JP2002099555A - Apparatus and method for document classification - Google Patents

Abstract

PROBLEM TO BE SOLVED: To overcome the problem of a prior art such that the time required for updating a database increases as the number of documents stored in the database for use in document classification becomes enormous, because database is to be updated when there are documents having much effect on a classification specifying error, and it becomes difficult to adjust the database frequently. SOLUTION: The degree of similarity of a plurality of registration document data stored in a main database 4a to document data is found (step 307). After classification and specifying (step 310), when the classification with high degree of similarity is included in erroneous classification of an erroneously specified database 4b (step 312), the degree of similarity is found from the document data and the registration document data wherein the erroneous classification of the erroneously specified database is provided (step 315). If there are registration document data with high degree of similarity, correct classification that is previously imparted therein is added to specified classification (step 317). In such an arrangement, the classification specifying error can be decreased.

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 and a document classification method for determining which classification new document data corresponds to, using a database of a plurality of documents that have been digitized and registered.

[0002]

2. Description of the Related Art In recent years, a large amount of digitized document data has been distributed, and a technique for automatically classifying to which classification the document data belongs has been put to practical use. As a general technique, a plurality of documents representing various classifications are registered in a database, and the similarity between the input document and the registered document is obtained using a vector space method or the like, and the similarity is determined. There is a method in which a classification to which an input document is supposed to be specified is specified with reference to a classification to which the input document belongs. This is based on the fact that, in a document included in each classification, generally used terms are biased.

[0003] However, in a classification for a specific technology, an explanation of an application example of the specific technology or the like is often described in a document included therein. There are cases where many words that are not directly related to the word to be included are included. For example, there is a case where a document relating to classification for an image pickup device (CCD) as a specific technique describes a video camera, a digital still camera, a scanner, or the like as an application example. If such a document is registered in the database, words that are not directly related to the document classification affect the similarity between the documents, and the similarity between the document to be the correct answer and the document of a different classification Was sometimes higher.

As a method of solving such a problem, Japanese Patent Application No. 11-33544 filed by the present inventors previously has filed a patent application.
To improve the field-specific correct answer rate by preferentially deleting from the database documents that strongly affect domain-specific incorrect answers, such as those described in the specification 2 and drawings. There is a way to update the database itself.

[0005]

When a database is updated as in the prior art, as the number of documents stored in the database used for classification of classification becomes enormous, the time required for updating increases, and the update is frequently performed. It becomes difficult to adjust the database. Also, deleting documents in a specific field from the database will break the balance of the entire database, and will affect the accuracy of classifying documents other than the class from which the documents were deleted. There is room for improvement in that it is difficult to perform database adjustment by word weight adjustment and the like.

The present invention provides a document classifying apparatus and a document classifying method which can reduce a classifying error without updating the entire database even if a document having a large influence on the classifying error exists. The purpose is to:

[0007]

Means for Solving the Problems In order to achieve the above object, according to the first aspect of the present invention, a plurality of registered document data to which classification has been given in advance and the similarity of the document data based on the similarity between the document data. In a document classifying device for specifying a classification, a first database storing a plurality of registered document data to which a classification is previously assigned, and an error classification different from the correct classification when specifying the classification of the document data. And a second database for storing the registered document data causing the error, together with the correct answer classification and the error classification, and a similarity between the document data, the registered document data in the first database, and the registered document data in the second database. And a class specifying means for specifying the class based on the similarity. With such a configuration, not only the classification result specified in the first database but also registered document data in the second database in which classification specification is apt to be erroneous, and there are similar data in the second database. In such a case, since this is also output, it is possible to reduce classification identification errors. In addition, even when the database needs to be updated in order to improve the classification error, it is efficient because only the second database needs to be updated instead of the entire database.

