JP2009230427A - Method, device and program for detection and estimation of electronic document attribute, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method, device and program for detection and estimation of electronic document attribute, capable of appropriately detecting and estimating an attribute corresponding to a document without attribute association from a preliminarily accumulated electronic document database, and a recording medium therefor. <P>SOLUTION: The method for detection and estimation of electronic document attribute for detecting and estimating an attribute corresponding to a document without attribute association from a preliminarily accumulated electronic document database includes a search procedure for searching an electronic document with high similarity to the document from electronic documents accumulated in the database; a comparison procedure for comparing, for each attribute of the electronic document searched by the search procedure, the similarity of the document for the attribute concerned with a threshold corresponding to an importance adjusted for this attribute; and an estimation procedure for estimating, when the similarity of each attribute of the searched electronic document exceeds the threshold, the attributes of the electronic document as the attributes of the document. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an electronic document attribute detection / estimation method, an electronic document attribute detection / estimation apparatus, and an electronic document attribute detection / estimation program for detecting and estimating an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance. And a storage medium.

  In general, in an enterprise, the information assets are formed, stored, and used in the form of electronic documents (hereinafter simply referred to as documents). Considering the confidentiality of this in-company document, it is very important to control the handling of the corporate document according to the confidentiality. Against this background, there are various techniques for restricting the handling of corporate documents.

  For example, in the technology described in Patent Document 1, similar document search technology is used to analyze the contents of unprocessed work documents and the like, and their attributes (for example, classification, category, confidential level, This is a system for estimating the attributes and the like for managing and organizing types. In the similarity determination, the reliability value indicating the similarity, which is given by the similar document search engine, is compared with a preset threshold value, and if it exceeds, it is determined as similar, and if it does not exceed, it is determined as dissimilar. By this determination, the attribute of the in-process document as the estimation target is estimated as the attribute value of the document determined to be similar.

Accordingly, access control or the like can be performed on a document that has not been managed in accordance with a policy that is an access restriction rule based on the estimated attribute value.
JP 2006-185153 A

  However, according to the technique described in Patent Document 1 and the like, the threshold value at the time of similarity determination is set to be constant regardless of the document attribute, which may cause inconvenience in operation. For example, if the product A theme that is the attribute of the document is in secret development that has not been disclosed to the public, the document relating to the product A does not want to be known by anyone other than the developer. It is necessary to strictly check outflows to the outside due to copying, scanning, printing, etc. due to a document having no access authority for a document estimated as a document related to the product A theme.

  On the other hand, if the product has already been released and the functions and performance are disclosed, there are cases where it is desired to check the outflow of only the confidential document related to the product B to the outside.

  In the case of the product A theme, it is necessary to set a low threshold for measuring the similarity and to check a wide range of documents with low similarity, but in the case of the product B theme, the threshold for measuring the similarity is set. I want to check only information with high similarity by setting it higher. In this way, for example, when a warning message is sent to the administrator by changing and adjusting the threshold according to the importance of the document attribute for each document attribute without setting the threshold at the time of similarity determination constant. However, it is possible to make adjustments suitable for actual operation so that the check function is exhibited only when necessary, and to send an appropriate warning message.

  The present invention has been made in view of the above points, and an electronic document attribute detection estimation method that appropriately detects and estimates an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance, An object is to provide an electronic document attribute detection estimation apparatus, an electronic document attribute detection estimation program, and a storage medium.

  In order to solve the above-described problems, the present invention is characterized by the following measures.

  The present invention relates to an electronic document attribute detection and estimation method for detecting and estimating an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance, and the electronic document among the electronic documents stored in the database A search procedure for searching for an electronic document having high similarity to the document, and for each attribute of the electronic document searched by the search procedure, the similarity of the document with respect to the attribute and an importance adjusted for the attribute A comparison procedure for comparing a threshold corresponding to the degree, and an estimation procedure for estimating the attribute of the electronic document as the attribute of the document when the similarity for each attribute of the retrieved electronic document exceeds the threshold. Have.

  Further, as a means for solving the above-described problems, the present invention is an electronic document attribute detection / estimation apparatus that detects and estimates an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance. Search means for searching for an electronic document having a high similarity to the document among the electronic documents stored in the database, and for each attribute of the electronic document searched by the search means, the similarity of the document to the attribute And a comparison means for comparing the threshold value corresponding to the importance adjusted for the attribute, and when the similarity for each attribute of the searched electronic document exceeds the threshold value, the attribute of the electronic document is Estimating means for estimating the attribute of the document is executed.

  Furthermore, as means for solving the above-described problems, the present invention may be a program for causing a computer to execute the procedure in the above-described electronic document attribute detection estimation method, and a computer-readable storage medium.

  According to the present invention, an electronic document attribute detection / estimation method, an electronic document attribute detection / estimation device, and an electronic document that appropriately detect and estimate an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance. An attribute detection estimation program and a storage medium can be provided.

  Next, the best mode for carrying out the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram illustrating a schematic configuration example of the first system 101. A system 101 shown in FIG. 1 includes an electronic document attribute detection / estimation apparatus 10, an administrator terminal 20, a multifunction machine 30-1, a document management DB (database), a file server 40 as a document management apparatus, etc. (Whether wired or wireless). The multifunction machine 30-1 includes at least a scanner and a plotter, and enables a plurality of image processing such as a printer, a copy, and a scanner. Further, it is assumed that the document attribute detection estimation apparatus 10, the administrator terminal 20, and the like are configured in a space where the confidentiality of information should be maintained, such as in the same company or office.

  The document attribute detection estimation apparatus 10 installs the document attribute learning program 11 from the document learning cooperation program 41 configured by a plurality of document learning program groups acquired from the outside via a communication network. The document attribute detection estimation apparatus 10 acquires various document data from the document management DB of the file server 40 connected via the communication network, analyzes it using the document attribute learning data program 11, and is generated as a result. Are output to the attribute feature base DB 12-1. The document attribute learning program 11 and the attribute feature base DB 12-1 constitute a content analysis engine 16-1.

  Further, the document attribute detection estimation apparatus 10 refers to feature information of various attributes obtained from the attribute feature base DB 12-1 and a preset policy 14-1 using the document attribute estimation program 13-1. Based on these pieces of information, the document attribute detection / estimation apparatus 10 estimates an attribute adapted to the analysis target document 31-1 obtained from the multifunction machine 30-1 provided outside via the communication network, and estimates the attribute. The obtained analysis result 15 is output.

  The document attribute detection estimation apparatus 10 communicates warning information or the like indicating that the analysis target document 31-1 is output based on the policy information set in the policy 14-1, based on the result of the analysis result 15. The data is transmitted to the administrator terminal 20 that is a computer device connected to the outside via the network.

  The administrator terminal 20 can acquire and display the transmitted warning information from the document attribute detection estimation apparatus 10.

  FIG. 2 is a diagram illustrating a hardware configuration example of the electronic document attribute detection estimation apparatus 10. In FIG. 2, the electronic document attribute detection / estimation apparatus 10 is a computer apparatus, and includes an input device 51, an output device 52, a drive device 53, and an auxiliary storage device 54 that are connected to each other via a bus B. A memory device 55, an arithmetic processing device 56, and an interface device 57.

