JP2000222431A - Document classifying device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To select a category to be appropriately classified and make document information storable according to the category by selecting a category from which a characteristic pattern whose weight is the highest is extracted among collected characteristic patterns and storing document information that is not classified yet in the selected category. SOLUTION: A storing means 10 classified document information having a keyword set having a keyword characterizing the contents of a document corresponding to a document number, etc., to a corresponding category and stores it. An extracting means 20 extracts a characteristic pattern having a keyword whose rate of being included in the keyword set of the stored document information is high in each category and gives weight. Then, a characteristic pattern collecting means 30 collects character patterns including a part of a keyword set of unclassified document information. A selecting means 40 selects a category from which a characteristic pattern whose weight is the highest among the collected characteristic patterns and stores the unclassified document information.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a method of classifying document information which has not been classified yet (hereinafter referred to as "unclassified document information") in a pre-classified and stored version of document information. The present invention relates to a document classification device that stores documents in this category.

[0002]

2. Description of the Related Art As a technique for performing classification based on statistical information of individual keywords, Japanese Patent Application Laid-Open Nos. 63-214832, 1-188934, 5-54.
037, JP-A-5-342272 and JP-A-6-75995. In these publications, information such as the frequency of appearance of individual keywords in documents belonging to a certain category is used as the degree of contribution to the category, and the degree of contribution of each word included in the unclassified document information is classified for each category. The sum is set as the degree of relevance to each category, classified into the category having the highest degree of relevance, and stored. Japanese Patent Application Laid-Open No. 8-221439 discloses a classification using a neural network.

[0003] Japanese Patent Application Laid-Open No.
JP-A-233706, JP-A-5-234726,
JP-A-9-16570 and the like are available. In the publications described above, a tree for deciding a classification based on the presence or absence of a keyword or other document information is configured in advance, and the classification is determined using the tree.

[0004]

When classification is performed based on statistical information of individual keywords, it is difficult to determine the combination of keywords that contributed to the classification as a result of adding the contribution of each keyword to the category. Therefore, it is difficult for the user to judge the quality of the classification. For example, a document including both the keyword a and the keyword b is included in a certain category (for example, category A) of a plurality of categories, and any of the keywords a and b is included in the other categories (that is, categories other than the category A). Classify documents that contain other keywords along with one or the other,
Consider a case where an unclassified document including both keywords a and b is classified.

When calculating the contribution of each keyword, if a weighting method is used such that the weight decreases according to the total number of documents appearing in the keyword, an unclassified document including both the keyword a and the keyword b is used. The degree of relevance of the information to category A is calculated to be small, and the keywords included in the unclassified document information are classified into other categories depending on the contribution degree of the keyword with a small total number of appearing documents. there were.

On the other hand, in the method of constructing a classification decision tree, in order to construct a decision tree having no redundancy, for example, some keywords (keyword a, keyword a, If you try to classify documents that contain any of the keywords b)
The document that was accidentally classified into category B happens to be keyword b.
If the decision tree contains only the keyword b, for example, if there is a keyword b, the classification tree is classified into the category B, and the knowledge for classifying a document including only the keyword a cannot be obtained.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems in the prior art, and a document information set classified into several categories and unclassified document information are provided. An object of the present invention is to provide a document classifying apparatus which can cope with a case in which a keyword appears as a combination and which can select and store an appropriate category to be classified.

[0008]

SUMMARY OF THE INVENTION A document classification apparatus according to the present invention classifies document information having a document number and a keyword set having a keyword characterizing the contents of a document corresponding to the document number into corresponding categories. Storage means for storing, and information for characterizing each category of the storage means, wherein a feature pattern having a keyword which is included in a keyword set of document information stored in the category at a high rate is stored in each category of the storage means. Extraction means for extracting each of the extracted feature patterns and weighting the extracted feature patterns, and feature pattern collection means for collecting, for each category, feature patterns including a part or all of a keyword set of document information that has not been classified. Among the feature patterns collected by the feature pattern collection means, Select a category extracting the symptoms pattern, it is characterized in that a selection means for storing the selected the document information that has not yet been classified into categories.

