” are adjacent to each other so as to be higher than the case where the main keyword “(stock) XX” is adjacent. Therefore, typically, the value of the weighting factor α is set so that the value of α22c is larger than α22d, the value of α22b is larger than α22c, and the value of α22a is larger than α22b.

  However, in practice, there is a case where the degree of cooperation between the main keyword and the auxiliary keyword extracted from the areas separated from each other should be calculated high. For example, in the example of FIG. 5B, the main keyword “xx” is extracted from the footer, and the auxiliary keyword “specification” is extracted from the title. When it is desired to extract a keyword pair arranged in this way as a linked keyword pair, the value of α32 of the arrangement cost table 0900 corresponding to the keyword pair may be set larger than other values. However, in that case, column 0912 corresponds to not only the text but also the title.

  When both the main keyword and the auxiliary keyword are extracted from the table, the weighting factor α is further determined according to the position of the main keyword and the auxiliary keyword in the table, as in the case of α22 described above.

  For example, when the main keyword and the auxiliary keyword are extracted from cells adjacent to each other in the table, the value of the weighting coefficient α is “α11a”. When the main keyword and the auxiliary keyword are extracted from the same table (but not from cells adjacent to each other), the value of the weight coefficient α is “α11b”. When the main keyword and the auxiliary keyword are extracted from different tables, the value of the weighting factor α is “α11c”. As in the case of α22, the user can arbitrarily set these values. For example, for the same reason as in the case of α22 described above, α11b may be set larger than α11c, and α11a may be set larger than α11b.

  Similarly, when both the main keyword and the auxiliary keyword are extracted from the header or footer, the weighting factor α is further determined according to the position of the main keyword and auxiliary keyword in the header or footer.

  For example, when the main keyword and the auxiliary keyword are extracted from the same row, the value of the weighting factor α is “α33a”, and when they are extracted from different rows, the value of the weighting factor α is “α33b”. . As in the case of α22, the user can arbitrarily set these values. For example, α33a may be set larger than α33b for the same reason as in the case of α22 described above.

  The secure document dictionary 0504 may include a plurality of arrangement cost tables 0900. For example, as shown in FIG. 8C, when the character string “Hokkaido Seisakusho” is to be extracted as a linked keyword pair, but “Hokkaido” included in the string “Hokkaido Sapporo City” is not to be included in the linked keyword pair, An arrangement cost table 0900 relating to a combination of “Hokkaido” and “Sapporo City” (or similar city name in Hokkaido) is further created, and the value of α22a in the placement cost table 0900 is extracted as a linked keyword pair. A value smaller than the value given to the arrangement (for example, “0”) may be used. In this case, since the degree of cooperation regarding the character string “Hokkaido Sapporo City” is low, the influence of the character string such as “Hokkaido Sapporo City” on the secure information likelihood determination of the document can be suppressed.

  FIG. 10 is an explanatory diagram of the secure document dictionary 0504 according to the first embodiment of this invention.

  The secure document dictionary 0504 includes a secure dictionary header 1001, a plurality of keywords 1011 and the like, one or more arrangement cost tables 1021 and the like, and one or more specific forms 1031 and the like.

  The secure dictionary header 1001 includes information indicating the version of the secure document dictionary 0504 and information describing the contents of the dictionary.

  Each of the keywords 1011 and the like includes a character string designated as a keyword and additional information related to the keyword. The additional information includes information indicating whether the keyword is a main keyword or an auxiliary keyword, information specifying a main keyword or an auxiliary keyword combined with the keyword, a part of speech of the keyword (for example, a proper noun such as a company name or Information indicating a common noun such as “secret”), information indicating the importance of the keyword, and the like. This additional information may further include information for specifying a combination of keywords that should not affect the secure information likelihood determination, as in the example of “Hokkaido Sapporo City”.

  Further, each of the keywords 1011 and the like may include vector data indicating the positional relationship between keyword pairs including a main keyword and an auxiliary keyword. This vector data is compared with vector data indicating the positional relationship of keyword pairs extracted from the input document, for example, in step 0406 of FIG.

  FIG. 10 shows a keyword 1 — 1011 and a keyword 2 — 1012 as examples of the keyword 1011 and the like, but the secure document dictionary 0504 may include more keywords.

  Each of the arrangement cost table 1021 and the like corresponds to the arrangement cost table 0900 described with reference to FIG. 9, information indicating the types of main keywords and auxiliary keywords corresponding to the arrangement cost table 1021, and the like Includes information indicating the weight (importance) of. As described with reference to FIG. 9, a plurality of arrangement cost tables 0900 may be created. For example, for each type of keyword pair, an arrangement cost table 1021 corresponding to the keyword pair may be created. Or the arrangement | positioning cost table 1021 etc. corresponding to only a specific keyword pair may be produced.