  The input device 51 includes a keyboard and a mouse, and is used to input various signals. The output device 52 is composed of a display device or the like, and is used for displaying various windows and data. The interface device 57 includes a modem, a LAN card, and the like, and is used for connecting to the network 3.

  The electronic document attribute detection / estimation apparatus 10 obtains various document data from the file server 40 or the like provided outside via the network 3 by the interface device 57. Further, the electronic document attribute detection / estimation apparatus 10 generates the attribute feature base DB 12-1 by the document attribute learning program 11 configured as the content analysis engine 16-1 from the acquired various document data.

  In addition, the electronic document attribute detection / estimation apparatus 10 acquires a document to be analyzed from the multifunction machine 30-1 provided outside via the network 3 by the interface device 57. Furthermore, the electronic document attribute detection estimation apparatus 10 transmits information indicating that the document to be analyzed has been output to the administrator terminal 20 provided outside via the network 3.

  The document attribute learning program 11 and the document attribute estimation program 13-1 according to the present embodiment are at least a part of various programs that control the electronic document attribute detection estimation apparatus 10. The document attribute learning program 11 and the document attribute estimation program 13-1 are provided by, for example, distribution of the storage medium 58 or downloading from the network 3. The storage medium 58 storing the document attribute learning program 11 and the document attribute estimation program 13-1 stores information optically, electrically, or magnetically like a CD-ROM, flexible disk, magneto-optical disk, or the like. It can be provided using various types of storage media such as a storage medium, a ROM (Read Only Memory), a semiconductor memory that electrically stores information such as a flash memory.

  When the storage medium 58 storing the document attribute learning program 11 and the document attribute estimation program 13-1 is set in the drive device 53, the document attribute learning program 11 and the document attribute estimation program 13-1 are read from the storage medium 58. It is installed in the auxiliary storage device 54 via the drive device 53.

  Further, the document attribute learning program 11 and the document attribute estimation program 13-1 may be downloaded from the network 3 and installed in the auxiliary storage device 54 via the interface device 57.

  The electronic document attribute detection estimation apparatus 10 stores the installed document attribute learning program 11 and the document attribute estimation program 13-1, and stores necessary files, data, and the like. The memory device 55 reads and stores the document attribute learning program 11 and the document attribute estimation program 13-1 from the auxiliary storage device 54 at the time of activation. Then, the arithmetic processing unit 56 performs generation processing of the attribute feature base DB 12-1 as the content analysis engine 16-1 according to the document attribute learning program 11 stored in the memory device 55, and the document attribute estimation program 13- 1, various processes as described later are realized.

  Next, a schematic configuration example of the second system will be described. Compared with the first system 101, the second system 102 has a configuration in which a multifunction peripheral is provided in the electronic document attribute detection estimation apparatus 10.

  FIG. 3 is a diagram illustrating a schematic configuration example of the second system 102. In the schematic configuration example shown in FIG. 3, the document attribute learning / estimation multifunction device 60 includes a document attribute learning program 11, an attribute feature base DB 12-2, a document attribute estimation program 13-2, a multifunction device 30-2, a copy / scan. A facsimile application 61 is used. Note that the multifunction machine 30-2 includes at least a scanner and a plotter, and enables a plurality of image processing such as a printer, a copy, and a scanner.

  The document attribute learning / estimating multifunction device 60 is connected to a file server 40 provided outside via a communication network, and acquires various types of document data from a document management DB of the file server 40 or the like. Further, the document attribute learning / estimation complex machine 60 uses the document attribute learning program 11 to request acquisition of various document data, analyze the acquired various document data, and use the generated various data as an attribute feature base. Output to DB12-2. The document attribute learning program 11 and the attribute feature base DB 12-2 constitute a content analysis engine 16-2.

  Further, the document attribute learning / estimation complex machine 60 uses the document attribute estimation program 13-2 to obtain the feature information of various attributes obtained from the attribute feature base DB 12-2 and the policy information set in the policy 14-1 in advance. refer. Then, the document attribute learning / estimating multifunction device 60 estimates an attribute to be applied to the analysis target document 31-2 obtained from the multifunction device 30-2 via the copy / scan / facsimile application 61, and an analysis obtained by the estimation. The result 15 is output.

  Note that the document attribute learning / estimation multifunction device 60 obtains information indicating that the analysis target document 31-2 has been output based on the policy information set in the policy 14-1 based on the analysis result 15. Warning information or the like is transmitted on the screen of the multifunction machine 30-2 to the user who has output the analysis target document 31-2 through the copy / scan / facsimile application 61.

  FIG. 4 is a diagram illustrating a hardware configuration example of the document attribute learning / estimation multifunction device 60. In FIG. 4, the document attribute learning / estimating multifunction device 60 is a computer device, and is an input device 51, an output device 52, a drive device 53, and an auxiliary storage device 54 that are connected to each other via a bus B. A memory device 55, an arithmetic processing device 56, and an interface device 57.

  The input device 51 includes a keyboard and a mouse, and is used to input various signals. The output device 52 is composed of a display device or the like, and is used for displaying various windows and data. The interface device 57 includes a modem, a LAN card, and the like, and is used for connecting to a network.

  The document attribute learning / estimating complex machine 60 acquires various document data from the file server 40 provided outside via the network 4 by the interface device 57. Further, the document attribute learning / estimating multifunction device 60 generates the attribute feature base DB 12-2 from the acquired various document data by the document attribute learning program 11 configured as the content analysis engine 16-2.

  The document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61 according to the present embodiment are at least a part of various programs that control the document attribute learning / estimation multifunction device 60. The document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61 are provided by, for example, distribution of the storage medium 58 or download from the network 4. The storage medium 58 storing the document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61 stores information optically such as a CD-ROM, a flexible disk, and a magneto-optical disk. The present invention can be provided using various types of storage media such as a storage medium that stores information electrically or magnetically, a semiconductor memory that stores information electrically such as a ROM and a flash memory.

  When the storage medium 58 storing the document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61 is set in the drive device 53, the document attribute learning program 11, the document attribute estimation program 13-2 and the copy / scan / facsimile application 61 are installed in the auxiliary storage device 14 from the storage medium 58 via the drive device 53. The document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61 downloaded from the network 4 are installed in the auxiliary storage device 54 via the interface device 57.

  The document attribute learning / estimation multifunction device 60 stores the installed document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61, and also stores necessary files, data, and the like. The memory device 55 reads and stores the document attribute learning program 11, the document attribute estimation program 13-2, and the copy / scan / facsimile application 61 from the auxiliary storage device 54 at the time of activation. Then, the arithmetic processing unit 56 performs generation processing of the attribute feature base DB 12-2 as the content analysis engine 16-2 according to the document attribute learning program 11 stored in the memory device 55, and the document attribute estimation program 13- According to 2, various processes as described later are realized.