According to the document classification device of the present invention, the weight given to the extracted feature pattern is set to the document information having a keyword set including the feature pattern among all the document information stored in the category corresponding to the feature pattern. The ratio is characterized by:

[0010] The document classification device according to the present invention sets the weight given to the extracted feature pattern to the ratio of the documents belonging to the category corresponding to the feature pattern among the documents of all categories having the keyword set including the feature pattern. It is characterized by having done.

[0011] The document classification apparatus according to the present invention is characterized in that, in a set of predetermined document information stored in the storage means, conditional entropy for a feature pattern of a category is weighted to be given to the feature pattern. is there.

[0012] The document classification device according to the present invention is characterized in that a weight obtained by multiplying the weight given to the extracted feature pattern by the number of keywords constituting the feature pattern is used as the weight given to the feature pattern. is there.

[0013] A document classifying apparatus according to the present invention comprises: storage means for classifying and storing document information having a document number and a keyword set having a keyword characterizing the content of the document corresponding to the document number into a corresponding category; Extraction means for extracting, for each category, information for characterizing each category of the storage means, the feature pattern having a keyword having a high rate of being included in a keyword set of document information stored in the category. A characteristic pattern collection unit that collects, for each category, a characteristic pattern including a part or all of the keyword set of the document information that has not been classified, and it is appropriate that the document information that has not been classified belongs to the category. The power probability is calculated for each category, and the category having the highest probability is selected. It is characterized in that a selection means for storing the selected the document information that has not yet been classified into categories.

[0014] In the document classification apparatus according to the present invention, the selection unit includes a category in which a feature pattern having a high weight is extracted from the feature patterns collected by the feature pattern collection unit, or the document information that has not been classified is a category. A category from which a characteristic pattern having a high probability of belonging to the category is selected is selected, and information on the category selected by the selection means is displayed.

[0015] The document classification apparatus according to the present invention is characterized in that it is provided with document information search means for searching whether or not the input characteristic pattern is present in the document information stored in the storage means.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 Hereinafter, an embodiment of the present invention will be described. FIG. 1 is a diagram for explaining the document classification device according to the first embodiment. In the figure, 10 is storage means, 20 is extraction means, 30 is feature pattern collection means,
A selection means, 40 is a selection means, and 50 is a document information input means.

The storage means 10 is, for example, a hard disk,
It has an information storage medium having a high storage capacity, such as a magnetic recording medium such as a flexible disk and a magneto-optical recording medium such as an MO disk. The storage means 10 has several categories in advance. Each category stores a document ID which is a document number to which a number corresponding to the document is assigned in advance, and document information including a keyword set having at least one keyword characterizing the content of the document.

The classification into categories may be performed by using any clustering algorithm such as the K-means method, or may be performed manually. As a simple example of the storage data format, for example, document ID 1: keyword 1, keyword 2. . . Document ID 2: Keyword 1, Keyword 2. . . As described above, there is a type in which a plurality of pieces of document information are collected into one file as a set of document information.

The contents of the document itself are described in Document I of this document.
It is stored so that it can be taken out in association with D. For example, information on the contents of the document may be stored in each category of the storage unit 10 corresponding to the document, or may be stored in a storage unit (not shown) separate from the storage unit 10.

The keyword set may be manually added in advance from the contents of the document information, or may be set in advance by a machine (not shown).
For example, the content of the document information may be analyzed, and the result may be used to provide the document information. Also, if a term having the same meaning is expressed as one unified keyword, such as expressing "computer" or "computer" as a unified keyword such as "computer", the so-called swaying of the notation and synonym Considered and unified keywords can be assigned.

The feature pattern extracting means 20 includes the storing means 1
From the keyword set of the document information stored in 0, a feature pattern that is a conjunction of one or more keywords characterizing the document information of each category is extracted for each category, and a weight is assigned to the extracted feature pattern. Is what you do. The above processing can be performed before the classification is actually performed by giving the unclassified document information.

The unclassified document input means 50 sends to the feature pattern collection means 30 a set of keywords included in the document information (referred to as unclassified document information) of a document not yet classified in the storage means 10. The feature pattern collection unit 30 collects, for each category, feature patterns that include a part or all of the keyword set of the unclassified document information among the feature patterns extracted by the feature pattern extraction unit 20.

The selecting means 40 determines the feature pattern having the largest weight among the feature patterns of each category collected by the feature pattern collecting means as the relevance to the category. After selecting the category having the highest relevance, Store the unclassified document information in the selected category.