  FIG. 10 shows an arrangement cost table 1_1021 and an arrangement cost table 2_1022 as examples of the arrangement cost table 1021 and the like, but the secure document dictionary 0504 may include more arrangement cost tables.

  Each of the specific forms 1031 and the like is a method and range for extracting the form of the document (specifically, vector data corresponding to a specific format or graphic as shown in FIGS. 8D and 8E), and extraction. Format vector data (that is, vector data such as ruled lines or seals previously extracted from the registered secure document example 0501 and registered). FIG. 10 shows a specific form 1_1031 and a specific form 2_1032 as examples of the specific form 1031 and the like, but the secure document dictionary 0504 may include more specific forms.

  In the first embodiment, an example is shown in which the extracted positional relationship is compared with the previously registered positional relationship for a keyword pair composed of two keywords extracted from the document. However, the same processing as described above may be executed for a group of keywords including three or more keywords. For example, vector data indicating three keywords and their positional relationship may be registered in the secure document dictionary 0504. In that case, vector data indicating the three keywords and their positional relationship are extracted from the input document. Then, based on the similarity between the extracted vector data and the registered vector data, it is determined whether or not the input document is a secure document.

  According to the first embodiment of the present invention described above, whether or not the document is secure is automatically determined based on a keyword or the like included in the input document. Can be stored with encryption or the like. In particular, according to the present embodiment, secure document determination is performed in consideration of the association of a plurality of keywords in the context based on the combination of the plurality of keywords and the context distance of the combination. Based on the degree of cooperation of the keyword combinations extracted from the document, it is determined whether or not to use the combination for secure document determination. As a result, it is possible to reduce both the omission of detection of a document that should be originally secure and the erroneous detection of a document that is not inherently secure, thereby realizing highly accurate secure document determination. As a result, it is possible to reliably protect a secure document while suppressing management costs by the user.

<Second Embodiment>
FIG. 11 is a block diagram illustrating a hardware configuration of the OCR integrated secure document detection apparatus 0200 according to the second embodiment of this invention.

  The OCR integrated secure document detection apparatus 0200 is an example of an apparatus that realizes the OCR integrated paper document management of the present invention shown in FIG.

  The OCR integrated secure document detection device 0200 of this embodiment includes an operation terminal device 0201, a display terminal device 0202, an external storage device 0203, a memory 0204, a central processing unit 0205, a communication device 0207, an image imaging device 0208, a sorter device 0209, and A communication line 0206 is provided for interconnecting them.

  The operation terminal device 0201, the display terminal device 0202, the external storage device 0203, the memory 0204, the central processing unit 0205, the communication device 0207, and the communication line 0206 are respectively the operation terminal device 0101, the display terminal device 0102, and the external Since the storage device 0103, the memory 0104, the central processing unit 0105, the communication device 0107, and the communication line 0106 are the same, detailed description thereof will be omitted.

  The image capturing apparatus 0208 includes an optical scanner that reads characters, ruled lines, graphics, and the like described in the input paper document 0306 and converts them into data. The data read at this time may be stored in the external storage device 0203 as a file including text data and image data.

  The sorter device 0209 is a device that discharges the paper document 0306 after reading by the image pickup device 0208 is completed. For example, the sorter apparatus 0209 may include a plurality of shelves as the discharge destination of the paper document 0306. In this case, the sorter device 0209 can discharge the paper document 0306 to the selected shelf as necessary.

  Note that the OCR integrated secure document detection device 0200 may be realized by adding a conventional OCR device as the image capturing device 0208 and the sorter device 0209 to the secure document detection device 0100 of the first embodiment. In this case, for example, the image capturing device 0208 and the sorter device 0209 in FIG. 11 correspond to the OCR device 0307 in FIG. 1, and the remaining portion in FIG. 11 corresponds to the computer 0308 in FIG. Alternatively, the entire OCR integrated secure document detection device 0200 may be realized as one OCR device.

  The central processing unit 0205 of the OCR integrated secure document detection device 0200 uses the data read by the image capturing device 0208 to determine whether or not the input document is a secure document. Since the process is the same as that of the first embodiment described with reference to FIGS. 3 to 10, the description thereof is omitted.

  Furthermore, the OCR integrated secure document detection device 0200 according to the present embodiment not only outputs the secure electronic document 0415 based on the determination result of step 0407 but also the paper document after the reading by the image capturing device 0208 is completed. 0306 can be selected. This is because when the paper document 0306 is a secure document, the secure information contained therein is protected from outflow.

  For example, the OCR integrated secure document detection apparatus 0200 may discharge the paper document 0306 determined to be a secure document to a position different from that of the paper document 0306 that is not. Here, the “different positions” may be, for example, “different shelves” or “different positions in the same shelf”.

