JP2001060199A - Device and method for classifying document and computer readable recording medium storing document classfication program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform high-accuracy document classifying processing by classifying a document after the document is converted so as to include necessary and sufficient keywords in the document. SOLUTION: This device is provided with an input part 105 for inputting a document to be classified and attribute information attached to the document, a document analytic part 111 for extracting a keyword by analyzing the inputted document on the basis of morpheme analysis, a classification rule storage part 103 for storing a classification rule for describing the prescribed combination composed of at least one of the keyword showing the characteristic of a group and the attribute information for each group, a similarity degree calculating part 115 for calculating the degree of similarity between each of classification rules and the keyword or attribute information of the document and a document classifying part 116 for finding a group corresponding to the classification rule, with which the document is most similar, on the basis of the calculated similarity degree and making the document correspondent to the group.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document classification apparatus for classifying documents described in a natural language into a predetermined number of groups.
Document classification method and a computer-readable recording medium storing a document classification program, particularly, when classifying documents described in natural language into a predetermined number of groups,
The present invention relates to a technology that enables highly accurate document classification processing by using attribute information attached to a document and complementing a keyword.

[0002]

2. Description of the Related Art FIG. 8 shows a prior art of a document classifying apparatus for classifying documents described in a natural language into a predetermined number of groups.

Conventionally, in a document classifying apparatus for classifying documents described in a natural language into a predetermined number of groups, as shown in FIG.
The typical documents are manually classified into several groups, and the results are stored in a suitable recording medium. In the actual document classification process, the similarity between a document to be classified (hereinafter referred to as a target document) and a typical document in a group is compared and calculated, and it is automatically determined to which group the target document is classified. (See, for example, JP-A-11-4
No. 5247).

[0004]

However, the above-described conventional document classification technology has the following problems.

[0005] First, in the conventional document classification processing, when a necessary and sufficient amount of keywords are included in a target document, high-precision classification processing is possible. In the case of a document category such as a sentence, a part of the keyword is omitted, or conversely, a keyword more than necessary appears frequently. In this case, since the information used for the classification process is extremely limited, a highly accurate classification process cannot be performed.

Second, the document classification processing is sometimes performed using information attached to a document provided from the outside, such as information on a speaker or information on a speech scene, obtained from a document category such as an interview or a questionnaire. But,
In the conventional document classification process, since the document classification process is performed using only the keywords in the target document, it is practically impossible to reflect information attached to the externally provided document in the document classification process. As a result, the accuracy of the document classification process is reduced.

The present invention has been made to solve the above problems. The object of the present invention is to provide a document classification device, a document classification method, and a computer-readable program that stores a document classification program that enables high-accuracy document classification processing by using attribute information attached to a document and complementing keywords. It is to provide a possible recording medium.

Another object of the present invention is to provide a document classifying apparatus, a document classifying method, and a computer storing a document classifying program which greatly reduce the labor and time required for the document classifying process by facilitating the document classifying process. It is to provide a readable recording medium.

[0009]

In order to solve the above technical problem, the present inventors have proposed a classification rule which describes a predetermined combination of at least one of a keyword in a document and attribute information attached to the document. The document classification process is performed on the basis of, and the keyword that is omitted in the target document is complemented, so that a high-precision document classification process can be realized.

A first feature of the present invention based on this idea is a document classifying device for classifying documents described in a natural language into a predetermined number of groups, the classifying documents and attribute information attached to the documents. , An input unit for analyzing the input document based on the morphological analysis, and extracting a keyword, and, for each group, one or more of keyword and attribute information indicating characteristics of the group. A classification rule storage unit that stores a classification rule describing a predetermined combination, a similarity calculation unit that calculates a similarity between each of the classification rules and a keyword or attribute information of the document, and a similarity calculation unit that calculates a similarity based on the calculated similarity. And a document classification unit that obtains a group corresponding to a classification rule with which a document is most similar, and associates the document with the group.

According to the above configuration, since information other than the keyword can be reflected in the document classification result, the accuracy of the document classification process can be improved.

A second feature of the present invention is that a determination unit for determining whether or not a keyword of a document has been omitted based on the keyword or attribute information, and a determination unit that has determined that the keyword has been omitted. And a keyword complementer for restoring the omitted keyword in the document and / or adding an expression similar to the keyword and outputting the expression to the similarity calculator.