With this configuration, it is possible to sequentially store the unclassified document information in a category according to the content of the document.

The feature patterns extracted from each category by the feature pattern extracting means 20 include, for example, a feature pattern having a high category support rate, a feature pattern having a high category certainty factor, a feature pattern having both of these two high features, and the like.

Here, the support rate and the certainty factor are defined as follows. The category approval rate refers to document information having a feature pattern having at least one keyword.
It is a ratio of document information provided with a keyword set including all keywords constituting the feature pattern in the document information set belonging to the category. Category confidence is
The ratio of the document information belonging to the category among all the categories storing the document information including the keyword set including all the keywords constituting the feature pattern. As described above, a feature pattern having a high support rate and a high degree of certainty can be information that characterizes document information belonging to the category.

As the characteristic pattern extracted by the characteristic pattern extracting means 20, for example, the number of stored document information is 50.
It is assumed that a feature pattern having both “computer” and “network” as keywords corresponding to category C is extracted with a support rate of 40% and a certainty factor of 80%. This is of all the document information stored in category C
Those having a keyword set including the keywords "computer" and "network" correspond to 40% of all the document information of category C, that is, 20 pieces of document information, which are stored in all categories in the storage means 10. In the obtained document information, 80% of the document information having a keyword set including the keywords “computer” and “network” is document information stored in category C.

In the first embodiment, the feature pattern extracting means 20 uses, for example, a feature pattern having a minimum support rate equal to or higher than a predetermined value and a minimum certainty factor equal to or higher than a predetermined value in a certain category. The extraction will be described as an example.

FIG. 2 is a diagram for explaining an example of a specific configuration of the storage means 10 and the characteristic pattern extraction means 20. In the figure, components denoted by the same reference numerals as those in FIG. 1 are the same or corresponding components. In the figure, 11 is
The recording unit is a recording unit that classifies and stores a document ID and a keyword set corresponding to the document ID into categories, and the recording unit 11 is, for example, a hard disk device, a disk drive device equipped with an MO disk or a floppy disk, or the like. Reference numeral 12 denotes a document ID recorded in the recording unit 11 and a document I
A keyword set corresponding to D is read and stored for each category, and a document ID stored in a category specified by the control unit 21 described later, and document information having a keyword set corresponding to the document ID are generated as candidate pattern generation described later. A document information storage unit that outputs to the unit 24.

When a category and a pattern are specified by the multi-support pattern set generation unit 22 or the high confidence pattern output unit 23, the document information storage unit 12 stores a keyword set including the specified pattern in the specified category. Is calculated and output to the multi-support pattern set generation unit 22 or the high confidence pattern output unit 23. When checking whether or not a specified pattern appears in each piece of document information, the document information storage unit 12 determines whether a part or all of a keyword set constituting the document information matches the pattern. inspect.

Therefore, in order to perform this check efficiently, a pattern that may be a subset of the keyword set of each document information is narrowed down using a hash table or a hash tree, or each pattern, document information, , The keyword set may be arranged or represented by a bit pattern. The document information storage unit 12
May be connected to a predetermined computer network, and the document information may be read from another recording unit via the computer network.

24 generates a candidate pattern composed of one keyword for each keyword corresponding to a document ID of a certain category supplied from the document information storage unit 12, and sends the candidate pattern to the multi-support pattern set generation unit 22. Output and supplied from the multi-support pattern set generation unit 22,
Assuming that the number of keywords constituting the multi-support pattern of the multi-support pattern set is n, (n + 1) is calculated from any two multi-support patterns having (n−1) keywords in common.
1) A pattern composed of keywords is generated, and a new candidate pattern set is generated by using, as candidate patterns, a pattern in which arbitrary n keywords of the pattern are included in the multi-support pattern set. 22 is a candidate pattern set generation unit that outputs to 22.

Reference numeral 22 denotes, for each candidate pattern supplied from the candidate pattern set generation unit 24, in the category specified by the control unit 21, the number of document information in which the candidate pattern appears and all the document information belonging to the category. Is calculated using the document information storage unit 12, and a pattern whose support rate is equal to or greater than a predetermined threshold (or minimum support rate) is selected as a new multi-support pattern. Then, a set of those multi-support patterns is output to the high confidence pattern output unit 23 and the candidate pattern set generation unit 24.
Is a multi-support pattern set generation unit that outputs the multi-support pattern set.

