JPH1145247A - Document classification device, storage medium for storing document classification program and document classification method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform a more precise classification of an object document by using both classification results by a manual classification and an automatic classification. SOLUTION: An evaluation function by weighting or the like obtained from the rate of correct answer for the past classification is made to be a database by each classification personnel and, at the same time, a typical document for characterizing the classification by each classification is prepared in advance. Then, a scoring of a manual classification of each classification is executed from a classification result by the classification personnel (manual) and the evaluation function of the object document. Also, the degree of similarity of the object document and the typical document is calculated and the scoring of an automatic classification of each classification is performed by using this degree of similarity. The classification of the highest value obtained by summing up both of these scores by each classification is regarded as the final classification result. In this way, by fusing the manual classification and the automatic classification, a more precise classification result can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document classification device, a storage medium storing a document classification program, and a document classification method, and more particularly, to an improvement in classification accuracy of an acquired target document.

[0002]

2. Description of the Related Art When a document is filed, distributed electronically, or stored in a storage medium, the target document may be classified into a predetermined category. Conventionally, when classifying a target document in this way, the person in charge of the classifier performs a manual classification after reading the target document, or automatically analyzes the target document by analyzing the document content using a computer system. Or categorized.

[0003]

However, in the conventional manual classification of documents, classification is not always accurate. On the other hand, a computer system can classify a large number of documents at high speed, but this classification is not always accurate. In addition, since the conventional manual classification and the automatic classification are completely different in classification form, a seamless usability combining the two is not realized.

[0004] The present invention has been made to solve such problems of the prior art, and uses both classification results of manual classification and automatic classification to classify a target document with higher accuracy. It is a first object of the present invention to provide a document classification apparatus that can perform the document classification. Further, the present invention provides a storage medium on which a document classification program capable of performing more accurate classification on a target document by using both classification results of manual classification and automatic classification is recorded. This is the first purpose.
A third object of the present invention is to provide a document classification method capable of performing more accurate classification on a target document by using both classification results of manual classification and automatic classification. .

[0005]

According to the first aspect of the present invention, as shown in FIG. 8, a manual classification result obtained by manually classifying target documents within a predetermined set of plural classifications is obtained. A manual classification result acquiring unit, a target document acquiring unit for acquiring the target document, and automatically classifying the target document acquired by the target document acquiring unit within the set of the plurality of classifications. The object is achieved by providing a document classification device with an automatic classification result obtaining unit for obtaining a classification result, and a classification determination unit for finally determining a classification for the target document from the manual classification result and the automatic classification result. According to the second aspect of the present invention, as shown in FIG. 9, in the document classification apparatus according to the first aspect, the automatic classification result obtaining means obtains a target document vector characterizing the target document. Means, a typical document vector obtaining means for obtaining a typical document vector of a typical document characterizing each of the classes, and calculating a similarity between the target document vector and each of the typical document vectors to obtain a similarity for each class. A similarity calculating unit, and the similarity for each classification obtained by the similarity calculating unit is set as a classification result. In the invention described in claim 3, FIG.
As shown in FIG. 0, within a set of a plurality of predetermined classifications, a manual classification result obtaining function for manually obtaining the classification result of manually classifying the target document, a target document obtaining function for obtaining the target document, An automatic classification result acquisition function for automatically classifying a target document acquired by the target document acquisition function and obtaining an automatic classification result within the range of the plurality of classification sets, the manual classification result and the automatic classification The second object is achieved by storing a computer-readable document classification program for causing a computer to implement a classification determination function for finally determining the classification of the target document based on the result, on a storage medium. According to the invention described in claim 4, as shown in FIG. 11, the automatic classification result obtaining function includes a document vector obtaining function for obtaining a target document vector characterizing the target document, and a typical document characterizing each of the classifications. A typical document vector acquisition function of acquiring a typical document vector of the same type, and a similarity calculation function of calculating a similarity between the target document vector and each of the typical document vectors to obtain a similarity for each classification. The similarity for each classification obtained by the degree calculation function is defined as a classification result.
In the invention described in claim 5, as shown in FIG. 12, the target document is automatically classified within a predetermined range of a plurality of classification sets, and the result of the automatic classification and the range of the plurality of classification sets are determined. The third object is achieved by finally determining the classification of the target document from the result of manual classification in which the target document is manually classified.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of a document classification apparatus, a storage medium storing a document classification program, and a document classification method according to the present invention will be described below with reference to FIGS.
This will be described with reference to FIG. (1) Overview of Embodiment In the document classification processing according to the present embodiment, an evaluation function based on weights and the like obtained from the correct answer rate for the classification performed in the past is stored in a database for each person in charge of classification, and the classification is performed for each classification. A typical document characterizing the classification is prepared in advance. Then, based on the result of classification of the target document by the person in charge of classification (manually) and the evaluation function, scoring of the manual classification for each classification is performed. In addition, the similarity between the target document and the typical document is calculated, and the automatic classification is scored for each classification using the similarity. The classification having the highest sum of the two scores for each classification is defined as the final classification result. In this way, by merging the manual classification and the automatic classification, a more accurate classification result can be obtained.