According to a third aspect of the present invention, there is provided a document classification apparatus which specifies a classification of a document data based on a similarity between the document data and a plurality of pre-registered document data. A first database that stores a plurality of registered document data to which a classification has been assigned in advance, and a registration that causes a classification specific error as an error classification different from the correct classification when specifying the classification of the document data. A second database for storing the document data together with the correct classification and the error classification; input means for inputting the document data; and calculating the similarity between the input document data and the registered document data in the first database. The similarity calculated by the similarity calculating means, and the similarity calculated by the similarity calculating means are added for each classification assigned to each registered document data, and the sum of all similarities is calculated. Means for calculating the ratio of the total value for each classification, classification specifying means for specifying the class based on the magnitude of the ratio calculated by this means, and the above-mentioned ratio in the classification specified by the classification specifying means A comparison unit that compares the document data with the registered data included in the error classification in the second database if the ratio is smaller than the specific value as a result of the comparison by the comparison unit; Calculating a degree, and if there is registered document data whose similarity is equal to or greater than a predetermined value, adding means for adding the correct classification assigned to the registered document data to the classification specified by the classification specifying means; Output means for outputting the classification specified as the classification of the document data, including the classification added by the means. By adopting such a configuration, it is possible to reduce classification identification errors, and it is not only efficient to update the database to improve the classification errors,
Instead of using the database, the second database is used when the total value of the similarities is smaller than the specific value, in other words, when the reliability of the classification specification is low, so that the processing efficiency can be improved. .

According to a fourth aspect of the present invention, there is provided the document classification device according to the third aspect, further comprising: a correct answer classification input means for inputting a correct answer classification to be given to the document data; The correct classification input by the correct classification input means is compared with the classification specified by the classification specifying means. If the classification specification result is incorrect, the document data is added to the second database. Data adding means. With this configuration, the second database can be automatically updated, and the maintenance of the database can be saved.

Further, the document classification apparatus of the present invention, in addition to the configuration of claim 3, further comprises a correct classification input means for inputting a correct classification to be given to the document data. The correct classification input by the correct classification input means is compared with the classification added by the additional means, and if the added classification is incorrect, the registered document data which is the basis for adding this classification is compared with the above. And deleting means for deleting from the second database. With such a configuration, the second database can be automatically updated, and the maintenance of the database can be saved.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing a hardware configuration of a document classification device according to an embodiment of the present invention. This device is constructed as one function on a computer having a general architecture.

As shown in FIG. 1, the present document classifying device
A control device 1 including a PU, a ROM, a RAM, and the like;
Input device 2 such as keyboard, pointing device, scanner, microphone, etc., display device 3 for displaying classification results, etc., and hard disk device, MO, DVD-RAM
And the like. The external storage device 4 stores document data, similarity information, and various setting values used for classification. In this embodiment, a main database (hereinafter, referred to as a “main DB”) 4a storing a plurality of registered document data to which classifications have been previously assigned, and an error classification different from the correct classification when specifying the classification of the document data. And an erroneous identification database (hereinafter, referred to as an “erroneous identification DB”) 4b that stores registered document data causing an categorization specific error together with the correct classification and the erroneous classification. The incorrect identification DB 4b is created by being derived from the main DB 4a. The details of the creation are as follows. That is, although the classification is specified by calculating the similarity between the registered document data and the classified document data using the main DB 4a, if the classification specification result is incorrect, the document data is classified according to the incorrectly specified classification. It is created by registering in the incorrect identification DB 4b. Further, by repeatedly performing this operation, the erroneously specified DB 4b stores erroneously specified document data for each classification that is erroneously specified in the main DB 4a. Therefore, the size of the erroneous identification DB 4b is naturally smaller than the main DB 4a, and the location is considerably smaller. Nevertheless, classification specific errors can be reduced. Therefore, the update is easier than updating the main DB 4a, and the erroneous identification DB is automatically updated as described later, so that resources required for database adjustment can be reduced.

FIG. 2 shows the configuration of the control device 1 in the present apparatus. In FIG. 2, the control device 1 includes a CPU 200 and a ROM.
Are functionally represented as a buffer unit 250.