Reference numeral 23 denotes, for each pattern supplied from the multi-support pattern set generation unit 22, the number of pieces of document information in which the pattern appears in the cluster designated by the control unit 21 and the document information in which the pattern appears in all the clusters. A certainty factor, which is a ratio to the number of pieces of information, is calculated using the document set storage unit 12, and the certainty factor is equal to or more than a predetermined minimum certainty factor from the pattern set supplied from the multi-support pattern set generating unit 22. This is a high confidence pattern output unit that selects a pattern and outputs the pattern. A control unit 21 controls the document set storage unit 12, the multi-support pattern set generation unit 22, the high confidence pattern output unit 23, and the candidate pattern set generation unit 24.

With the configuration shown in FIG. 2, a feature pattern having a high support rate can be extracted from the multi-support pattern set generating section 22 and a feature pattern having a high certainty can be extracted from the high certainty pattern output section 23. can do.

As described above, a feature pattern having a high support rate and certainty factor can be information characterizing document information belonging to the category. Document ID corresponding to unclassified document information,
Not only is it useful for classifying the corresponding keyword set into appropriate categories, but when performing a keyword search, priority can be given to the category related to the content of the keyword, thereby reducing the search time Becomes possible.

In general, when a threshold value of a support rate to be extracted (minimum support rate) and a threshold value of a certainty factor to be extracted (minimum certainty factor) are given, a plurality of feature patterns corresponding to a certain category are extracted. Is done. However, if the minimum support rate or the minimum certainty factor is too high, no feature pattern may be extracted. Conversely, if the minimum support rate or the minimum certainty factor is too low, it takes a long time to extract a feature pattern, or it becomes easy to extract more feature patterns than necessary.

Therefore, if the minimum support rate and the minimum certainty factor are set to high values in advance, and if the category corresponding to the feature pattern cannot be extracted, the minimum support rate and the minimum certainty factor are determined until an appropriate number of feature patterns can be extracted. It is desirable that the degree is gradually reduced. In the extraction of the feature pattern, by limiting the number of keywords included in the feature pattern, it is possible to suppress the extraction of a too complicated pattern.

The characteristic pattern extracting means 20 gives each extracted characteristic pattern a weight that can be compared with each other. As the weight w _ij of each feature pattern p _ij extracted in the category C _i , it is conceivable to use a support rate, a certainty factor, or a conditional entropy of the feature pattern p _ij for the category C _i . _{Alternatively,} as the weight w _ij of each feature pattern p _ij extracted in the category C _i , the number of keywords constituting the feature pattern p _ij itself is used as the weight w w of the individual feature pattern p _ij of the category C _i.
ij may be used.

Here, a set of predetermined document information stored in the storage means 10 is referred to as a document set D. The document set D is part or all of the document information stored in the storage unit 10. If conditional entropy Ent _i (D | p) for pattern p of category C _i in document set D is:

[0041]

(Equation 1)

Here, D _p is a set of documents in the document set D that includes the pattern p as a part of the keyword set.

[0043]

(Equation 2)

It is assumed that Here, Ent _i (D) is the entropy of cluster C _i in document set D,

[0045]

(Equation 3)

Then, from (3a) and (3b), Ent is obtained.
_i (D) is represented by the following equation.

[0047]

(Equation 4)

The feature pattern extracting means 20 assigns weights to the extracted feature patterns according to any of the indices. The above processing can be performed before classification is actually performed by giving unclassified document information.

The feature pattern collecting means 30 collects a feature pattern including a part or all of the keyword set of the unclassified document information for each category. For example, as a feature pattern of category C, a feature pattern consisting of "computer" and "network" as a feature pattern, a feature pattern consisting of "computer" and "control", and a feature pattern of "computer" Is extracted, and the keyword set of unclassified document information is “computer, processor, network, computation,
Program ". At this time, the feature pattern collection unit 30 collects the “computer”, the “network”, and the “computer” as the feature patterns of the category C related to the keyword set of the unclassified document information.

The selecting means 40 is the feature pattern collecting means 30
Of the feature patterns of each category collected by the above, the one with the largest weight is regarded as the relevance to that category, and after selecting the category with the highest relevance, the unclassified document information is classified into the selected category Store.