  The copy / scan / facsimile application 61 is an application for executing the copy / scan / facsimile functions of the multifunction machine 30-2. The copy / scan / facsimile application 61 outputs a document to be analyzed (analysis target document 31-2) to the document attribute estimation program 13-2 in response to a request from the document attribute estimation program 13-2.

  Next, a function for generating learning data in the first functional configuration of the document attribute detection estimation process will be described. FIG. 5 is a diagram for explaining a function of generating learning data in the first functional configuration.

  As shown in FIG. 5, various types of documents that are divided into folders or the like are stored in a storage device such as a hard disk included in the document management device 40 as data serving as learning data. At this time, the electronic document attribute detection / estimation apparatus 10 uses the content analysis engine 16-1 or the like described above, for example, to acquire the accumulated various document information, extract document features for each folder, and extract the extracted folder. Each document feature and folder name are associated with each other and stored in, for example, the attribute feature base DB 12-1 of the content analysis engine 16-1.

  Here, the content analysis engine 16-1 will be described. The content analysis engine 16-1 acquires various types of document information for estimating the attributes of the analysis target document acquired by, for example, a multifunction peripheral from the document management apparatus 40 in which various types of documents are stored and managed for each attribute. .

  The attributes here are, for example, classification names, category names, folder names, and the like used when various information is managed and organized as documents. Generally, when various information is stored as a document, related documents are grouped, organized, and managed for each folder. A label name for classifying, organizing, and managing documents collected in each folder can be considered as an attribute of the related document. Therefore, an attribute can be considered as a label (folder) name for classifying, organizing, and managing related documents into, for example, classification, category, confidential level, type, and the like.

  The content analysis engine 16-1 acquires the data of the documents summarized for each folder in advance from the document management apparatus 40, extracts the characteristics of the documents summarized for each folder from the acquired data, and extracts the data for each extracted folder. The feature of the document is associated with a folder name, that is, a label (or attribute), and is generated as learning data.

  More specifically, as shown in FIG. 5, the document management apparatus 40 has a plurality of product themes (for example, product A theme, product B theme,...) In the learning folder cabinet. Existing.

  In addition, in the folder of the product A theme, documents related to the product A such as a plan and a design document of the product A are classified and managed (product A plan.doc, product A design.doc,...). . Similarly, in the folder of the product B theme, documents related to the product B such as a planning document and a design document of the product B are classified and managed (product B design.doc, product A design.doc,...). .

  The content analysis engine 16-1 acquires document information for each folder from the document management apparatus 40, and extracts and analyzes document features for each folder. Further, the content analysis engine 16-1 learns by associating the characteristics and attributes (folder name, label, etc.) for each folder obtained as an analysis result. In this way, the content analysis engine 16-1 determines the characteristics of the document for each of the product A theme and the product B theme when, for example, the documents of the product A theme and the product B theme are classified and managed for each folder. Extracting and associating the features of the product A theme and the product A theme, which are folder names, or the features of the product B theme and the product B theme, which are folder names, and storing them in, for example, the attribute feature base DB 12-1 To learn in advance. In addition, a well-known method can be used for the above-mentioned analysis / extraction / learning method.

  Next, the module configuration of the document attribute program 13-1 in the first functional configuration will be described with reference to FIG. As described below, the document attribute estimation program 13-1 receives an attribute estimation request for an analysis target document, performs attribute estimation, associates the attribute estimation result with the policy, and executes the policy based execution. It is a program to be performed.

  The document attribute estimation program 13-1 includes an attribute estimation request reception unit 70, an attribute estimation processing unit 71, a policy determination unit 72, and a policy setting unit 73.

  The attribute estimation request accepting unit 70 receives document data of the analysis target document 31-1 and context information (user information, operation) as material for policy determination from a client system such as the multifunction machine 30-1. The document data is output to the attribute estimation processing unit 71 described below, and the analysis result 15 is received from the attribute estimation processing unit 71. Thereafter, the analysis result 15 and the context information are output to the policy determination unit 72.

  The attribute estimation processing unit 71 is a control module that performs document attribute estimation processing. The attribute estimation processing unit 71 divides the document data into pieces by the following submodules, performs attribute estimation for each piece, performs threshold determination for each result, and outputs a final result. A specific processing procedure of the document attribute estimation process of the attribute estimation processing unit 71 will be described later.

  The attribute estimation processing unit 71 includes a piece division unit 71-1, a document analysis unit 71-2, and a similarity collection unit 71-3 described below.

  First, the piece dividing unit 71-1 divides a unit for determining the similarity of documents into finer parts, and determines which part of which document is similar to each part. Since we want to detect even some diversions, this process can be incorporated in the analysis process. The unit of this division is a piece. The division unit is, for example, a fixed length division of 500 character units, page unit division, paragraph unit division (extracted by identifying blank lines on the page image), one line unit division, sentence unit division, etc. Can be considered.

  Next, the document analysis unit 71-2 analyzes the divided document (text) by similar document search, and extracts the attribute and the similarity to the Nth place as the analysis result 15. A specific processing procedure of the similar document search processing will be described later.

  Next, the similarity collection unit 71-3 extracts the result of each divided document (piece), merges overlapping attributes, extracts the highest similarity for each attribute, and performs threshold determination. A specific processing procedure of the similarity collection processing will be described later.

  Next, the threshold management unit 71-4 manages the threshold list 80.

  The policy determination unit 72 refers to the preset policy 14-1, and based on the attribute of the analysis result 15 output from the attribute processing unit 71 and the context information such as user information and operation received from the client system. Make a policy decision. Also, actions such as this policy determination result and accompanying warning mail, log recording are executed. A specific example of policy information will be described later.

  The policy setting unit 73 is a module having a user interface for setting a policy. In the present embodiment, for example, the fine adjustment setting unit 73-1 which is a sub module has a function of setting the strength of the threshold on the policy setting screen. A specific example of the setting screen will be described later.

  The fine adjustment setting unit 73-1 sets the strength of the threshold for each attribute when setting the policy, and describes the fine adjustment level in the policy 14. At the time of attribute estimation, the threshold management unit 71-4 refers to this fine adjustment level.

  By having the above-described module, the document attribute estimation program 13-1 performs attribute estimation, executes the attribute estimation result in association with the policy, and executes based on the policy.

  Next, document attribute estimation processing performed by the attribute estimation processing unit 71 of the document attribute estimation program 13-1 described above will be described. FIG. 7 is a flowchart for explaining the document attribute estimation processing in the first functional configuration.

  The attribute estimation processing unit 71 first acquires the threshold list 80 from the threshold management unit 71-4 (S1) and receives the analysis target document 31-1 (S2).

  Next, the piece dividing unit 71-1 performs piece division of the analysis target document 31-1 acquired in the process of S <b> 2 (S <b> 3), and the document analyzing unit 71-2 performs similar documents for each divided document piece. Document analysis is performed by search processing (S4). In the document analysis of S4, the document analysis unit 71-2 cuts off at the lowest value in the threshold list 80 and outputs the analysis result X as the analysis result 15-1. A specific process of the similar document search process in S4 and a specific example of the analysis result X will be described later.