(2) Details of the Embodiment The document classification apparatus according to the present embodiment is not only configured by a computer system including a personal computer, a word processor, etc., but also by a LAN (local area network) server and a computer (personal computer). ) It can also be constituted by a communication host, a computer system connected to the Internet, and the like. Further, it is also possible to distribute the function to each device on the network and configure the document classification device on the entire network.

FIG. 1 is a block diagram showing the configuration of the document classification device. As shown in FIG. 1, the document classification device includes a control unit 11 for controlling the entire system. The control unit 11 includes a keyboard 12 and a mouse 13 as input devices via a bus line 21 such as a data bus.
A display device 14, a printing device 15, a storage device 16, a storage medium driving device 17, a communication control device 18, an input / output I / F 19,
And a character recognition device 20 are connected. Control unit 11
Includes a CPU 111, a ROM 112, and a RAM 113. The ROM 112 is a read-only memory in which various programs and data for the CPU 111 to perform various controls and calculations are stored in advance.

The RAM 113 is a random access memory used as a working memory for the CPU 111. In the RAM 113, a classification processing table storage area is provided as an area for performing the document classification processing according to the present embodiment, in which a classification processing table for converting the classification results of the automatic classification and the manual classification into scores and performing processing such as normalization is stored. 1131, target document storage area 11 for storing target documents to be classified
32, a target document vector storage area for storing a target document vector characterizing the target document using the importance of the extracted keyword as an element value, a typical document vector storage area 1 for storing a typical document vector characterizing a typical document
134, similarity storage areas 1135,... For storing the similarity between the target document and each typical document, and other various areas are secured.

The keyboard 12 constitutes a part of a target document obtaining means for creating a target document in the apparatus itself and a part of a manual classification result obtaining means for inputting a classification result by a group person in charge. Various keys such as a kana key and a numeric keypad for inputting, a function key for executing various functions, a cursor key, and the like are arranged. The mouse 13 is a pointing device, and is an input device for designating a corresponding function by left-clicking a key, an icon, or the like displayed on the display device 14. As the display device 14, for example, a CRT or a liquid crystal display is used. This display device displays input results from the keyboard 12 and the mouse 13 and displays final classification results. The printing device 15 is for printing a document displayed on the display device 14, a document stored in the document storage unit 164 of the storage device 16, and the like. This printing device includes laser printers, dot printers, inkjet printers, page printers, thermal printers,
Various printing devices such as a thermal transfer printer are used.

The storage device 16 comprises a readable and writable storage medium and a drive device for reading and writing various information such as programs and data on the storage medium.
As a storage medium used for the storage device 16, a hard disk is mainly used, but a readable and writable storage medium among various storage media used in a storage medium driving device 17 described later may be used. Good. The storage device 16 stores the kana-kanji conversion dictionary 161 and the program storage unit 1
62, a data storage unit 163, a document database 164,
It has an evaluation function database 165, a document vector database 166, and other storage units (not shown) (for example, storage units for backing up programs and data stored in the storage device 16). The program storage unit 162 converts a kana character string input using the kana-kanji conversion dictionary 161 into a kanji mixed sentence, in addition to various programs such as a document classification processing program and a document vector creation processing program in the present embodiment. Various programs such as a kana-kanji conversion program are stored. The data storage unit 163 stores various data required by the system, such as data on the user.