According to the above configuration, even when keywords in the target document are omitted, necessary and sufficient keywords can be replenished in the target document, so that highly accurate document classification processing can be performed. it can.

A third feature of the present invention is that the attribute information includes identification information for identifying a user corresponding to the document, question information corresponding to the document, scene information, creation time of the document, a document category for the proposition, and a document category. And one or more pieces of connection information of the preceding utterance.

A fourth feature of the present invention resides in that the classification rules are described by including a predetermined combination of at least one of a nominal element, a verbal element, and a modifier element.

A fifth feature of the present invention resides in that groups are formed hierarchically according to the abstraction level of the corresponding classification rule.

Further, a sixth feature of the present invention is to provide a display unit for displaying, for each group, a list of documents belonging to the group together with a classification rule corresponding to the group or a similar expression thereof.

A seventh feature of the present invention resides in that the display unit displays a list of documents that are not associated with any group.

An eighth feature of the present invention resides in that an editing unit is provided for instructing another group to change a group associated with each displayed document.

Further, a ninth feature of the present invention resides in that each group includes a classification rule generation unit that recursively generates a classification rule from a document group belonging to the group.

[0021] A tenth feature of the present invention is that the similarity calculation unit can variably set the weight of the keyword according to the attribute information.

According to the above configuration, since a difference in importance can be provided between keywords, the points of document classification processing can be narrowed and classification accuracy can be improved. Further, for example, by setting the weight variably according to the part of speech, the accuracy of classification can be improved.

An eleventh feature of the present invention is a document classification method for classifying a document described in a natural language into a predetermined number of groups, and inputs a document to be classified and attribute information accompanying the document. An input step, a document analysis step of analyzing the input document based on the morphological analysis and extracting a keyword, and a predetermined combination including at least one of keyword and attribute information indicating characteristics of the group, for each group. A similarity calculation step of calculating a similarity between each of the described classification rules and the keyword or attribute information of the document; and, based on the calculated similarity, a group corresponding to the classification rule to which the document is most similar, And a document classification step of associating a document with a group.

According to the above configuration, since the document classification process is performed after the document is converted so that the document includes a necessary and sufficient keyword, a highly accurate document classification process can be performed.

A twelfth feature of the present invention resides in that, based on a keyword or attribute information, a determination unit for determining whether or not a keyword of a document has been omitted, and a determination unit that determines that the keyword has been omitted. Another object of the present invention is to provide a document classification method having a keyword complementing step of restoring an omitted keyword to a document and / or adding an expression similar to the keyword and outputting the result.

A thirteenth feature of the present invention is a computer-readable recording medium storing a document classification program for classifying a document described in a natural language into a predetermined number of groups. Is analyzed based on the morphological analysis, a document analysis process for extracting a keyword, an input process for inputting a document to be classified and attribute information attached to the document, and an input document is analyzed based on the morphological analysis,
A document analysis process for extracting a keyword, a classification rule describing a predetermined combination of at least one of the keyword and attribute information indicating the characteristics of the group for each group, and a keyword or attribute information of the document. A similarity calculation process for calculating the similarity and a group corresponding to the classification rule with which the document is most similar based on the calculated similarity, and a document classification process for associating the document with the group. A computer-readable recording medium that stores a document classification program to be executed by a computer.

According to the above configuration, the document classification process is performed after the document is converted so as to include a necessary and sufficient keyword in the document, so that a highly accurate document classification process can be performed.

A fourteenth feature of the present invention resides in that, based on a keyword or attribute information, a determination unit that determines whether a keyword of a document has been omitted, and a case that it is determined that the keyword has been omitted. And a computer-readable record storing a document classification program for causing a computer to execute a keyword supplementation process for restoring an abbreviated keyword in a document and / or adding an expression similar to the keyword and outputting the same. Being a medium.

[0029]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A document categorizing apparatus, a document categorizing method, and a computer-readable recording medium storing a document categorizing program according to an embodiment of the present invention will be described in detail below with reference to FIGS. .

In this embodiment, the keyword is complemented and attribute information other than the keyword is used. By converting the target document so that the target document includes a necessary and sufficient keyword, the accuracy of the target document is improved. Provides a function that enables high document classification processing.

First, the configuration of the document classification device according to the embodiment of the present invention will be described.