  Next, the similarity collection unit 71-3 acquires the content of the fine adjustment level 90 from the preset policy 14-1 (S5), collects the similarity result of the similarity collection process, and analyzes the result 15-3. As a result, the final result Y is output (S6). In the S6 process, the threshold value can be readjusted for each attribute at the fine adjustment level 90. A specific process of the similarity collection process in S6 and a specific example of the final result Y will be described later.

  Here, an example of how to determine the similarity threshold used in the present invention will be described. FIG. 8 is a diagram showing the correctness / error distribution of similarity with respect to the similarity threshold.

  When the threshold value of similarity is set, for example, since the value and distribution of similarity are different depending on a preset similar document search classifier or the like, an experiment is performed in advance with a sample data set. As shown in FIG. 8A, due to sampling of similar document search, there is a correct answer distribution of similarity hitting an expected attribute and an incorrect answer distribution of similarity hitting an unexpected attribute. Therefore, for example, the basic threshold value positioned in the graph (3) in FIG. 8A is set so that there are many correct answers and fewer incorrect answers.

  Next, as described above, the threshold is adjusted for each attribute in order to actually perform operations according to various situations. This threshold value adjustment is defined in a policy 14 set in advance, for example, a threshold value adjustment list 78 as shown in FIG. 8B can be created.

  As shown in FIG. 8B, the threshold adjustment list 78 includes “higher (80)”, “slightly higher (75)”, “normal (70)”, and “slightly lower (65)” indicating attribute importance. , “Lower (60)” or the like is used to specify only the step value in the policy 14. By specifying the threshold adjustment list 78 in this way, for the above-described product A theme, for example, when “lower” is specified, more information that flows out to information on similar product A theme is monitored. be able to. In addition, it is possible to avoid the complexity of operation defined in the policy 14.

  An example of the preset threshold list 80 will be described in relation to the attribute importance shown in the threshold adjustment list 78. FIG. 9 is a diagram illustrating an example of the threshold list 80. As illustrated in FIG. 9, the threshold list 80 is a list in which thresholds are added and the level is set to “+2 (80)”, “+1 (75)”, “0 (70)”, “−1”. (65) "and" -2 (60) ". By providing a minimum threshold value in the threshold value list 80 (for example, “60” in FIG. 9), it can also be used as a cut-off in a similar document search process described later.

  Further, an example of the fine adjustment level 90 will be described. FIG. 10 is a diagram illustrating an example of the fine adjustment level. The fine adjustment level 90 shown in FIG. 10 is a level for defining a threshold value for each attribute in the policy 14-1, for example, “(attribute) A → (level) −1” “(attribute) B → (Level) +2 ”,“ (attribute) C → (level) +1 ”, etc. The fine adjustment level 90 corresponds to the threshold adjustment list 78 and the threshold list 80. Therefore, for example, when the fine adjustment level 90 “attribute A → level (−1)” is defined, the threshold value of the attribute A can be changed to “65” corresponding to the threshold list 80 shown in FIG. With the fine adjustment level 90, it is possible to easily change the threshold according to the importance of the attribute for each attribute. The correspondence relationship between the threshold adjustment list 78, the threshold list 80, and the fine adjustment level 90 will be described later.

  In the setting of the threshold value in the present embodiment, five levels are set. However, in the present invention, the present invention is not limited to this, and a plurality of levels may be set.

  Next, a similar document search process performed by the document analysis unit 71-2 of the document attribute estimation program 13-1 described above will be described. FIG. 11 is a flowchart for explaining similar document search processing in the first functional configuration.

  In the similar document search process shown in FIG. 11, first, the document analysis unit 71-2 acquires the document piece divided by the piece division unit 71-1, and refers to the feature of the document for each attribute from the attribute feature base DB 12-1. Then, a similar document search is performed for each acquired document piece, and document analysis is performed (S10). The similar document search may be performed using a known technique, but for example, a method described in JP-A-2006-185153 can also be used.

  Next, the document analysis unit 71-2 determines whether document analysis has been completed for all document pieces (S11). If it is determined that the document analysis has ended, an analysis result X is generated (S12), and the process ends. If it is determined that the document piece analysis has not ended, one document piece is acquired (S13).

  Next, the document analysis unit 71-2 performs a similar document search for one piece acquired by the processing in S13, and acquires the similarity with similar documents up to rank N (S14). The rank N can be determined in advance as a system parameter.

  Next, the document analysis unit 71-2 determines whether all N ranks acquired in S14 have been checked (S15). If it is determined that all N ranks have not been checked, a rank record is acquired (S16). Next, with respect to the similarity of the rank record, the threshold list 80 is acquired from the threshold management unit 71-4 and compared with the minimum threshold to determine whether the rank record has a similarity equal to or higher than the minimum threshold (S17). .

  When the document analysis unit 71-2 determines that the similarity of the rank record is equal to or higher than the minimum threshold, the document analysis unit 71-2 acquires an attribute corresponding to the similar document (S18). Further, the document analysis unit 71-2 outputs and stores the analysis result X indicating the piece ID, rank, attribute, and similarity of the rank record in the analysis result X list (S19).

Further, when the document analysis unit 71-2 determines that the similarity of the rank record is smaller than the minimum threshold in the process of S17, the document analysis unit 71-2 cuts off the record (S17). As a result, the data to be processed can be reduced and the efficiency can be improved. If the document analysis unit 71-2 determines that all N ranks acquired in S14 have been examined in the process of S15, the process of S11 is performed. Return to.

  Next, the similarity collection process performed by the similarity collection unit 71-3 of the document attribute estimation program 13-1 described above will be described. FIG. 12 is a flowchart for explaining similarity collection processing in the first functional configuration.

  In the similarity collection process shown in FIG. 12, the similarity collection unit 71-3 acquires the analysis result X output by the document analysis unit 71-2, merges attributes with respect to the acquired analysis result X, and the similarity is The highest value is left for each attribute (S20). Therefore, in the process of S20, the similarity collection unit 71-3 outputs the intermediate result Y ′ as the obtained analysis result 15-2.

  Next, the similarity collection unit 71-3 performs threshold fine adjustment determination for each of the output intermediate results Y ′ (S21). Next, it is determined whether the threshold fine adjustment determination for each intermediate result Y ′ has been completed (S22). When it is determined that the threshold fine adjustment determination has been completed, the similarity collection unit 71-3 generates a final result Y as the analysis result 15-3 (S23) and ends the process.

  Next, when it is determined that the threshold fine adjustment determination has not ended, the similarity collection unit 71-3 acquires one record of the intermediate result Y ′ (S24). Next, the similarity collection unit 71-3 acquires the fine adjustment level 90 for each attribute from the policy 14 for the record acquired in S24 processing (S25). Next, the similarity collection unit 71-3 refers to the threshold list 80 of the threshold management unit 71-4, and acquires a threshold level corresponding to the fine adjustment level 90 for each attribute acquired in S25 processing ( S26). Next, the similarity collection unit 71-3 compares the threshold for each attribute acquired in S26 with the similarity for each attribute of the record acquired in S24, and determines whether the record has a similarity greater than or equal to the threshold. Judgment is made (S27).