The program section 200 includes an initialization section 201,
A classification document input unit 202, a registered document reading unit 203, a similarity calculation unit 204, a classification-similarity list creation unit 205, a classification identification unit 206, a similarity occupancy ratio shortage condition setting unit 207,
Similarity occupancy ratio insufficiency determination setting unit 208, similarity determination condition setting unit 209, similarity determination unit 210, classification specification result output unit 21
1. It has 14 functions of a correct answer classification information input unit 212, a classification specification result right / wrong judgment unit 213, and an incorrect specification DB update unit 214.

The buffer section 250 includes a classification document storage buffer section 251, a registered document storage buffer section 252, a similarity calculation result storage buffer section 253, a classification-similarity list storage buffer section 254, a classification specification result storage buffer section 255,
It has eight areas: a similarity occupancy ratio insufficiency determination condition storage buffer unit 256, a similarity determination condition storage buffer unit 257, and a correct answer classification information storage buffer unit 258.

The initialization section 201 clears data stored in each buffer section in the buffer section 250.

The classification document input unit 202 is a device for inputting classification data (hereinafter, referred to as classification data) input by the user using the input device 2.
(Referred to as “classified document data”) in the classified document storage buffer unit 251. At this time, a classification document ID is issued, and the classification document storage buffer unit 251
Is stored in

The registered document reading unit 203 reads document data (hereinafter, referred to as “registered document data”) stored in the external storage device 4, and stores the data in the registered document storage buffer unit 2.
52.

The similarity calculating section 204 divides the classified document data stored in the classified document storage buffer section 251 and the registered document data stored in the registered document storage buffer section 252 into words, respectively. The degree of similarity is calculated by a vector space method using the number of appearances of a vector as a component of the vector, and a classification document ID, a registered document ID, a degree of similarity, and classification information to which the registered document data belong are set, and a similarity calculation result storage buffer unit 253. Note that the similarity obtained here may be calculated by the number of common words instead of the vector space method, or other values indicating similarity between documents may be used.

The classification-similarity list creation unit 205 generates a classification-similarity list obtained by adding the similarity for each classification from the similarity information with the registered document data stored in the similarity calculation result storage buffer unit 253. Then, the ratio of the sum of the similarities of the respective fields to the sum of all the similarities is stored as the similarity occupancy in the classification-similarity list storage buffer unit 254.

The classification specifying unit 206 outputs, from the information stored in the classification-similarity list storage buffer unit 254, the one with the highest similarity occupancy to the classification specification result storage buffer 255. In addition to the classification specifying method based on the classification-similarity list, a classification to which a document having a high similarity belongs may be used as it is.

Similarity occupancy ratio insufficient judgment condition setting unit 207
The similarity occupancy ratio insufficiency determination condition input from the input device 2 by the user is stored in the similarity occupancy ratio insufficiency determination condition storage buffer unit 2.
55.

The similarity occupancy shortage judging section 208 includes the classification specifying result stored in the classification specifying result storage buffer 254 and the similarity occupancy shortage judging condition stored in the similarity occupancy shortage judging storage buffer 255. From erroneous identification DB4
It is determined whether or not to perform the classification specifying process using b.

Next, the similarity determination condition setting unit 209 stores the similarity determination condition input by the user from the input device 2 in the similarity determination condition storage buffer unit 256.

The similarity determination unit 210 determines whether or not the similarity is high based on the similarity calculation result stored in the similarity calculation result storage buffer unit 253 and the similarity determination condition stored in the similarity determination condition storage buffer unit 256. I do.

The classification specification result output section 211 outputs the classification specification result stored in the classification specification result storage buffer section 254 to the display device 3.

The correct answer classification information input unit 212 stores the correct answer classification information input by the user from the input device 2 in the correct answer classification information storage buffer 257.

The classification specification result right / wrong determination unit 213 determines whether the classification specification result is correct based on the classification specification result stored in the classification specification result storage buffer unit 254 and the correct answer information stored in the correct classification information storage buffer 257. .