The document database 164 stores typical documents characterizing each classification, ordinary documents other than the typical documents, and the like. The format of each document stored in the document database 164 is not particularly limited, and text documents, HTML (Hyper Text Markup Language)
ge) format documents and various types of documents such as JIS format documents can be stored. Classifications characterized by this typical document include in-house classifications such as technical trend reports, assertion reports, and new projects, general classifications such as politics, economy, and health, and general books and science used by disciples such as libraries. Classifications related to technical literature and other various classifications can be appropriately selected according to the purpose of use.

FIG. 2 conceptually shows the contents of the evaluation function database 165. As shown in FIG. 2, the evaluation function stores "weight" for each classifier A, B, C,... As an evaluation value for each classifier Hanako, Taro, Shiro,. The “weight” is determined based on the correct answer rate (or error rate) of the person in charge of classification for each classification. This “weight” is changed by comparing with the final classification result each time each person in charge determines the classification of the target document. As shown in FIG. 2, it is understood that Hanako, who is in charge of classification, has a low correct answer rate for Classification A and a high correct answer rate for Classification Hei.

FIG. 3 shows a document vector database 166.
Are conceptually represented. As shown in FIG. 3, the importance f (x) obtained for the keyword x automatically extracted from the document Ajk is stored as the element value f (x) of the document vector. This document vector is stored for each document jk (j = 1, k = 1), and is associated with each document stored in the document database 164. The dimension of each document vector is the number of keywords x (keywords) to be adopted. When the similarity between two documents is obtained from both document vectors, the number of unions of keywords of both documents is Be a dimension.
In this case, the element value of the other document vector for the keyword included in only one document vector is defined as “0”.

For example, in FIG. 3, the keyword of document B is “important, important word, importance,...”, The keyword of document C is “important,..., Politics,. It is as follows. Document vector of document B = (1,18,19, ...) Document vector of document C = (18, ..., 21, ...) On the other hand, when calculating the similarity between document B and document C, Keywords for both documents are "important, important words, importance,
..., politics, ... ", and the document vectors of both documents are defined as follows. Document vector of document A = (1,18,19, ...,
0,...), Document vector of document C = (18, 0, 0,.
1,…)

The storage medium drive 17 (FIG. 1) has a CPU
Reference numeral 111 denotes a driving device for reading data including computer programs and documents from an external storage medium.
The computer programs and the like stored in the storage medium include various processing programs such as a document classification process executed by the document classification device of the present embodiment, and dictionaries and data used therein. Here, the storage medium refers to a storage medium in which a computer program, data, and the like are stored, and specifically, a magnetic storage medium such as a floppy disk, a hard disk, and a magnetic tape, and a semiconductor storage medium such as a memory chip and an IC card. , CD-ROM or M
O, PD (phase change rewritable optical disk) and other storage media from which information can be read optically, storage media using paper (such as paper cards and tapes) (and media having functions equivalent to paper), and various other types A storage medium in which a computer program or the like is stored in the method is included. As a storage medium used in the document classification device of the present embodiment, a storage medium such as a CD-ROM or a floppy disk is mainly used. The storage medium drive 17 can read data stored in the RAM 113 or the storage device 16 into a writable storage medium such as a floppy disk in addition to reading a computer program from these various storage media. It is.

In the document classification device of the present embodiment, the control unit 1
One CPU 111 reads a computer program from an external storage medium set in the storage medium drive 17 and stores (installs) it in each unit of the storage 16. When executing various processes such as the document classification process according to the present embodiment, the corresponding program is read from the storage device 16 into the RAM 113 and executed.
However, not from the storage device 16 but the storage medium drive 1
7, it is also possible to read the program directly from the external storage medium into the RAM 113 and execute it. Further, depending on the document classification device, the document classification processing program or the like of the present embodiment may be stored in the ROM 112 in advance, and the program may be executed by the CPU 111. Further, various programs and data such as the document classification processing program of the present embodiment may be downloaded from another storage medium via the communication control device 18 and executed.