  When the similarity collection unit 71-3 determines that the similarity of the record acquired in S24 processing is equal to or greater than the threshold, the similarity collection unit 71-3 stores the final result Y for the list for the attribute and similarity of the record (S28).

  Furthermore, when the similarity collection unit 71-3 determines in the process of S27 that the similarity of the record acquired in the process S24 is smaller than the threshold, the record is cut off (S27).

  Next, the analysis result X, the intermediate result Y ′, the analysis result, and Y as the analysis result 15 generated by the above-described process will be described. FIG. 13 is a diagram illustrating an example of the analysis result X, the intermediate result Y ′, and the final result Y generated by the above-described processing.

  FIG. 13A is an example of the analysis result X generated by the similar document search process. The analysis result X shows the divided piece IDs and ranks, and the rank record shows similar attributes and similarities. For example, the piece ID “1” has the first place “A, 70”, and the piece ID “1” is 70% similar to the attribute A, and it is understood that there is no one that falls under the second place. . The piece ID “2” has the first place “C, 70”, the second place “A, 65”, the third place “B, 60”, the piece ID “2” is 70% similar to the attribute C, and the attribute A It can be seen that it is 65% similar to attribute B and 60% similar to attribute B.

  The document analysis result X is obtained by performing a similar search in S14 of the similar document search process, and then comparing with similar documents up to a preset rank N (“third” in FIG. 13A). Since the similarity is set to be acquired, the fourth and lower similarities are cut off and do not appear on the list. In S17 of the similar document search process, with reference to the threshold list 80, those with a similarity equal to or lower than the minimum threshold (“60” in FIG. 13A) are set to be cut off. No one with a similarity of 60 or less appears.

  FIG. 13B is an example of the intermediate result Y ′ generated by the similarity collection process. The intermediate result Y ′ is generated by merging the attributes of the analysis result X, and the similarity is the process of leaving the highest value for each attribute. Therefore, the merged attribute and the similarity of the highest value for each attribute are indicated. Yes. Therefore, in the intermediate result Y ′, the merged attribute of the analysis result X and the records of “B, 90”, “A, 70”, “C, 70” which are the similarity of the highest value in the attribute are indicated. ing.

  FIG. 13C is an example of the final result Y generated by the similarity collection process. The final result Y indicates “B, 90” and “A, 70”. Although “C, 70” is shown in the list of intermediate results Y shown in FIG. 13B, the threshold value of attribute C is obtained by acquiring the fine adjustment level 90 for each attribute in S25 of the similarity collection process. Becomes “75”. Therefore, since “C, 70” is shaken off by the attribute-specific determination in S27, it does not remain in the final result Y list.

  Here, an example of the policy 14-1 in the first functional configuration will be described. FIG. 14 is a diagram illustrating an example of the policy 14-1.

  As shown in FIG. 14A, for example, a label as an attribute, a threshold determination standard indicating a fine adjustment level for each attribute, an operable target person, an operation content, and an operation are executed in the policy 14-1. The contents of later warnings are described.

  Specifically, as shown in FIGS. 14A and 15A, it is shown that “the threshold judgment criterion is“ lower ”for the label“ product A ”” ”in FIG. The description 14a in FIG. 15B shows that “other than the related person is a violation if copying, scanning, printing, and faxing”. Based on this policy 14-1, as shown in FIG. 14 (B), when the determination result indicating “Copied to Mr. X (not related person)” is obtained, a warning is given to the manager of the product A An email is issued.

  A description example of the policy 14-1 will be described. FIG. 15 is a diagram illustrating an example in which the above-described policy example is expressed in XML. As shown in FIG. 15B, in the XML 14-2 of the policy 14-1, the above description 14a includes <Resource type = “LABEL”> <value> PROJECT-A </ value> <ThresholdLevel> LOWER <// ThresholdLevel>.

  Further, the description 14b described above is indicated as <User> <Operation id = “ALL” /> other than those related to <User>.

  In accordance with this description, the policy 14-1 “When a violation is detected, a warning mail is sent to the administrator of the product A” is described in the description 14c <Oblation id = “ALARM_MAIL”> <address> aaa @ in FIG. xxx. xxx. xx </ address> is indicated.

  Next, an example of the policy setting screen described above will be described. FIG. 16 is a diagram illustrating an example of a policy setting screen. 16A shows the layout configuration of the policy setting screen 100, and FIGS. 16B and 16C show examples of interfaces for adjusting the fine adjustment level.

  For example, the policy setting screen 100 shown in FIG. 16A includes a “classification label” for setting an attribute, an “operation performer” for setting an operable target person, an “operation type” for setting operation details, and an operation. Each setting area includes a “warning method” for setting a later action and an “attribute importance” for adjusting a fine adjustment level. After the setting, for example, by selecting a registration button, the setting items are registered and defined in the policy 14-1.

  As shown in FIG. 16A, for the attribute importance 100-2, one of “high”, “slightly high”, “normal”, “slightly low”, and “low” is selected by a radio button. However, the present invention is not limited to this. For example, the slider bar format 101 shown in FIG. 16B or the dial format 102 shown in FIG. Each time the attribute importance 100-2 is changed by the person who manages the attributes of the document, the fine adjustment list 90 of 14-1 is updated.

  Further, “higher”, “slightly higher”, “normal”, “slightly lower”, and “lower” of the attribute importance 100-2 are the threshold adjustment list 78 of the policy 14-1 described in FIG. 8B. “High (80)”, “Slightly high (75)”, “Normal (70)”, “Slightly low (65)”, and “Low (60)” are set. Further, the threshold adjustment list 78 and the threshold list 80 are related in importance, and the threshold list 80 and the fine adjustment list 90 are related in level.

  Therefore, for example, when the attribute importance 100-2 “normal” is selected for the classification level “product A theme” on the policy setting screen 100 shown in FIG. 16, “normal” is displayed in the threshold adjustment list 78 of the policy 14-1. (70) ". Since “threshold 70 → level 0” is set in the threshold list 80, the fine adjustment level 90 of the policy 14-1 is defined as the attribute “product A theme” → level “0”. Therefore, for example, in the threshold fine adjustment determination in the above-described first and second functional configurations described later, when the threshold value of the level corresponding to the fine adjustment level 90 is acquired, the threshold value of the attribute “product A theme” is “70”. Is obtained.

  As described above, the attribute importance 100-2 is provided on the policy setting screen 100, thereby facilitating the setting of the threshold according to the importance of the attribute. Accordingly, a person who manages the attribute of the document stored in advance can easily change the setting of the threshold according to the importance of the attribute that varies depending on the time.

  By making it possible to easily adjust the threshold value for each attribute, it is possible to adjust the threshold value according to the importance level of the attribute when performing a similarity search for estimating the attribute of a document that is not associated with an attribute. The adjustment of the threshold value makes it possible to easily change the extraction range of similar documents. Therefore, the person who manages the attribute of the document sets the threshold value low for the attribute with increased importance and widens the extraction degree range of similar documents. On the other hand, it is possible to closely monitor the outflow. On the other hand, for attributes with low importance, the threshold is set high and the extraction range of similar documents is narrowed so that only documents with high similarity are monitored. be able to. This makes it possible to perform a similarity search according to the importance of attributes that vary from day to day.