For example, the certainty factor is used as the weight of the feature pattern, and the feature patterns collected by the feature pattern collection means 30 correspond to the “computer” and the “network” with the certainty factor of 80% corresponding to the category C; Similarly, if the feature pattern corresponds to “computer” having a certainty factor of 50% in category C and corresponds to “processor” having a 60% certainty factor in category D, the selecting unit 40 stores the unclassified document information. It is stored in category C, which is one of the ten categories.

As described above, according to the first embodiment, a feature pattern having one or more keywords characterizing document information of each category is extracted for each category, and can be compared with the extracted feature pattern. Weights (e.g., support rate, certainty factor, conditional entropy, weight of feature pattern multiplied by the number of keywords, etc.) are given, and among the weights assigned to a plurality of feature patterns extracted from each category, The feature having the largest weight is stored by selecting the category with the largest value as the relevance to the category, selecting the category with the largest relevance, and classifying the unclassified documents into the selected category. It is possible to select a category and store unclassified document information in this category.

Furthermore, a combination pattern of keywords characterizing document information of each category is extracted from the stored set of document information, and classification is performed using the extracted patterns. Since it is possible and the basis of the classification is the combination pattern of the keywords, it is easy for the user to intuitively understand, and it is easy to judge the quality of the classification.

When the certainty factor is used as the weight assigned to each feature pattern, unclassified document information is stored in a category in which document information having a keyword set including the feature pattern is biased.

When the same feature pattern is collected in a plurality of categories, if the support rate is used as the weight of the feature pattern extracted in each category, a feature in which the ratio of the document information which appears most in a certain category is high. Unclassified document information is stored in the category corresponding to the pattern.

When conditional entropy is used as a weight assigned to each feature pattern, this corresponds to considering both the support rate and the certainty factor. In this case, unclassified document information is stored in a category corresponding to a feature pattern in which the ratio of document information that appears in a category is high and that does not often appear in other categories. Therefore, a category in which a large number of keywords included in the keyword set of the unclassified document information are stored, and the percentage of the document information including the keywords included in the keyword set of the unclassified document information is high, is stored in the category. Therefore, unclassified document information can be stored in a more appropriate category.

Further, the weight of any one of the above-mentioned characteristic patterns is multiplied by the number of keywords constituting the characteristic pattern, and this is set as a new weight, and the selecting means corresponds to the one having the maximum new weight. If the configuration is such that a category for storing a feature pattern to be stored is selected, unclassified document information is stored in a category corresponding to a feature pattern having a large number of keywords constituting the feature pattern.

Embodiment 2 Feature pattern extraction means 20
After extracting a feature pattern characterizing each category from the document information set stored in the storage unit 10 for each category, instead of assigning a weight to each feature pattern as in the first embodiment, The number of pieces of document information including the pattern is assigned to each of the corresponding category and the total of all other categories, and the selecting unit 40 determines a probability (referred to as an estimated probability) that it is appropriate that the unclassified document belongs to a certain category. In addition to the calculation, the estimated probability is regarded as the degree of relevance to the category, and an unclassified document is stored in the category having the highest degree of relevance.

The estimated probability in each category is calculated as follows. First, if a feature pattern corresponding to a keyword included in unclassified document information is not collected at all in a certain category, the estimated probability is set to 0. Next, if only one feature pattern is collected, the certainty factor is directly used as the estimated probability. On the other hand, when a plurality of feature patterns are collected corresponding to a certain category, various methods can be considered as a method of calculating the estimated probability for the category.

For example, a configuration in which the certainty factor of each feature pattern is used as the estimated probability is the same as the example described in the first embodiment.

As another method, an estimation method focusing on the fact that a plurality of feature patterns collected corresponding to a certain category appear simultaneously can be considered. The estimation method is described below.

First, as a simple case, in a category C _i in a set D of document information in which the number of document information stored in the storage means is N, a feature pattern pa composed of _a keyword _a and a keyword b are formed. It is assumed that the feature pattern p _b has been collected. Also, sets of document information having a keyword set including each feature pattern are D _pa and D _pb , respectively.

[0063]

(Equation 5)

Assume that Estimated probability belonging unclassified document containing this case a pattern p _a and p _b at the same time the category C _i is expressed as follows.