The communication control device 18 is a control device for making a network connection between the document classification device and various electronic devices such as other personal computers and word processors. The communication control device 18 can access a document in the same language as the target document of these various electronic devices, a document in another language input, and a database of documents in the same language or another language as search targets. It is possible. The target document includes various types of data such as bitmap data in addition to various types of documents such as a text format and an HTML format. The input / output I / F 19 is an interface for connecting various devices such as a speaker that outputs audio, music, and the like. The character recognition device 20 is a device for recognizing characters written on paper or the like in various formats such as a text format or HTML, and is configured by an image scanner, a character recognition program, and the like.

In the present embodiment, in addition to a document created by an input operation of the keyboard 12 (stored in a predetermined storage area of the RAM 113), a document created externally and stored in a predetermined storage medium is read from the storage medium driving device 17. Documents, documents stored in advance in a document database, documents downloaded from the communication control device 18, and the character recognition device 2.
It is possible to acquire various documents such as a document whose character has been recognized with 0 as a target document to be searched (document acquiring means).

The operation of the document classification processing of the document classification device of the present embodiment configured as described above will be described with reference to FIG. FIG. 4 is a flowchart showing the main operation of the document classification process. First, the CPU 111 acquires the target document T to be classified and stores it in the target document storage area 1132 of the RAM 113 (step 11).

Then, the CPU 111 includes:
The result of the manual classification classified by the person in charge of classification is acquired and stored in the classification processing table in the classification processing table storage area 1131 of the RAM 113 (step 12). FIG. 6 shows a RAM
This is a conceptual representation of the contents of an automatic classification table in which an area is secured as a work area 1131. If the classification determined by the classifier Hanako after reading the target document is the classification A, as shown in FIG.
In the column 61, the score of the classification A is 1 point and the other classifications are 0 points.

Next, the CPU 111 executes the acquired target document T
It is confirmed whether or not the document vector Bt has already been created and stored in the document vector database 166 (step 14). If it is stored (; Y), the document vector Bt is read and the It is stored in the document vector storage area 1133 (step 1
5). When the document vector Bt of the target document is not stored in the document vector database 166 (Step 1)
4; N), the CPU 111 creates a document vector Bt for the target document (step 16).

FIG. 5 is a flowchart showing the operation of the document vector creation processing. The CPU 111 extracts a self-sufficient word from the target document T by performing morphological analysis (step 131), and extracts candidate words (phrases) including a noun phrase, a compound noun phrase, etc. from the target document T, and performs a predetermined operation of the RAM 113. It is stored in the area (step 132). Then, the degree of importance f (x) of each candidate word (phrase) is determined from the frequency of appearance of the extracted candidate word (phrase) in the target document T and the evaluation function (step 133). Here, as the evaluation function, for example, when a predetermined important word is specified in advance, weighting for the important word, weighting according to the type of a candidate word (phrase) such as a word, a noun phrase, a compound noun phrase, and the like are used. used. Further C
The PU 111 determines the keywords a, b,... Of the target document T from the determined value of the importance f (x) (step 13).
4). Then, the document vector B = (f (a), f (b),...) Is stored in the target document vector storage area 1133 of the RAM 113 using the importance f (x) of each keyword as an element (step 135). The process returns to the document classification processing routine of FIG.

Next, the CPU 111 calculates the similarity S between the target document T and each of the typical documents of the classifications A, B, C,... (Step 17). That is, as shown in FIG. 7, the CPU 111 compares the document vector Bt of the target document with the document vector Bjk of the typical document, and calculates the similarity S between the two documents based on the cosine depending on the angle of both vectors. Generally, the angle between the document vector Bx of the document Ax and the document vector By of the document Ay is θ, the inner product of both document vectors is Bx · By, and the magnitudes of both document vectors are | Bx | and | By | In this case, the similarity S between the two document vectors is obtained by the following equation 1.