  Next, an outline of processing in the first functional configuration of document attribute detection estimation processing will be described. FIG. 17 is a diagram for explaining the flow of the processing outline in the first functional configuration.

  According to FIG. 17, when the document attribute estimation program 13-1 receives an analysis target document 31-1 whose attribute is unknown by the multifunction machine 30-1 or the like, the document attribute estimation program 13-1 acquires document data of the analysis target document 31-1. Next, the document attribute estimation program 13-1 analyzes the “characteristics of the product A theme document” and “characteristics of the product B theme document” which are document characteristics for each attribute learned in advance by the content analysis engine 16-1. Similar search is performed on the document data of the document 31-1, and the document data is analyzed. From this analysis, an intermediate result Y ′ is generated. Next, the policy 14-1 is referred to, and the fine adjustment level 90 for each attribute is acquired. The threshold for each attribute after obtaining the fine adjustment level 90 is compared with the similarity of the document of the intermediate result Y ′, and the final result Y is generated.

  For example, according to FIG. 17, “product A: low (similarity 60 or higher)” and “product B: slightly higher (similarity 75 or higher)” are set in the policy 14-1. Since the intermediate result Y ′ is “product A theme (similarity 65%)” and “product B theme (similarity 70%)”, “product B theme (similarity 70%) is determined by similarity determination for each attribute. ) ”Was cut off, and the final result Y was“ product A (similarity 65%) ”. As a result, the analysis target document 31-1 is estimated as “product A (similarity 65%)”.

  As described above, according to the first functional configuration, when a document with no attribute association is printed on a copier or the like, document data (information) with no attribute association is stored in advance. Similar analysis is performed with the document features of the attribute. As a result, it is possible to estimate an attribute of a document having a high similarity to a document that is not associated with an attribute as an attribute of a document that is not associated with an attribute, and to apply a policy defined for the attribute. When estimating the attribute of a document that is not associated with an attribute, it is possible to adjust the similarity extraction range for each attribute by changing the threshold according to the importance of the attribute stored in advance. . Thereby, it is possible to estimate the attribute of the document in accordance with the importance of the attribute that varies depending on the situation. Thus, for example, the similarity extraction range is narrowed by setting the similarity threshold value of the attribute having a low importance level to be narrow, and the similarity threshold value of the attribute having a high importance level is set to be low. It is possible to appropriately estimate the attribute of a document that is not associated with the attribute, and to perform an appropriate action suitable for the operation based on the attribute policy applied to the document.

  Next, a function for generating learning data in the second functional configuration of the document attribute detection estimation process will be described. FIG. 18 is a diagram for explaining a function of generating learning data in the second functional configuration.

  As shown in FIG. 18, various types of documents that are divided into folders or the like are stored in a storage device such as a hard disk included in the document management device 40 as data that is the basis of learning data. At this time, the document attribute learning / estimating multifunction device 60 acquires various types of document information using, for example, the content analysis engine 16-2 and extracts the document features for each folder. The document features and folder names for each folder are associated with the folder name, that is, the attribute, and stored in the attribute feature base DB 12-2, for example. Furthermore, at this time, the fine adjustment level 90 (lower, slightly higher, etc.) at the time of similarity determination for each attribute is also accumulated. For this analysis method, learning method, etc., known methods are used.

  Next, the module configuration of the document attribute estimation program 13-2 in the second functional configuration will be described with reference to FIG. The document attribute estimation program 13-2 is different from the document attribute estimation program 13-1 described in the first functional configuration in that the policy setting unit 73 has no fine adjustment setting unit 73-1, and the threshold management unit 71-4 has a policy. The difference is that 14-2 is not referenced. Other functions with the same reference numerals are the same as the functions described in the first functional configuration described above, and thus the description thereof is omitted.

  Next, document attribute estimation processing performed by the attribute processing unit 71 of the document attribute estimation program 13-2 described above will be described. FIG. 20 is a flowchart for explaining document attribute estimation processing in the second functional configuration.

  In the document attribute estimation process shown in FIG. 20, the attribute estimation processing unit 71 first acquires the threshold list 80 from the threshold management unit 71-4 (S30), and accepts the analysis target document 31-2 (S31).

  Next, the piece dividing unit 71-1 performs piece division of the analysis target document acquired in S31 processing (S32), and the document analyzing unit 71-2 acquires the fine adjustment level 90 from the attribute feature base DB (S33). ).

  Next, the document analysis unit 71-2 performs document analysis for each divided document piece by a similar document search process (S34). In S34 processing, the document analysis unit 71-2 refers to the threshold value of the fine adjustment level 90 and outputs the analysis result Y ″ as the analysis result 15-4. A specific processing procedure of the similar document search processing in S34 and a specific example of the analysis result Y ″ will be described later.

  Next, the similarity collection unit 71-3 collects the similarity results of the similar document search process in the S34 process, and outputs the final result Y as the analysis result 15-3 (S35). A specific processing procedure of the similar result collection processing in S35 and a specific example of the final result Y will be described later.

  Next, a similar document search process performed by the document analysis unit 71-2 of the document attribute estimation program 13-2 described above will be described. FIG. 21 is a flowchart for explaining similar document search processing in the second functional configuration.

  In the similar document search process shown in FIG. 21, first, the document analysis unit 71-2 acquires the document piece divided by the piece division unit 71-1, and obtains the document features for each attribute from the attribute feature base DB 12-2. The similar document search is performed for each acquired document piece by referring to the document analysis (S40). The similar document search may be performed using a known technique, but for example, a method described in JP-A-2006-185153 can also be used.

  Next, the document analysis unit 71-2 determines whether document analysis has been completed for all document pieces (S41). If it is determined that the document analysis has been completed, an analysis result Y ″ is generated (S42), and the process ends. If it is determined that the document piece analysis has not ended, one document piece is acquired (S43).

  Next, the document analysis unit 71-2 performs a similar document search for one piece acquired by the processing of S43, and acquires the degree of similarity with similar documents up to rank N (S44). The rank N can be determined in advance as a system parameter.

  Next, the document analysis unit 71-2 determines whether all N ranks acquired in S44 have been examined (S45). If it is determined that all N ranks have not been checked, a rank record is acquired (S46). Next, with respect to the similarity of the rank record, the threshold list 80 is acquired from the threshold management unit 71-4 and compared with the minimum threshold to determine whether the rank record has a similarity equal to or higher than the minimum threshold (S47). .

  When the document analysis unit 71-2 determines that the similarity of the rank record is equal to or higher than the minimum threshold, the document analysis unit 71-2 acquires an attribute corresponding to the similar document (S48). Next, the document analysis unit 71-2 acquires the fine adjustment level 90 corresponding to the attribute for the rank record from the attribute feature base DB 12-2 (S49). Next, the document analysis unit 71-2 refers to the threshold value list 80, and acquires a threshold level corresponding to the fine adjustment level 90 for each attribute of the rank record (S50).