[0065]

(Equation 6)

Here, N (X) is the number of documents belonging to the document information set X. Here, for example, N (C _i ∩D _pa ∩
D _pb ) can be actually calculated from the document information. However, there is a case where such document information does not exist by chance, and even if it exists, the number of the document information is small and has no statistical significance. There are cases. Here, a method of indirectly calculating the number of documents of each feature pattern in the corresponding category and all other categories is considered. Hypothetically therefor, D _pa and D _pb is C _i and

[0067]

(Equation 7)

Are independent of each other. At this time,

[0069]

(Equation 8)

The conditional probability is obtained from the number of documents appearing in the category corresponding to each feature pattern and all other categories, and the total number of documents in each category.

[0071]

(Equation 9)

Using this, P (C _i | D _pa ∩D _pb ) can be calculated as follows:

[0073]

(Equation 10)

However,

[0075]

[Equation 11]

It is assumed that As described above, the estimation probability can be obtained by Expression (9).

Next, consider a case where at least one of the feature patterns collected in a certain category has a plurality of keywords. If a plurality of feature patterns do not have overlapping constituent keywords, it can be considered in the same manner as described above. If there are duplicate keywords, it is necessary to consider how much the keywords contribute to the probability.

For example, in the category C _i , the keyword c,
It is assumed that a feature pattern _{pc, d} consisting of d and a feature pattern pc, e, f consisting of keywords _{c, e, f} have been collected. At this time, if the calculation is directly performed according to the equation (9), the contribution of the keyword c is considered redundantly. In order to avoid this, it is necessary to consider the contribution of each keyword, but as an assumption to facilitate the estimation, for example, each keyword in a certain feature pattern is independent, and each contribution is Suppose they are equal. According to this assumption, the contribution of the keywords c and d included in the feature pattern _{pc and d} to the expression (9) is

[0079]

(Equation 12)

Is obtained. Similarly, in the feature pattern _{pc, e, f} , the contribution of the keywords c, e, f is

[0081]

(Equation 13)

Is calculated. When a keyword is included in a plurality of collected feature patterns, the contribution of the keyword is, for example, the maximum value of the contribution calculated for each feature pattern. Category C in this way
_{If the} respective contributions R _i (k) are calculated as described above for all the keywords k included in the feature pattern p collected in _i , the estimated probability becomes

[0083]

[Equation 14]

Is calculated. As described above, according to the second embodiment, from the feature patterns collected for each category according to the unclassified document information, the estimated probability that the unclassified document should be appropriately assigned to each category is calculated for each category. Then, the category having the highest estimated probability is selected, and the unclassified document information is stored in this category. Therefore, the unclassified document information can be stored in an appropriate category. In particular, if the estimated probabilities are calculated in consideration of the constituent keywords of the feature patterns collected for each category, the amount of calculation is larger than in the first embodiment, but new keyword combinations that did not frequently appear in the past are also supported. Classification is possible.

Embodiment 3 FIG. 3 is a diagram for explaining the configuration of the document classification device according to the third embodiment. In the drawings, the components denoted by the same reference numerals as those in FIGS. 1 and 2 are the same or equivalent. Reference numeral 60 denotes a classification determination interface, and reference numeral 70 denotes a document information search unit. The classification determination interface 50 has a display screen. The display screen displays a category selected by the selection unit 40 and information related to the selection, for example, a feature pattern collected for each category, a weight for the feature pattern, or an unclassified information. And the degree of association between the document information and the category. The user determines a category to be finally classified based on the displayed information. Further, the user can select a characteristic pattern as necessary, send it to the document information search means 70, and refer to the result.

The document information retrieving means 70 retrieves document information having a keyword set including the given feature pattern from the set of document information stored in the storage means 10 and displays the result on the classification decision interface 60. Although it is not necessary, if the search target is narrowed down to a specific category, and if an AND search, an OR search, and the like relating to a plurality of feature patterns can be executed, a more efficient search can be performed.

As described above, according to the third embodiment, it is possible for the user to finally determine the classification while referring to the information on the classification selection. In particular, the feature patterns collected in each category are a series of keywords, and are intuitive to the user. Therefore, it is necessary to grasp the features of each category while performing document information search using feature patterns as necessary. The classification can be finally determined.

The embodiments described above are shown as one embodiment of the present invention, and the present invention should not be limited to these embodiments. The invention of this application includes what is described in the claims or an equivalent thereof.