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

The document classification device 10 according to the embodiment of the present invention
0 is a structured rule database 101 storing structured rules for extracting “document structure information”, a supplemental keyword database 102 storing “keyword information” used for complementation processing, and a group A classification rule database 103 that stores “classification rules” in which one or more combinations of predetermined keyword or attribute information (synonymous with the accompanying attribute information described later) indicating characteristics are described for each group, document information, and document-related information Document storage unit 10 for storing document information such as document classification results and the like
4. An input unit 105 for inputting a target document and "attached attribute information (= attribute information)" accompanying the target document, and displaying a list of documents belonging to the group for each group together with a classification rule corresponding to the group or a similar expression thereof. Part 10
7. An editing unit 106 for changing a group associated with each displayed document, and a classification processing unit 110 for performing a document classification process on the input target document.

The classification processing unit 110 analyzes a target document based on morphological analysis and extracts keyword information. The document analyzing unit 111 extracts document structure information from the target document, accompanying attribute information, and keyword information. An extraction unit 112, a determination unit 113 that determines whether or not the keyword of the target document is omitted based on the keyword or attribute information of the target document, a complement that complements a keyword lacking in the target document by using the keyword information and the attribute information A keyword complementing unit 114 that performs processing, a similarity calculating unit 115 that calculates the similarity between each of the classification rules and the keyword information or the attribute information, and a target document corresponding to the classification rule with which the target document is most similar based on the calculated similarity. A document classification unit 11 that determines a group and associates the target document with the group.
6. A classification rule generation unit 117 that recursively extracts the characteristics of the group from the document group belonging to the group for each group (for example, using a known regression analysis method or the like) and updates the classification rule.

Further, the document information storage unit 104 stores, in a format as shown in FIG. 5, information about the speaker, the utterance time, the keyword, the complementary keyword, the analysis attribute, and the belonging group together with the input document. .

Here, the "document structure information" includes information on the structure of the target document, such as the number of characters, the number of phrases, the sentence pattern, the sentence / colloquial language, the genre of the document (for example, impressions, requests, etc.), and themes. , And the number of characters and the number of phrases are extracted from the target document and the result of the morphological analysis. In addition, the sentence pattern referred to here is S obtained from a morphological analysis result.
(Main part) + V (predicate), S + O (target part) + V, etc. Judgment of colloquial / colloquial is determined by examining phrases used only in colloquial or colloquial,
If the sentence / colloquial cannot be distinguished, it cannot be determined. Spoken documents are more likely to have the same keyword repeatedly or to have omitted keyword information than textual documents. For this reason, for example, the determination criteria of the classification rule are determined by using the determination result of the language rule as a part of the classification rule, or by weighting the coefficients of the classification rule based on the determination result of the phrase language / spoken language. By making it different, classification accuracy can be improved. The genre of a document is a result of judging the content of the document as to whether the document is an impression of a product, a request, or a question. For example, if the ending is “to want”, it is determined that the content of this document is a request. The theme is a set of character strings, and indicates what topic the target document relates to. For example, if the target document is a topic related to a personal computer, "PC"
(Soccer is a high-level concept of soccer, so sports should be included in the theme.) Each character string included in the theme is added to the keyword information at the time of keyword completion processing. As a result, when the topic is related to “soccer”, even when the target document does not include the keyword “soccer”, the keyword information includes “soccer”. In addition, by treating the theme as document structure information, the accuracy of classifying documents of a type that treats the theme consistently within the same article, such as newspapers and news, is increased.

The "keyword information" is information composed of a set of a character string serving as a keyword and a part of speech information of the character string. In the keyword information, the weight of the keyword can be variably set according to the accompanying attribute information.

Further, the "classification rule" is a description including a predetermined combination of any one or more of a nominative element, a verbal element, and a modifier element in a document. The classification rules are stored in the classification rule database 103 together with the documents (affiliation documents) associated with the classification rules, as shown in FIG. Further, the classification groups may be formed hierarchically according to the abstraction level of the corresponding classification rule, so that the search of the classification result is facilitated.

Further, the "associative attribute information (= attribute information)" is identification information for identifying the user corresponding to the document.
(For example, the name of the speaker), creation time of the target document, question information corresponding to the target document, scene information (information about a scene when the target document was created, for example, the speaker is an elderly person, etc.), This means a document category for a proposition, connection information between documents appearing before and after (for example, dependency on words) and the like.