The incorrectly specified DB updating unit 214 adds a classified document stored in the classified document storage buffer unit 251 to the incorrectly specified DB, or deletes a document registered in the incorrectly specified DB.

Next, the operation of the document classification device of this embodiment will be described. The operation described here is performed by the CPU of the control device 1.
Is executed using a program in the ROM and a storage area in the RAM.

The present embodiment mainly includes a first step and a first step.
It consists of two steps. The first step is a step of comparing the input classification document data with the registered document data registered, and determining a classification to which the classification document data belongs. The second step is to input the correct answer classification information of the document data classified in the first step, and if the result of the classification in the first step is incorrect, an erroneous identification DB 4
This is a step of adding classified document data to the document data b and deleting document data from the erroneously specified DB 4b.

First, the first step will be described.

First, the user uses the input device 2 to store, in the external storage device 4, document data to be referred to when classifying documents (step 301).

Subsequently, the initialization section 201 clears and initializes all buffers (step 302). In this state, when the user inputs the similarity occupancy ratio insufficient judgment condition from the input device 2, the similarity occupancy ratio insufficient judgment condition setting unit 206 stores the similarity occupancy ratio insufficient judgment condition storage buffer unit 253 (step 303). FIG. 4 is an example of storage when 0.3 is set as the similarity occupancy ratio shortage determination condition.

Subsequently, when the user inputs similarity determination conditions from the input device 2, the similarity determination condition setting unit 208 stores the condition values in the similarity determination condition storage buffer unit 254 (step 304). FIG. 5 shows a similarity determination condition of 0.4.
This is a storage example in the case where is set.

After inputting the similarity occupancy ratio insufficient judgment condition and the similarity judgment condition, the user uses the input device 2 to input classified document data. When the classification document data is input, the classification document input unit 202 changes the classification document storage buffer unit 251.
Stored as the search key. (Step 305). FIG. 7 shows an example of classified document data stored in FIG.

When the classified document data is stored, the registered document reading unit 203 reads a plurality of document data from the external storage device 4 and stores the read document data in the registered document storage buffer unit 252 as registered document data (step 306). As shown in FIG. 7, the registered document data is provided with a registered document ID for identifying the document and classification information indicating the classification of the document, in addition to the text.

Subsequent to step 306, similarity calculating section 2
04 obtains the similarity between the registered document data and the classified document data (step 307). That is, the text of the classified document data stored in the classified document storage buffer unit 251 is compared with the text of the registered document data stored in the registered document storage buffer unit 252, and a numerical value indicating the degree of similarity of each data is used. A certain similarity is calculated using a vector space method. The calculated similarity is stored in the similarity calculation result storage buffer unit 253 together with the registered document ID and the classification information indicating the classification of the registered document data. At this time, a predetermined number of similarities may be stored in descending order of similarity, or only those having a similarity higher than a previously specified similarity may be stored. In the storage example of the similarity calculation result in FIG. 8, the first data is the document I
D = 1023, classification = television, and similarity = 0.378 are stored.

When the similarity has been stored, it is determined whether or not there remains any registered document data for which the similarity has not been calculated (step 308). Step 30 as the target
6 and 307 are repeated. On the other hand, if it is determined that there is no other registered document data, the process proceeds to step 309.

In step 309, the classification-similarity list creation unit 205 sums the similarity calculation results stored in the similarity calculation result storage buffer unit 253 for each classification to which each registered document belongs, and stores the classification-similarity list. Buffer unit 25
4 and further calculates the ratio of the sum of the similarities to the sum of all the output similarities and stores the ratio as the similarity occupancy. In the example of the classification-similarity list in FIG. 9, the classification = the sum of the similarities of the documents belonging to the personal computer is 0.729, and the classification = the sum of the similarities of the documents belonging to the video is 1.5.
93. The sum of all the similarities is 0.514 + 1.593 + 0.729 + 1.782.
Assuming a value of 8.5 as +..., The similarity occupancy is 0.514 / 8.5 = 0.0 in the case of classification = TV.
60 and 1.593 / 8.5 when classification = video
= 0.187.