  Alternatively, the OCR integrated secure document detection device 0200 may process and discharge the paper document 0306 determined to be a secure document. Here, “processing” may mean printing a display (for example, a character such as “secret”) indicating that the paper document 0306 is a secure document, or a character on the paper document 0306. A predetermined graphic pattern or the like for making it difficult to read may be printed, or the paper document 0306 may be crushed to make it impossible to read those characters or the like. In this case, the sorter device 0209 includes a document processing device such as a printing device or a shredder.

  Alternatively, the OCR integrated secure document detection device 0200 may discharge the paper document 0306 determined not to be a secure document to a normal shelf and may not discharge the paper document 0306 determined to be a secure document. In this case, the paper document 0306 determined to be a secure document is stored in the OCR integrated secure document detection apparatus 0200 until it is taken out by a user having appropriate authority.

  In the above, an example in which it is determined in step 0407 whether or not the document is a secure document has been described. However, in step 0407, multivalue determination may be performed instead of such binary determination. For example, the OCR integrated secure document detection device 0200 may determine the rank of the secure information likelihood by comparing the calculated secure information likelihood with a plurality of threshold values. In that case, a discharge method of the paper document 0306 may be selected according to the determined rank. For example, the integrated secure document detection apparatus 0200 may crush and discharge the highest-ranked paper document 0306 and discharge the paper documents 0306 of other ranks to the shelves assigned to the respective ranks.

  According to the second embodiment of the present invention described above, high-accuracy secure document determination similar to that of the first embodiment is realized for a document printed on paper, thereby ensuring a secure electronic document. Can be protected. Further, by linking the result of the secure document determination and the discharge of the input paper, the secure paper document can be reliably protected while suppressing the management cost of the user.

0100 Secure document detection device 0101, 0201 Operation terminal device 0102, 0202 Display terminal device 0103, 0203 External storage device 0104, 0204 Memory 0105, 0205 Central processing unit 0106, 0206 Communication line 0107, 0207 Communication device 0200 OCR integrated secure document detection Apparatus 0208 Image imaging apparatus 0209 Sorter apparatus 0301, 0306 Paper document 0302 Optical character reader (OCR apparatus)
0303, 0305, 0309, 0310, 0312 Document file 0304, 0308, 0311 Computer 0501 Secure document example 0502 Secure term definition 0504 Secure document dictionary 0511 Unmanaged document 0900 Arrangement cost table

Claims (20)