[0025]

## EQU1 ## Similarity S = COS (θ) = (Bx · By) /
(| Bx | × | By |)

The value of the similarity S takes a value of -1.ltoreq.S.ltoreq.1, and as the value is closer to 1, the two document vectors are closer to each other, and it is considered that the two documents Ax and Ay are similar to each other. be able to.

Next, the CPU 111 normalizes the similarity S calculated for the typical document of each classification so that the total value becomes 1, and uses the normalized similarity as the score of the automatic classification to obtain a classification processing table area. It is stored in the automatic b column 62 (FIG. 6) of 1131 (step 18).

Then, the CPU 111 performs an evaluation function process on the scores obtained by the manual classification and the automatic classification (step 19). That is, among Hanako's evaluation functions,
Evaluation function when the target document is classified into the classifier (weight w =
0.5) from the evaluation function database 165,
In the classification processing table of FIG. 6, the score of each classification in the Hanako a column 61 is multiplied and stored in the Hanako c column 63. Further, the score of each classification in the automatic b column 62 is multiplied by (1−w = 0.5) and stored in the automatic d column 64.

Further, the CPU 111 calculates the total value of the points of the manual classification after the evaluation function processing (Hanako c column 63) and the points of the automatic classification after the evaluation function processing (Auto d column 64) ( c + d) is obtained for each classification, and the classification having the largest total value is finally determined as the classification for the target document T (step 20; classification determining means). The CPU 111 processes the target document according to the classification purpose based on the finally determined classification (step 21), and ends the processing. As an example of processing of a target document, if the classification purpose is distribution, the target document is distributed to users belonging to the classification.

As described above, according to the present embodiment,
Based on the result of the manual classification by each classifier, the points of the manual classification for each classification are converted into points, and the automatic classification of each classification is performed based on the similarity between the document vector of the typical document characterizing each classification and the target document vector of the target document. By performing scoring, manual classification and automatic classification can be combined, and a more accurate classification result can be obtained.

Although the configuration of the present embodiment and the process of searching for a document in another language have been described above, the present invention is not limited to each of these embodiments, and various modifications are possible within the scope of the invention described in the claims. Can be modified. For example, the typical document does not need to be selected in advance, and a normal document stored in the document database 164 may be used as the typical document. Further, a document automatically extracted by clustering processing from documents stored in the document database 163 may be used as a typical document.

In the described embodiment, the typical document and its typical document vector are stored in the document database 164, respectively.
The description has been made on the premise that they are stored in the document vector database 166, but both need not necessarily be present. That is, if there is a typical document vector for the typical document (stored in the document vector database 166), the similarity S with the target document T can be calculated, so the typical document itself is not necessarily required. Conversely, if there is a typical document characterizing the classification for each classification (if stored in the document database 164),
Since the typical document vector can be created by the document vector creation process shown in FIG. 5, the typical document vector itself is not always necessary.

In the embodiment described above, the number of typical documents for one class is not particularly limited. However, the typical document does not necessarily need to be one typical document for one class, and a plurality of typical documents may be included for one class. May be prepared. In this case, a total value or an average value (either of which can be used since normalization processing is performed) is used as the similarity of the target document for each classification. In this way, by making the document into a plurality of typical documents of one classification, each can be characterized more accurately, and the accuracy of the automatic classification side can be improved.

When the final classification result is different from the classification result by the person to be classified, learning may be performed by changing the weight of the evaluation function. It is also possible to configure a document classification device by using. Further, a weight (an evaluation function for the automatic classification) may be defined for the classification result by the automatic classification (for example, the value of the similarity after the normalization by the step 18), as in the case of the manual classification. . The weighting in this case may be changed by learning.

Further, in the above-described embodiment, the language of the target document is not particularly described. However, the present invention is not limited to Japanese, and can be applied to target documents in any language. . In this case, it is only necessary to change a part that does not affect the configuration of the present invention, such as using a morphological analysis algorithm for the language of the target document. However, the language of the typical document needs to be the same as the language of the target document.

Each device, each unit, each operation, each process, and the like described in the above embodiment can be constituted by each unit as a high-level concept including these units. It is. For example, “CPU 111
Compares the document vector Bt of the target document with the document vector Bjk of the typical document, as shown in FIG. 7, and calculates the similarity S between the two documents by a cosine depending on the angle of both vectors.
Is calculated. May be configured as “similarity calculating means”. Similarly, the embodiment may be configured with a higher concept such as “「 (operation) means ”for various other operations.