The classification specifying unit 20 sequentially sorts the similarities in the classification-similarity list obtained in this manner in descending order of similarity.
6 is stored in the classification specification result storage buffer 255 (step 310). At this time, the database used for the classification is the main DB 4a, and the similarity occupancy of the specified classification is stored in association with each other. In the example of storing the classification specification results in FIG. 10, the first candidate is classified as video camera, similarity occupancy = 0.210, the second candidate is classified as video = video, similarity occupancy = 0.187, and the third candidate is third camera. Classification = PC, similarity occupancy = 0.086 is output.
Although the storage area of the similar registration document ID is also provided in the classification specification result storage buffer unit 255, at this time,
Nothing is stored in this storage area.

The similarity occupancy rate of the first candidate among the data stored in the classification specification result storage buffer section 255 is stored in the similarity occupancy rate shortage determination section 208 by the similarity occupancy rate shortage determination section 208. Similarity degree occupancy ratio shortage determination condition stored in buffer unit 256 (0.3 in the example of FIG. 4)
(Step 311). When the similarity occupancy of the first candidate is larger, the process proceeds to Step 318. When the similarity occupancy is smaller, the process proceeds to Step 312. In the present embodiment, the similarity occupancy of the first candidate is set as the target of the similarity occupancy determination in order to use a case in which a classification with high certainty is not included in the classification identification result as a condition for the determination.

In step 312, it is determined whether the classification of the first candidate stored in step 310 is registered in the erroneous identification DB 4b. If registered, the process proceeds to step 313, and if not, the process proceeds to step 318. . FIG.
1 is an example of the document ID of the document data registered in the erroneous identification DB 4b. This example shows that document data of document IDs 3, 19,... Are registered in the classification = video camera. In this case, the classification of the first candidate of the classification specification result in step 310 is a video camera, and the determination result in step 312 is Yes.
Will go to.

In step 313, the registered document reading unit 203 stores the registered document data registered in the first candidate classification identified in step 310 among the registered document data registered in the erroneous identification DB 4b in the external storage device. 4 and stored in the registered document storage buffer unit 252.

Next, similar to step 307, the similarity calculating section 204 calculates the similarity between the classified document data stored in the classified document storage buffer 251 and the registered document data stored in the registered document storage buffer 252. Is calculated and stored in the similarity calculation result storage buffer unit 253 (step 3).
14). Steps 313 and 314 are repeatedly executed for all registered document data stored in the registered document storage buffer unit 252, and similarity calculation and storage of the similarity for all registered document data stored in the buffer 252 are performed. Upon completion, the process proceeds to step 316 (step 3
15). FIG. 12 is an example of the calculation result of the similarity with the document registered in the video camera in the classification of the erroneous identification DB 4b. The document data of the registered document ID = 3 is the classification = camera and the similarity is 0.226. The document data of the registered document ID = 19 indicates that the classification = digital camera and the similarity is 0.423.

In step 316, the similarity determination condition (0.4 in the example of FIG. 5) stored in the similarity determination condition storage buffer 257 and the similarity calculation result storage buffer 25
The similarity stored in No. 3 is compared. If the similarity determination condition is higher, the process proceeds to step 318. If the similarity determination condition is lower, the process proceeds to step 317. In step 317, the classification of the corresponding document data, the similar registration document I
D and the fact that it is the classification specification result in the incorrect specification DB 4b is added to the classification specification result storage buffer unit 255. FIG.
In the example, the similarity of the document data with the registered document ID = 19 exceeds the similarity determination condition of 0.4, and is added to the classification specification result storage buffer. And this registration document ID = 1
Since the registered document data of No. 9 is stored in the erroneous identification DB 4b also indicating that the correct classification = digital camera,
Classification result storage buffer unit 2 as the classification result in 4b
In 55, the classification = digital camera is stored. If there is no registered document data to be added according to the classification specification result by the erroneous specification DB 4b, it remains unchanged in FIG.