A secure document detection method executed by a secure document detection device,
The secure document detection device includes an arithmetic device and a storage device that holds a dictionary,
In the dictionary, a plurality of keyword pairs each including at least two keywords, and information indicating a positional relationship in the document of the two keywords included in each keyword pair are registered,
The secure document detection method includes:
A first procedure for extracting keyword pairs registered in the dictionary from input document data;
A second procedure for determining whether or not the input document data is a secure document based on the positional relationship in the input document data of two keywords included in the extracted keyword pair; A secure document detection method comprising: The second procedure includes
A third procedure for determining whether or not the extracted keyword pair is a cooperative keyword pair based on the strength of association between the two keywords included in the extracted keyword pair;
The secure information likelihood of the input document data is increased so that the similarity between the positional relationship between the two keywords included in the cooperative keyword pair and the positional relationship between the two keywords registered in the dictionary increases. A fourth procedure for calculating the degree;
The secure document detection method according to claim 1, further comprising: a fifth step of determining that the input document data is a secure document when the secure information likelihood is greater than a predetermined threshold. . The secure document detection method further includes a procedure of classifying text included in the input document data into a plurality of regions,
The third procedure includes
A sixth procedure for calculating the strength of association between the two keywords based on the region including the two keywords;
When the strength of the relation exceeds a predetermined threshold, or when the rank of the relation strength in the input document data exceeds a predetermined threshold, the extracted keyword pair is a linked keyword pair The secure document detection method according to claim 2, further comprising: a seventh procedure for determining that there is. The plurality of areas include at least one of a title, text, table, header, or footer,
The dictionary further includes placement cost information that associates the combination of the regions with a predetermined weight,
The sixth procedure includes a procedure of calculating the strength of association between the two keywords so that the larger the predetermined weight corresponding to the combination of regions to which the two keywords belong, the larger the predetermined procedure is. The secure document detection method according to claim 3. In the dictionary, keyword pairs that should not be extracted as the linked keywords are further registered,
5. The arrangement cost information is registered with a value smaller than the weights corresponding to other keyword pairs as the weight corresponding to a keyword pair that should not be extracted as the cooperation keyword. The described secure document detection method. The secure document detection method further includes a procedure of rearranging text included in the input document data in the order in which the text is read.
The sixth procedure calculates a contextual distance between the two keywords based on the order in which the text is read, and relates the relation between the two keywords so as to increase as the calculated contextual distance decreases. The secure document detection method according to claim 3, further comprising a procedure for calculating the strength.   The sixth procedure includes a procedure of calculating a strength of association between the two keywords so that the Euclidean distance in the input document data of the two keywords is shorter. 4. The secure document detection method according to 3.   The method of claim 1, wherein the secure character detection method further includes a step of encrypting and outputting the input document data when it is determined that the input document data is a secure document. The described secure document detection method.   In the secure character detection method, when a document to be determined to be a secure document and information specifying two keywords included in the document are input, information indicating the positional relationship between the two specified keywords The secure document detection method according to claim 1, further comprising a step of: extracting the specified keyword and information indicating the extracted positional relationship in the dictionary.   The information indicating the positional relationship between two keywords included in the keyword pair includes vector data indicating a direction indicating the arrangement of the two keywords in the document and a distance between them. The described secure document detection method. A secure document detection program executed by a computer,
The computer includes an arithmetic device, a memory that stores the secure document detection program, and a storage device that holds a dictionary.
In the dictionary, a plurality of keyword pairs each including at least two keywords, and information indicating a positional relationship in the document of the two keywords included in each keyword pair are registered,
The secure document detection program is:
A first procedure for extracting keyword pairs registered in the dictionary from input document data;
A second procedure for determining whether or not the input document data is a secure document based on the positional relationship in the input document data of two keywords included in the extracted keyword pair; A secure document detection program executed by the computer. The second procedure includes
A third procedure for determining whether or not the extracted keyword pair is a cooperative keyword pair based on the strength of association between the two keywords included in the extracted keyword pair;
The secure information likelihood of the input document data is increased so that the similarity between the positional relationship between the two keywords included in the cooperative keyword pair and the positional relationship between the two keywords registered in the dictionary increases. A fourth procedure for calculating the degree;
The secure document detection program according to claim 11, further comprising: a fifth step of determining that the input document data is a secure document when the secure information likelihood is greater than a predetermined threshold. . The secure document detection program further causes the computer to execute a procedure for classifying text included in the input document data into a plurality of regions,
The third procedure includes
A sixth procedure for calculating the strength of association between the two keywords based on the region including the two keywords;
When the strength of the relation exceeds a predetermined threshold, or when the rank of the relation strength in the input document data exceeds a predetermined threshold, the extracted keyword pair is a linked keyword pair The secure document detection program according to claim 12, further comprising a seventh procedure for determining that there is. The plurality of areas include at least one of a title, text, table, header, or footer,
The dictionary further includes placement cost information that associates the combination of the regions with a predetermined weight,
The sixth procedure includes a procedure of calculating the strength of association between the two keywords so that the larger the predetermined weight corresponding to the combination of regions to which the two keywords belong, the larger the predetermined procedure is. The secure document detection program according to claim 13.   The secure character detection program further causes the computer to execute a procedure of encrypting and outputting the input document data when it is determined that the input document data is a secure document. The secure document detection program according to claim 11.   In the secure character detection method, when a document to be determined to be a secure document and information specifying two keywords included in the document are input, information indicating the positional relationship between the two specified keywords The secure document detection program according to claim 11, wherein the computer is caused to execute a procedure of extracting the specified keyword and the information indicating the extracted positional relationship in the dictionary. An optical character reader that reads character information from an input paper document,
The optical character reader includes an arithmetic device, a storage device that holds a secure information dictionary, an image pickup device that reads the input paper document, and a paper discharge device that discharges the input paper document. Prepared,
In the secure information dictionary, a plurality of keyword pairs each including two keywords, and information indicating the positional relationship in the document of the two keywords included in each keyword pair are registered,
The optical character reader is
Create document data by reading character information from the input paper document,
Extracting keyword pairs registered in the secure document dictionary from the document data;
Determining whether or not the input paper document is a secure document based on the positional relationship in the document data of two keywords included in the extracted keyword pair;
An optical character reader that controls a method of discharging the input paper document in accordance with the result of the determination. The paper discharge device includes a plurality of shelves,
The optical character reader discharges the input paper document so that a paper document determined to be a secure document and a paper document determined to be not a secure document are discharged to different shelves. The optical character reader according to claim 17, wherein: The paper discharge device includes a printing device,
When it is determined that the input paper document is a secure document, the optical character reader prints a display indicating that the input paper document is a secure document on the input paper document, and the printed The optical character reader according to claim 17, wherein the paper document is discharged. The paper discharge device includes a processing device that processes the input paper document so as to be difficult to read,
When it is determined that the input paper document is a secure document, the optical character reader processes the input paper document so as to be difficult to read, and discharges the processed paper document The optical character reader according to claim 17.