It is assumed that the accompanying attribute information is information that accompanies the target document from the beginning, whereas the document structure information is the structure information of the target document obtained by analyzing the target document. In that sense, the keyword information is also a type of document structure information because it is information obtained by performing morphological analysis on the target document.However, since keywords are more frequently incorporated into classification rules than other structural information, here Distinguish from document structure information.

In addition, in the complementing process performed by the keyword complementing unit, for a keyword having a high degree of relevance, if one of the keywords comes out, the other is automatically complemented. It is good to insert the keyword that complements the place.

Next, a processing procedure of the document classification processing according to the embodiment of the present invention will be described.

FIG. 2 is a flowchart showing a processing procedure of the document classification processing according to the embodiment of the present invention.

(1) A target document is input to the input unit 105 (target document input step, S201).

(2) The accompanying attribute information accompanying the target document is input to the input unit 105 (attribute information input step, S202).

(3) The document analysis section 111 analyzes the input target document based on morphological analysis and extracts keyword information (keyword information extraction step, S2
03). Here, when the keyword is a part of speech that is used, the part of speech is preferably put into an end form, so that a similarity calculation step described later can be executed regardless of the use form.

(4) The document structure information extraction unit 112 extracts document structure information using the target document, accompanying attribute information, and keyword information (document structure information extraction step S204).

(5) The determining unit 113 determines whether the keyword of the target document is omitted based on the keyword or the attribute information (determination step, S205). As a result of the determination, the process proceeds to (keyword complementing step, S300) if omitted, and to (similarity calculating step, S207) if not omitted.

(6) The keyword complementing unit 114 extracts a keyword to be complemented using the target document, accompanying attribute information, keyword information, and document structure information (keyword complementing step, S300).

(7) The similarity calculator 115 calculates the similarity between each of the classification rules and the keyword or attribute information of the target document (similarity calculation step, S207).

(8) The document classification unit 116 extracts a group corresponding to the classification rule to which the target document is most similar based on the calculated similarity, and associates the group with the extracted group (document classification step, S208).

(9) The display unit 107 displays a list of documents belonging to the group together with a classification rule corresponding to the group or a similar expression thereof for each group (classification result display step, S209). Here, a list of documents that are not associated with any group may be displayed. With this configuration of the display unit 107, the number of target documents to be manually classified can be significantly reduced, and the time required for the classification process can be reduced.

(10) The group associated with each displayed document is changed via the editing unit 106 (classification result change step, S210). Note that this processing step may be omitted.

(11) The classification rule generation unit 117 recursively extracts the characteristics of the group from the document group belonging to the group for each group by using, for example, a known regression analysis or the like, and updates the classification rule ( Classification rule update step, S211). It should be noted that this processing step should be performed when the classification result change step is performed. However, in other cases, it is not necessary to perform the processing step, so that the time required for the document classification processing can be reduced.

(12) The determining unit 116 determines whether there is an unclassified document (unprocessed document determining step, S21
2). If there is an unprocessed document, the process returns to (target document input step, S201). If there is no unprocessed document, the document classification process ends.

Next, the above-described keyword complementing step S3
Details of the process at 00 are described below.

(5-1) First, a complement rule to be processed is extracted from the complementary keyword database 102, and a complement source keyword set Cx is extracted (complement source keyword set extraction step, S301). FIG. 7 (a),
FIG. 2B illustrates a complementary rule including the keyword information in the complementary keyword database 102.

(5-2) It is determined whether or not all keywords of the complement source keyword set Cx are included in the keyword information (determination step, S302). As a result of the determination,
If it is included (complement destination keyword storage step, S
303), and if not included (unprocessed keyword determination step, S304).

(5-3) A complement destination keyword is derived based on a complement rule (refer to a specific example described later for a complement rule which is a complement destination keyword derivation rule), and all keywords of a complement destination keyword set Cy are complemented. The keyword is added to the keyword database 102 (complement destination keyword saving step, S303).

(5-4) It is determined whether there is an unprocessed complement rule (unprocessed complement rule determining step, S30)
4). As a result of the determination, if there is an unprocessed complement rule, the process returns to the step of “complement source keyword set extraction step, S30
If there is not, the (keyword complementing step, S300) ends, and the (similarity calculating step, S2)
07).