The classification specification result obtained in this manner and corresponding to the classification document data stored in the classification specification result storage buffer unit 255 is output to the display device 3 by the classification specification result output unit 211 so that the user can recognize it. (Step 318). FIG. 14 shows an output example in this case.

After outputting the result of specifying the classification of the target classified document data, it is determined whether or not the classified document data which has not been classified yet remains in the classified document storage buffer unit 251 (step 319). If so, the process returns to step 305, and steps 305 to 318 are repeated. On the other hand, if no classified document data remains, the classification process ends. This ends the first step.

Next, the second step will be described.
FIG. 15 is a flowchart showing the procedure of the second step.

First, the initialization unit 201 clears all buffer units other than the classification specification result storage buffer unit 255 (step 351).

Next, in this state, when the classification document ID of the classification document data classified in the first step and the correct classification of the document data are input from the input device 2, the correct classification information input unit 212 causes the correct classification This input information is stored in the information storage buffer unit 258 (step 352). FIG.
This is an example of storage when the correct classification of the document data of the classification document ID = 1 is input to the personal computer.

Next, the classification specification result correctness judgment unit 213 uses the main DB 4a stored in the classification specification result storage buffer unit 255 to classify the classification result and the correct classification information stored in the correct classification information storage buffer unit 258. Compare.
First, in step 353, the misidentification D is added to the classification identification result.
It is determined whether there is a classification specification result by B4b. If there is, the process proceeds to step 354; otherwise, the process proceeds to step 356.
The classification specification result is the content of FIG. 13 and the correct classification information is FIG.
In the case of the content, the classification specification result includes the classification specification result in the erroneous specification DB 4b.

In step 354, the classification specification result right / wrong determination unit 213 determines whether the classification specification result by the incorrect specification DB 4b is the same as the correct classification information, and if they are the same, proceeds to step 358; Then, the process proceeds to step 355, where the similar registered document data is deleted from the erroneous identification DB 4b. In the present embodiment, the classification result using the main DB 4a stored in the classification specification result storage buffer unit 255 is the contents of FIG. 13, and the correct classification information stored in the correct classification information storage buffer unit 258 is the contents of FIG. Since the classification specification result by the erroneous identification DB 4b is a digital camera and the correct classification information is different from a personal computer, in step 355, the document data of the similar registration document ID = 19 of the classification identification result in the erroneous identification DB 4b is erroneously specified. Will be deleted.

After the deletion of the registered document data from the erroneous identification DB 4b in step 355, in step 356, the classification identification result correctness judgment unit 213 determines that the classification identification result by the main DB 4a of the classification identification result is the same as the correct answer classification information. Is determined, and if they are the same, the process proceeds to step 358, and if they are different, the process proceeds to step 357 to add the classified document data to the erroneous identification DB 4b. For example, when the classification result is the content shown in FIG. 17 and the correct classification information is the content shown in FIG. 16, since the classification result in the main DB 4a does not include the personal computer of the correct classification, the process proceeds to step 357, and the incorrect specification is performed. The classified document data is incorrectly specified D by the DB update unit 214.
It is additionally registered in B4b.

In step 358, it is determined whether there is correct information that has not been input, and if so, the process returns to step 352 to repeatedly execute the processing of steps 352 to 357. On the other hand, the present invention, in which the process is terminated if there is no unanswered correct answer information, is not limited to the above embodiment as long as it does not depart from the gist of the present invention. For example, instead of inputting the similarity occupancy ratio insufficient judgment condition or the similarity judgment condition each time,
It is also possible to store it in advance as a default value.

[0057]

As described above in detail, according to the present invention, even if there is a document having a great influence on a classification error, it is possible to reduce the classification error without updating the entire database. Become.