[0037]

According to the present invention, manual classification and automatic classification are performed within the same set of plural classifications, and the final classification for the target document is determined by using both classification results. By using both the classification results of the manual classification and the automatic classification, it is possible to perform classification with higher accuracy by adding to the target document.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a document classification device according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram conceptually showing the contents of an evaluation function database in the embodiment.

FIG. 3 is an explanatory diagram conceptually showing the contents of a document vector database in the embodiment.

FIG. 4 is a flowchart illustrating a main operation of a document classification process according to the embodiment.

FIG. 5 is a flowchart showing an operation of a document vector creation process in the document classification process of the embodiment.

FIG. 6 is an explanatory diagram showing processing in a classification processing table up to the final determination of classification in the embodiment.

FIG. 7 is an explanatory diagram showing a similarity between a target document and a typical document in the embodiment using a document vector.

FIG. 8 is a diagram corresponding to claims of the invention described in claim 1;

FIG. 9 is a diagram corresponding to claims of the invention described in claim 2;

FIG. 10 is a diagram corresponding to claims of the invention described in claim 3;

FIG. 11 is a diagram corresponding to claims of the invention described in claim 4;

FIG. 12 is a diagram corresponding to claims of the invention described in claim 5;

[Explanation of symbols]

 11 Control Unit 112 ROM 113 RAM 1131 Classification Processing Table 1132 Target Document Storage Area 1133 Target Document Vector Storage Area 1134 Typical Document Vector Storage Area 1135 Similarity Storage Area 12 Keyboard 13 Mouse 14 Display Device 15 Printer 16 Storage Device 161 Kana-Kanji Conversion Dictionary 162 Program storage 163 Data storage 164 Document database 165 Evaluation function database 166 Document vector database 17 Storage medium drive 18 Communication control device 19 Input / output I / F 20 Character recognition device

Claims (5)

[Claims] 1. A manual classification result obtaining means for obtaining a manual classification result obtained by manually classifying a target document within a predetermined set of a plurality of classifications; a target document obtaining means for obtaining the target document; An automatic classification result obtaining unit that automatically classifies the target document obtained by the target document obtaining unit and obtains an automatic classification result within the range of the set of the plurality of classifications; the manual classification result and the automatic classification result And a classification determining means for finally determining the classification of the target document from the following. 2. The automatic classification result obtaining unit includes: a document vector obtaining unit that obtains a target document vector characterizing the target document; and a typical document vector obtaining unit that obtains a typical document vector of a typical document characterizing each of the classifications. Means, and similarity calculating means for calculating a similarity between the target document vector and each of the typical document vectors to obtain a similarity for each class, and a similarity for each class obtained by the similarity calculating means. The document classification apparatus according to claim 1, wherein the degree is a classification result. A manual classification result acquisition function for manually acquiring a result of manually classifying the target document within a predetermined set of a plurality of classifications; a target document acquisition function for acquiring the target document; The target document acquired by the target document acquisition function, within the set of the plurality of classifications, an automatic classification result acquisition function to automatically classify and obtain an automatic classification result, the manual classification result and the automatic classification result A storage medium storing a computer-readable document classification program for causing a computer to realize a classification determination function for finally determining a classification of the target document. 4. The automatic classification result obtaining function includes: a document vector obtaining function of obtaining a target document vector characterizing the target document; and a typical document vector obtaining function of obtaining a typical document vector of a typical document characterizing each of the classifications. And a similarity calculation function for calculating a similarity between the target document vector and each of the typical document vectors to obtain a similarity for each class, and a similarity for each class obtained by the similarity calculation function. 4. The storage medium document classification device according to claim 3, wherein the degree is a classification result. 5. A target document is automatically classified within a predetermined set of a plurality of classifications, and the result of the automatic classification and the target document are manually classified within the range of the plurality of classifications. A document classification method, wherein a classification for the target document is finally determined from a result of the manual classification.