  Next, the document analysis unit 71-2 compares the threshold acquired in S50 with the similarity for each attribute of the rank record acquired in S46, and determines whether the rank record has a similarity greater than or equal to the threshold. (S51).

  If the document analysis unit 71-2 determines that the similarity of the rank record is equal to or greater than the threshold, the analysis result Y '' indicating the piece ID, rank, attribute, and similarity of the rank record is displayed as the analysis result Y ''. The data is output to and saved in the list (S52).

Furthermore, when the document analysis unit 71-2 determines that the similarity of the rank record is smaller than the minimum threshold in the process of S47, the document analysis unit 71-2 cuts off the rank record (S47). As a result, the data to be processed can be reduced and efficiency can be increased. Furthermore, when the document analysis unit 71-2 determines in the processing of S51 that the similarity of the rank record is smaller than the threshold, the rank record Is cut off (S51).

  If it is determined in the process of S45 that all N ranks acquired in S44 have been examined, the process returns to S41.

  Next, the analysis result Y ″ generated by the above process will be described. FIG. 22 is a diagram illustrating an example of the analysis result Y ″ generated by the above-described processing.

  As shown in FIG. 22, the analysis result Y ″ indicates the divided piece ID and rank, and the rank record indicates similar attribute values and similarities. For example, the divided piece ID “1” has the first place “A, 70”, the piece ID “1” is 70% similar to the attribute A, and there is no one that falls under the second place. I understand. Further, it is understood that there is no corresponding piece ID “2”.

  The document analysis result Y ″ is set so as to obtain the similarity with similar documents up to rank N (“3” in FIG. 22) preset in S44 of the similar document search process. Therefore, the 4th or lower similarity is cut off and does not appear on the list. In S47 of the similar document grinding process, with reference to the threshold list 80, those with a similarity equal to or lower than the minimum threshold (“60” in FIG. 22) are set to be cut off. Those with a degree of 60 or less will not come out. Further, the fine adjustment level 90 corresponding to the attribute value is acquired in S49, the threshold value of the level corresponding to the fine adjustment level 90 is acquired in S50, and the similarity of similar documents is greater than or equal to this threshold value in S51. Remain on the list.

  Next, a similarity result collection process performed by the similarity collection unit 71-3 of the document attribute estimation program 13-2 described above will be described. FIG. 23 is a flowchart for explaining similar result collection processing in the second functional configuration.

  In the similarity collection process illustrated in FIG. 23, the similarity collection unit 71-3 acquires the analysis result Y ″ generated by the document analysis unit 71-2, merges the attributes of the analysis result Y ″, and the similarity Leaves the highest value for each attribute (S60). Next, a final result Y list is generated and stored from the process of S60 (S61).

  Next, the final result Y generated by the above process will be described. FIG. 24 is a diagram illustrating an example of the final result Y generated by the above-described processing.

  The final result Y shown in FIG. 24 is generated by the process of merging the attributes of the analysis result Y ″ and the similarity level leaves the highest value for each attribute. Therefore, the merged attribute of the analysis result Y ″ and its Records of “B, 90” and “A, 70” which are the similarity of the highest value in the attribute are shown.

  Next, an outline of processing in the second functional configuration of document attribute detection estimation processing will be described. In this configuration, when the content analysis engine 16-2 learns the attributes of the electronic document stored in the file server document management DB 40, the content analysis engine 16-2 also learns the importance of the attribute for each attribute of the electronic document. It is an example. FIG. 25 is a diagram for explaining the flow of the processing outline in the second functional configuration.

  According to FIG. 25, when the document attribute estimation program 13-2 receives an analysis target document 31-2 whose attribute (label) is unknown by the multifunction machine 30-2 or the like, the document attribute estimation program 13-2 acquires the document data of the analysis target document 31-2. To do. Next, the document attribute estimation program 13-2 reads “features of the product A theme document” and “features of the product B theme document”, which are document features for each attribute learned by the content analysis engine 16-2, and details for each attribute. An adjustment level 90 is acquired. Next, the similarity search is performed between the document data of the analysis target document 31-2 and the document feature for each attribute, and analysis is performed. At this time, since the similarity of the document data is compared and determined with reference to the fine adjustment level 90 for each attribute, the final result Y that takes into account the degree of the threshold for each attribute is generated. From this final result Y, the document attribute estimation program 13-2 can determine attributes (including similarity) for the analysis target document 31-2.

For example, according to FIG. 25, “features of product A theme document, threshold: lower” and “features of product B theme document, threshold: slightly higher” are stored as data in the content analysis engine 16-2. The fine adjustment level 90 is set to “lower (similarity 60 or higher)” “slightly higher (similarity 75 or higher)” or the like for each attribute. The document attribute estimation program 13-2 performs similarity search and analysis using the analysis target document data of the analysis target document 31-2, the content analysis engine 16-2, and the threshold list 80, and obtains the final result Y (attribute B (similarity Degree 90%)). As a result, the document attribute estimation program 13-2 estimates that the analysis target document 31-2 is “attribute B (similarity 90%)”.

  Here, an example of the policy 14-2 in the second functional configuration will be described with reference to the drawings. FIG. 26 is a diagram illustrating an example of the policy 14-2.

  As shown in FIG. 26A, the policy 14-2 describes, for example, contents such as a label as an attribute, an operable target person, an operation content, a warning after the operation is executed, and the like. Specifically, as shown in FIG. 26, “For a document whose label is“ product A ”, if a person other than the related party copies, scans, prints, or faxes it is“ violation ””, “ “Send a warning email to the administrator of A”. Based on this policy 14-2, as shown in FIG. 26 (B), if a determination result indicating “Copied to Mr. X (not a related person)” is obtained, a warning is given to the administrator of product A An email is issued.

  As described above, according to the second functional configuration, when a document with no attribute association is printed on a copier or the like, document data (information) with no attribute association is stored in advance. Similar analysis is performed with the document features of the attribute. As a result, it is possible to estimate an attribute of a document having a high similarity to a document that is not associated with an attribute as an attribute of a document that is not associated with an attribute, and to apply a policy defined for the attribute. When estimating the attribute of a document that is not associated with an attribute, it is possible to adjust the similarity extraction range for each attribute by setting a threshold according to the importance of the attribute stored in advance. . Thereby, it is possible to estimate the attribute of the document in accordance with the importance of the attribute that varies depending on the situation. Thus, for example, the similarity extraction range is narrowed by setting the similarity threshold value of the attribute having a low importance level to be narrow, and the similarity threshold value of the attribute having a high importance level is set to be low. It is possible to appropriately estimate the attribute of a document that is not associated with the attribute, and to perform an appropriate action suitable for the operation based on the attribute policy applied to the document.

  The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to such specific embodiments, and various modifications can be made within the scope of the gist of the present invention described in the claims. Can be changed.