Here, not only the keyword information but also both the keyword information and the complement destination keyword may be used to recursively perform the keyword complementing process. This makes it possible to extract another complement destination keyword based on the keyword complemented earlier.

The theme information in the document structure information is extracted using accompanying attribute information, keyword information, and other document structure information. Basically, the supplementary rule A (described later) is based on the keyword information. It is advisable to use a database of the same format as above. For example, "mouse, keyboard"
Is included in the keyword information, the “theme” is set to “the personal computer”. Here, when the attribute information on the theme is included in the accompanying attribute information from the beginning, it may be used. When the target document continues the topic of the upstream document, the sum of the theme extracted from the target document itself and the theme of the upstream document is set as the theme of the target document. Therefore, as long as the topic continues, the number of character strings included in the theme increases toward downstream documents. Conversely, if the topic does not continue, the theme is cleared. In addition,
Which document is upstream of the target document is determined based on the accompanying attribute information. Whether the topic is continuing may be determined by referring to the character string of the dependency. For example, if the document is news, it is determined that the continuation is continued if "Receive this" or "Response" is received, and "Next"
If "continue" or the like comes, it is good to judge that it is not continuing.

The document classifying apparatus 100 according to the embodiment of the present invention has an editing unit 106 and a display unit 107, and displays a list of documents belonging to the group for each group together with a classification rule corresponding to the group or a similar expression thereof. I do. Furthermore, since the group associated with each displayed document can be changed via the editing unit 107, the labor and time required for document classification processing that has been performed manually in the past can be greatly reduced. can do.

The document classifying apparatus 100 according to the embodiment of the present invention has, for example, an appearance as shown in FIG. That is, the document classification device 100 according to the embodiment of the present invention is configured by incorporating each element of the document classification device in the computer system. The computer system 40 includes a floppy disk drive 41 and an optical disk drive 43. Then, the floppy disk 42 is inserted into the floppy disk drive 41 and the optical disk 44 is inserted into the optical disk drive 43, and a predetermined read operation is performed.
The document classification program stored in these recording media can be installed in the computer system 40. Further, by connecting a predetermined drive device, for example, using a ROM 45 serving as a memory device or a cartridge 46 serving as a magnetic tape device,
You can also install and read and write data.

Further, the document classification device 100 according to the embodiment of the present invention may be programmed and stored in a computer-readable recording medium. Then, when performing the document classification process, the recording medium is read into a computer system, the program is stored in a storage unit such as a memory in the computer system, and the document classification program is executed by the arithmetic unit, thereby realizing the present invention. A document classification device and a method thereof can be realized. Here, the recording medium includes, for example, a computer-readable recording medium capable of recording a program such as a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, and a magnetic tape.

As described above, it should be sufficiently understood that the present invention includes various embodiments and the like not described herein. Therefore, the present invention must be limited only by the matters specifying the invention according to the claims that are reasonable from this disclosure.

Finally, a specific example of the supplementary processing operation in the document classification processing according to the embodiment of the present invention will be briefly described.

The complementary processing patterns according to the embodiment of the present invention are roughly classified into two types of complementary rules shown in FIGS. 7A and 7B. That is, a supplementary rule A for restoring the omitted keyword (FIG. 7A) and a supplementary rule B for adding a similar expression (FIG. 7B). The complementing rule A is, for example, by complementing “mouse” as a complementing keyword with respect to the complementing keyword “click”, so that even if “mouse” is not included in the original document, the classification rule “ The original document can be classified into a group of "mouse". On the other hand, complementary rule B
In, for example, by complementing the keywords “PC”, “PC”, and “personal computer”, even if the original document uses any of the three keywords, the same applies. Classification can be performed with high accuracy. In addition, since the classification rule only needs to describe any one of the three expressions, the classification rule can be simplified.

In the process of deriving the complement destination keyword in the complement rule A, the if-then rule (if A then add B) is mainly used. Hereinafter, some specific examples of the supplementary rule A will be described.

(1) “wear” → “clothes”, “brew” →
Complements the paired nomenclature, such as "tea", which is usually used only for a specific object.

(2) “Yen appreciation” → “dollar”, “young” →
Complement words that are frequently used in a set, such as “child”.

(3) A word specifying a specific type, such as “beer”, “whiskey” → “alcohol”, is complemented with a word of a more abstract type including the word.