[Brief description of the drawings]

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

FIG. 2 is a functional block diagram of a control device in the document classification device of FIG. 1;

FIG. 3 is a diagram showing a procedure for executing classification and identification of document data using a main DB and an erroneous identification DB;

FIG. 4 is a diagram showing an example of storing a similarity occupancy ratio insufficient judgment condition;

FIG. 5 is a diagram showing an example of storing similarity determination conditions.

FIG. 6 is a diagram illustrating an example of storing classified document data.

FIG. 7 is a diagram showing an example of storing registered document data.

FIG. 8 is a diagram showing an example of storing a similarity calculation result;

FIG. 9 is a diagram showing an example of a classification-similarity list.

FIG. 10 is a diagram showing an example of classification identification result storage by the main DB

FIG. 11 is a diagram illustrating an example of storing a document ID in an erroneous identification DB for each category;

FIG. 12 is a diagram illustrating an example of a similarity calculation result using an erroneous identification DB;

FIG. 13 is a diagram showing an example of adding a classification specification result by an incorrect specification DB;

FIG. 14 is a diagram showing a classification result output example.

FIG. 15 is a diagram showing a procedure for executing a classified document addition / deletion process of an erroneously specified DB;

FIG. 16 is a diagram showing an example of correct answer classification information.

FIG. 17 is a diagram showing another example of a classification specification result.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Control device 2 ... Input device 3 ... Display device 4 ... External storage device 200 ... Program part 206 ... Classification specifying part 207 ... Similarity occupancy ratio shortage judgment condition setting part 208 ... Similarity occupancy ratio shortage judgment part 209 ... Similarity judgment Condition setting unit 210: similarity determination unit 211: classification identification result output unit 212: correct answer classification information input unit 213: classification identification result right / wrong determination unit 214: incorrect identification DB update unit 250: buffer unit 254: classification similarity list storage buffer unit 255: classification specification result storage buffer unit 256: similarity occupancy ratio insufficiency determination condition storage buffer unit 257: similarity determination condition storage buffer unit 258 ... correct answer classification information storage buffer unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Tsuyoshi Matsukuma 3-3-3 Shinmachi, Ome-shi, Tokyo Toshiba Digital Media Engineering Co., Ltd. In-house (72) Yukio Nakamoto 3-3-1 Shinmachi, Ome-shi, Tokyo 1 Toshiba Digital Media Engineering Co., Ltd. In-house (72) Inventor Takuya Nishina 3-3-3 Shinmachi, Ome-shi, Tokyo In-house Toshiba Digital Media Engineering Co., Ltd. In-house (72) Inventor Hiroshi Yamazaki 3-3-3, Shinmachi, Ome-shi, Tokyo 1 Toshiba Digital Media Engineering Corporation In-house F-term (reference) 5B075 NR03 NR12 PR06 QM08 UU06

Claims (10)