It is a figure which shows the example of schematic structure of a 1st system. It is a figure which shows the hardware structural example of an electronic document attribute detection estimation apparatus. It is a figure which shows the schematic structural example of a 2nd system. FIG. 2 is a diagram illustrating an example of a hardware configuration of a document attribute learning / estimating multifunction peripheral. It is a figure for demonstrating the function which the learning data in a 1st function structure produces | generates. It is a figure for demonstrating the module structure of the document attribute program in a 1st function structure. It is a flowchart for demonstrating the document attribute estimation process in a 1st function structure. It is a figure which shows the correct answer / error distribution of the similarity with respect to the threshold value of similarity. It is a figure which shows an example of a threshold value list. It is a figure which shows an example of a fine adjustment level. It is a flowchart for demonstrating the similar document search process in a 1st function structure. It is a flowchart for demonstrating the similarity collection process in a 1st function structure. FIG. 6 is a diagram illustrating an example of an analysis result X, an intermediate result Y ′, and a final result Y. It is a figure which shows an example of a policy. It is a figure which shows an example which described the policy example in XML. It is a figure which shows an example of a policy setting screen etc. It is a figure for demonstrating the flow of the process outline | summary in a 1st functional structure. It is a figure for demonstrating the function which the learning data in a 2nd function structure produces | generates. It is a figure for demonstrating the module structure of the document attribute estimation program in a 2nd function structure. It is a flowchart for demonstrating the document attribute estimation process in a 2nd function structure. It is a flowchart for demonstrating the similar document search process in a 2nd function structure. It is a figure which shows an example about analysis result Y ''. It is a flowchart for demonstrating the similar result collection process in a 2nd function structure. It is a figure which shows an example about the final result Y. It is a figure for demonstrating the flow of the process outline | summary in a 2nd function structure. It is a figure which shows an example of a policy.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Electronic document attribute detection estimation apparatus 11 Document attribute learning program 12 Attribute feature DB
13 Document attribute estimation program 14 Policy 16 Content analysis engine 20 Administrator terminal 30 MFP 40 File server 51 Input device 52 Output device 53 Drive device 54 Auxiliary storage device 55 Memory device 56 Arithmetic processing device 57 Interface device 58 Storage device 60 Document attribute Learning / Estimating MFP 61 Copy / Scan / Facsimile Application 62 File Server 70 Attribute Determination Request Accepting Unit 71 Attribute Estimation Processing Unit 72 Policy Determination Unit 73 Policy Setting Unit 78 Threshold Adjustment List 80 Threshold List 90 Fine Adjustment List 100 Policy Setting Screen B bus

Claims (14)

An electronic document attribute detection estimation method for detecting and estimating an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance,
A search procedure for searching for an electronic document having a high similarity to the document among the electronic documents stored in the database;
For each attribute of the electronic document searched by the search procedure, a comparison procedure that compares the similarity of the document to the attribute and a threshold corresponding to the importance adjusted for the attribute;
An electronic document attribute detection estimation method comprising: an estimation procedure for estimating an attribute of the electronic document as an attribute of the document when a similarity for each attribute of the retrieved electronic document exceeds the threshold. The comparison procedure is:
For each attribute of the electronic document, the importance is acquired from the adjustment list in which the attribute is associated with the adjusted importance, and the importance is obtained from the threshold list in which the importance is associated with the threshold. The electronic document attribute detection estimation method according to claim 1, wherein a threshold value is acquired and the similarity is compared.   3. The electronic document attribute detection estimation method according to claim 2, wherein the adjustment list is updated in accordance with an adjustment of importance of the attribute of the electronic document by a user. The estimation procedure includes:
The information indicating that the document is output is output to an administrator based on a policy set in advance for each attribute when the document exceeding the threshold is output. The electronic document attribute detection estimation method according to any one of 1 to 3.   3. The electronic document attribute detection estimation according to claim 2, wherein the adjustment list is a list in which importance acquired when learning the attributes of the electronic document stored in the database is associated with each attribute. Method. An electronic document attribute detection / estimation device that detects and estimates an attribute corresponding to a document that is not associated with an attribute from an electronic document database stored in advance,
Search means for searching for an electronic document having a high similarity to the document among the electronic documents stored in the database;
Comparison means for comparing the similarity of the document to the attribute for each attribute of the electronic document searched by the search means and a threshold corresponding to the importance adjusted for the attribute;
An electronic document attribute detection / estimation device that executes estimation means for estimating the attribute of the electronic document as the attribute of the document when the similarity for each attribute of the searched electronic document exceeds the threshold value . The comparison means includes
For each attribute of the electronic document, the importance is acquired from the adjustment list in which the attribute is associated with the adjusted importance, and the importance is obtained from the threshold list in which the importance is associated with the threshold. The electronic document attribute detection estimation apparatus according to claim 6, wherein a threshold value is acquired and compared with the similarity.   8. The electronic document attribute detection estimation apparatus according to claim 7, wherein the adjustment list is updated in accordance with an adjustment of importance of the attribute of the electronic document by a user. The estimation means includes
The information indicating that the document is output is output to an administrator based on a policy set in advance for each attribute when the document exceeding the threshold is output. The electronic document attribute detection estimation apparatus according to any one of 6 to 8.   The electronic document attribute detection estimation according to claim 7, wherein the adjustment list is a list in which importance acquired when learning the attributes of the electronic document stored in the database is associated with each attribute. apparatus. An electronic document attribute detection and estimation program for detecting and estimating an attribute corresponding to a document that is not associated with an attribute from an electronic document database accumulated in advance,
Computer
Search means for searching for an electronic document having a high similarity to the document among the electronic documents stored in the database;
Comparison means for comparing the similarity of the document to the attribute for each attribute of the electronic document searched by the search means and a threshold corresponding to the importance adjusted for the attribute;
An electronic document attribute detection / estimation program which functions as an estimation means for estimating an attribute of the electronic document as an attribute of the document when the similarity for each attribute of the retrieved electronic document exceeds the threshold. The comparison means includes
For each attribute of the electronic document, the importance is acquired from the adjustment list in which the attribute is associated with the adjusted importance, and the importance is obtained from the threshold list in which the importance is associated with the threshold. 12. The electronic document attribute detection estimation program according to claim 11, wherein a threshold value is acquired and compared with the similarity. A computer-readable storage medium storing an electronic document attribute detection estimation program for detecting and estimating an attribute corresponding to a document that is not associated with an attribute from an electronic document database accumulated in advance,
Computer
Search means for searching for an electronic document having a high similarity to the document among the electronic documents stored in the database;
For each attribute of the electronic document searched by the search means, comparing means for comparing the similarity of the document to the attribute and a threshold corresponding to the importance adjusted for the attribute;
A computer-readable storage medium storing a program for functioning as estimation means for estimating the attribute of the electronic document as the attribute of the document when the similarity for each attribute of the searched electronic document exceeds the threshold . The comparison means includes
For each attribute of the electronic document, the importance is acquired from the adjustment list in which the attribute is associated with the adjusted importance, and the importance is obtained from the threshold list in which the importance is associated with the threshold. The computer-readable storage medium according to claim 13, wherein a threshold value is acquired and compared with the similarity.

Images