(4) "Keizo Kobuchi" → "Prime Minister", "Ichiro" → "Professional baseball player", "Yukikichi Fukuzawa" → "10,000 yen bill"
As described above, all other words or representative words when any one of the sets that can be paraphrased are included are complemented by using general knowledge, expertise, current events, and slang.

(5) It can be omitted because it is known that the advance information is shared between the sender and the recipient, such as “nearest station” → “Tokyo station”, “dinner” → “curry”. Complete a word.

(6) The rules are complemented by reflecting the world, such as “child + luxury goods” → “TV game”, “girl high school girl + luxury goods” → “PHS”.

(7) When a combination of "A + B" and "C + B" frequently appears, A and C are likely to be similar words and are complemented by each other.

On the other hand, in the complementing rule B, a complement set rule (if A then add ^ A) is mainly used. For example, all other words or representative words in the case where any one of the synonyms is included, such as "quick", "quick", and "quick", are complemented.

When the target document is a conversational sentence,
There are times when neither the if-then rule nor the complement rule can be processed. In other words, this is a case where the target document itself or a document having the attribute of the preceding statement with respect to the target document has information corresponding to the prior knowledge, and the word “Kadosado” included in the target document is replaced with a character string of the prior knowledge. (In English, these are words with the definite article (the). With some exceptions (the earth, etc.)). In such a case, for example, "in the land" → "in Tokyo", "for yesterday's product ..." → "for product A", other more detailed information included in the pre-shared information It is good to replace the statement document with a simple word.

As a result of the complementing process, the target document may belong to a plurality of group classification rules at the same time. For example, now the classification rule is, for example, "include (PC or
If you use "computer" or "include (information source or CM or commercial)", depending on the evaluation function, you may hit both, but in such a case,
(1) By classifying into all groups exceeding the threshold,
Belong to both groups, (2) prioritize classification rules and use the same degree of similarity, etc., to belong to one group, (3) confirm to the user without belonging to either group , Etc. are conceivable.
In the case of the method (1), similar opinions can be confirmed only by looking at one group without confirming all groups, and it can be understood that there are other classification destinations. Is desirable.

As an application field of the present invention, for example, a classification process of a large number of documents obtained from an interview, a questionnaire, or the like at the time of market research can be considered.

[0081]

As described above, according to the document classification device, the document classification method, and the computer-readable recording medium storing the document classification program of the present invention, the keyword or the keyword omitted in the document to be classified is provided. At the same time as supplementing new keywords similar to the extracted keywords, deleting unnecessary keywords in the document, converting the document to include the necessary and sufficient keywords in the document, and then converting the document Perform document classification processing. Further, since attribute information other than keywords can be reflected in the document classification result, highly accurate document classification processing can be performed. Further, since information other than the keyword can be reflected in the document classification result, it is possible to realize a more accurate document classification process.

[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 a flowchart illustrating a document classification method according to the embodiment of the present invention.

FIG. 3 is a flowchart illustrating details of a complementing process in step S300 of FIG. 2;

FIG. 4 is a diagram showing an overview of a document classification device according to an embodiment of the present invention.

FIG. 5 is a schematic diagram illustrating an internal configuration of a document information storage unit according to the embodiment of the present invention.

FIG. 6 is a schematic diagram showing an internal configuration of a classification rule database according to the embodiment of the present invention.

FIG. 7 is a diagram illustrating a supplementary process according to the embodiment of the present invention.

FIG. 8 is a schematic diagram for explaining a conventional document classification process.

[Explanation of symbols]

 Reference Signs List 40 computer system 41 floppy drive 42 floppy disk 43 optical disk drive 44 optical disk 45 ROM 46 cartridge 100 document classification device 101 structured rule database 102 complementary keyword database 60, 103 classification rule database 50, 104 document information storage unit 105 input unit 106 editing unit 107 display unit 110 classification processing unit 111 document analysis unit 112 document structure information extraction unit 113 discrimination unit 114 keyword complement unit 115 similarity calculation unit 116 document classification unit 117 classification rule generation unit

Continuation of the front page (72) Chie Sekimoto 70 Yanagicho, Saiwai-ku, Kawasaki-shi, Kanagawa Prefecture F-term in Toshiba Yanagimachi Plant (reference) 5B075 ND03 NK06 NK32 NR03 NR12 PR06 QM08 UU06