[Claims] 1. A document classification apparatus for identifying a class of document data based on a similarity between the document data and a plurality of registered document data to which a plurality of classifications are assigned in advance. And a second database that stores registered document data that causes a classification specific error as an error classification different from the correct classification when specifying the classification of the document data, together with the correct classification and the error classification. Classifying means for determining the similarity between the document data and the registered document data in the first database and the registered document data in the second database, and specifying a classification based on the similarity. Document classification apparatus characterized by the above-mentioned. 2. A document classification apparatus for identifying a class of document data based on a similarity between the document data and a plurality of registered document data to which a plurality of classifications have been assigned in advance. And a second database that stores registered document data that causes a classification specific error as an error classification different from the correct classification when specifying the classification of the document data, together with the correct classification and the error classification. Input means for inputting document data; classification specifying means for specifying a classification based on the similarity between the input document data and the registered document data in the first database; The classification is included in the error classification in the second database, and the document data is included in the registered document data corresponding to the error classification. If there is data whose similarity with the data is equal to or more than a predetermined value, additional means for adding the correct answer classification given to the registered document data to the classification specified by the classification specifying means, and Output means for outputting a classification specified as the classification of the document data, including a classification. 3. A document classification apparatus for identifying a class of document data based on a similarity between the document data and a plurality of registered document data to which a plurality of classifications are assigned in advance. And a second database that stores registered document data that causes a classification specific error as an error classification different from the correct classification when specifying the classification of the document data, together with the correct classification and the error classification. Input means for inputting document data, similarity calculating means for calculating the similarity between the input document data and the registered document data in the first database, and the similarity calculating means. Means for adding the similarity for each category assigned to each registered document data, and calculating the ratio of the total value for each category to the total value of all similarities; Classification specifying means for specifying a class based on the magnitude of the calculated ratio, comparing means for comparing the ratio in the classification specified by the classification specifying device with a specific value, and a result of the comparison by the comparing means Is smaller than the specific value, the similarity between the registered document data included in the error classification in the second database and the document data is calculated, and if there is registered document data whose similarity is equal to or greater than a predetermined value, Adding means for adding the correct classification assigned to the registered document data to the classification specified by the classification specifying means, and the classification specified as the classification of the document data including the classification added by the additional means. Output means for outputting a document. 4. A correct answer classification inputting means for inputting a correct answer classification to be given to the document data, and comparing the correct answer classification inputted by the correct answer input means with the classification specified by the classification specifying means. 4. The document classifying apparatus according to claim 3, further comprising: a document data adding unit that adds the document data to the second database when the classification specification result is incorrect. 5. A correct answer classification inputting means for inputting a correct answer classification to be given to the document data, and comparing the correct answer classification inputted by the correct answer input means with the classification added by the additional means, Deleting means for deleting, from the second database, the registered document data on which the added classification is incorrect when the added classification is incorrect;
The document classification device according to claim 3, comprising: 6. A similarity between the document data and a first plurality of registered document data to which a classification has been previously assigned, a first classification is specified based on the calculated similarity, and the document data is defined as: A similarity with a second plurality of registered document data that is recognized in advance as causing an erroneous classification such that the erroneous classification is different from the correct classification is calculated, and the calculated similarity and the correct classification are used to calculate a second similarity. A document classification method for specifying a classification and specifying the classification of document data by adding the second classification to the first classification. 7. A similarity between the document data and a first plurality of registered document data to which a classification is given in advance is calculated, and a classification is specified based on the calculated similarity. If there is a second plurality of registered document data that is recognized in advance to cause an erroneous classification as an error classification different from the correct answer classification, the registered document data and the document data If there is registered document data having a predetermined value or more in the calculated similarity, the classification of the document data is specified by adding the correct classification of the registered document data to the above specified classification. Document classification method to be used. 8. A similarity between the document data and a first plurality of registered document data to which a classification has been previously assigned is calculated, and the calculated similarity is added for each classification assigned to each registered document data. , Calculate the ratio of the total value of each class to the total value of all similarities, specify the class based on the magnitude of the calculated ratio, and classify the class based on the magnitude of the ratio in the specified class. Identify, compare the percentage in this identified classification with a specific value, and as a result of this comparison, if the percentage is smaller than the specific value, cause an incorrect classification as a wrong classification different from the correct classification The similarity between the previously recognized second plurality of registered document data and the document data is calculated. If the calculated similarity includes registered document data equal to or more than a predetermined value, the correct classification of the registered document data is determined. Identified above Document classification method of identifying a classification of the document data by adding the classification. 9. A correct classification to be given to the document data is input, and the input correct classification is compared with the specified classification. If the specified classification is incorrect, the document data is 9. The method according to claim 8, further comprising: adding to the second plurality of registered document data. 10. A correct classification to be given to the document data is input, and the input correct classification is compared with the added classification, and when the added classification is incorrect,
9. The document classification method according to claim 8, wherein the registered document data that is the basis for adding the classification is deleted from the second plurality of registered document data.