Claims (14)

[Claims] 1. A document classifying apparatus for classifying documents described in a natural language into a predetermined number of groups, comprising: an input unit for inputting a document to be classified and attribute information associated with the document; A document analysis unit that analyzes a document based on morphological analysis and extracts a keyword, and describes, for each group, a predetermined combination including at least one of the keyword indicating the characteristics of the group and the attribute information. A classification rule storage unit that stores a classification rule; a similarity calculation unit that calculates a similarity between each of the classification rules and the keyword or the attribute information of the document; based on the calculated similarity, A document classification unit for finding a group corresponding to a classification rule to which the document is most similar and for associating the document with the group. Classification device. 2. The document classification device, further comprising: a determination unit configured to determine whether the keyword of the document is omitted based on the keyword or the attribute information; and that the keyword is omitted. If it is determined,
2. The document according to claim 1, further comprising: a keyword complementer that restores an omitted keyword to the document and / or adds an expression similar to the keyword and outputs the keyword to the similarity calculator. 3. Classifier. 3. The attribute information includes identification information for identifying a user corresponding to the document, question information corresponding to the document, scene information, creation time of the document, a document category for a proposition, and a statement preceding the document. The document classification apparatus according to claim 1, further comprising one or more pieces of connection information of the document. 4. The classification rule according to claim 1, wherein the classification rule is described including a predetermined combination of at least one of a nominal element, a verbal element, and a modifier element. Document classification device. 5. The document classification apparatus according to claim 1, wherein the groups are formed hierarchically according to an abstraction level of a corresponding classification rule. 6. The document classifying apparatus further includes a display unit for displaying, for each group, a list of documents belonging to the group together with a classification rule corresponding to the group or a similar expression thereof. Claims 1, 2,
6. The document classification device according to 3, 4, or 5. 7. The document classification device according to claim 6, wherein the display unit displays a list of documents that are not associated with any group. 8. The document classification apparatus according to claim 6, wherein the document classification device further comprises an editing unit for instructing another group to change a group associated with each displayed document. Document classifier. 9. The document classification device according to claim 1, further comprising: a classification rule generation unit for inductively generating the classification rule from a document group belonging to the group for each group. The document classification device described in 2, 3, 4, 5, 6, 7, or 8. 10. The method according to claim 1, wherein the similarity calculating unit variably sets the weight of the keyword according to the attribute information. Or the document classification device according to 9. 11. A document classification method for classifying documents described in a natural language into a predetermined number of groups, comprising: an input step of inputting a document to be classified and attribute information associated with the document; A document analysis step of analyzing a document based on morphological analysis and extracting a keyword; and a classification in which a predetermined combination of at least one of the keyword and the attribute information indicating the characteristics of the group is described for each group. A similarity calculating step of calculating a similarity between each of the rules and the keyword or the attribute information of the document; and, based on the calculated similarity, a group corresponding to a classification rule to which the document is most similar. And associating the document with the group. 12. The document classification method, further comprising: determining whether the keyword of the document is omitted based on the keyword or the attribute information; and determining that the keyword is omitted. If it is determined,
12. The document classification method according to claim 11, further comprising: a keyword complementing step of restoring an omitted keyword to the document and / or adding and outputting an expression similar to the keyword. 13. A computer-readable recording medium storing a document classification program for classifying a document described in a natural language into a predetermined number of groups, and inputs a document to be classified and attribute information associated with the document. A document analysis process of analyzing the input document based on morphological analysis and extracting a keyword; and a predetermined process including at least one of the keyword and the attribute information indicating the characteristics of the group. A similarity calculation process for calculating a similarity between the classification rule describing the combination for each group and the keyword or the attribute information of the document; A document classification process for finding a group corresponding to a similar classification rule and associating the document with the group; Computer readable recording medium storing a document classification program for causing a computer to execute the. 14. A computer-readable recording medium storing the document classification program, further comprising: a determination process of determining whether the keyword of the document is omitted based on the keyword or the attribute information. , When it is determined that the keyword is omitted,
14. A keyword compensating process for restoring an omitted keyword in the document and / or adding an expression similar to the keyword and outputting the keyword, and causing a computer to execute these processes. And a computer-readable recording medium storing the document classification program.