[0089]

According to the document classification device of the present invention,
Storing means for classifying and storing document information having a keyword set having a document number and a keyword characterizing the content of the document corresponding to the document number into corresponding categories; and information for characterizing each category of the storing means. Extracting means for extracting, for each category, a feature pattern having a keyword which is included in a keyword set of document information stored in the category for each category, and assigning a weight to the extracted feature pattern A feature pattern collection unit that collects, for each category, a feature pattern including a part or the entirety of a keyword set of document information that has not been classified, and a feature pattern having the highest weight among feature patterns collected by the feature pattern collection unit. Select the category from which the high feature pattern was extracted, and select the category Since a selecting means for storing the document information that has not yet been classified into can store document information that has not yet been classified into appropriate categories.

According to the document classification device of the present invention, the weight given to the extracted feature pattern is determined by the keyword set including the feature pattern among the keyword sets of all the document information stored in the category corresponding to the feature pattern. , The document information that has not yet been classified can be stored in an appropriate category.

According to the document classification device of the present invention, the weight assigned to the extracted feature pattern is set to the weight of the document belonging to the category corresponding to the category corresponding to the category corresponding to the feature pattern among all the categories having the keyword set including the feature pattern. Since the ratio is set, document information that has not yet been classified can be stored in an appropriate category.

According to the document classifying apparatus of the present invention, in a set of predetermined document information stored in the storage means, conditional entropy for a feature pattern of a category is weighted to be assigned to the feature pattern. Document information that has not been stored can be stored in a more appropriate category.

According to the document classifying apparatus of the present invention, the weight given to the extracted feature pattern is multiplied by the number of keywords constituting the feature pattern, and the weight is given to the feature pattern. Document information that has not been stored can be stored in a more appropriate category.

According to the document classification device of the present invention, storage means for classifying and storing document information having a document number and a keyword set having a keyword characterizing the contents of the document corresponding to the document number into the corresponding category. And extracting, for each category of the storage unit, a feature pattern that is information for characterizing each category of the storage unit and has a keyword having a high ratio included in a keyword set of document information stored in the category. Extraction means, feature pattern collection means for collecting, for each category, a feature pattern including a part or all of a keyword set of document information that has not been classified, and it is preferable that the document information that has not been classified belongs to a category. Is calculated for each category, and the category with the highest probability is Constant and, since a selection means for storing the selected the document information that has not yet been classified into categories, it is possible to store the document information that has not yet been classified into more appropriate categories.

According to the document classifying device of the present invention, the selecting means selects the category in which the feature pattern having the higher weight is extracted from the feature patterns collected by the feature pattern collecting means or the document information which has not been classified yet. Is configured to select a category from which a feature pattern having a high probability that it should belong to the category is appropriate and to display information of the category selected by the selection means. Can be appropriately determined.

The document classification device according to the present invention includes the document information search means for searching whether or not the input characteristic pattern exists in the document information stored in the storage means.
Those using the apparatus can finally determine the classification based on the characteristics of each category while performing document information search using the characteristic pattern.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a document classification device according to a first embodiment.

FIG. 2 is a diagram for explaining a document classification device according to the first embodiment;

FIG. 3 is a diagram for explaining a document classification device according to the first embodiment;

[Explanation of symbols]

10: storage means 20: extraction means 30: feature pattern collection means 40: selection means 50: non-document information input means

Claims (8)

[Claims] 1. A storage unit for classifying and storing document information having a document number and a keyword set having a keyword characterizing the contents of a document corresponding to the document number into a corresponding category, and storing each category of the storage unit. A feature pattern, which is information for characterizing and has a keyword that is included in the keyword set of the document information stored in the category at a high rate, is extracted for each category of the storage unit, and a weight is assigned to the extracted feature pattern. Extracting means for assigning, to each category, a feature pattern collecting means for collecting a feature pattern including a part or all of a keyword set of document information which has not been classified, and among feature patterns collected by the feature pattern collecting means. Select the category from which the feature pattern with the highest weight is extracted, Selecting means for storing document information that has not yet been classified into the selected category. 2. The method according to claim 1, wherein the weight given to the extracted feature pattern is a ratio of document information having a keyword set including the feature pattern to all document information stored in a category corresponding to the feature pattern. 2. The document classification device according to claim 1, wherein: 3. The method according to claim 1, wherein the weight given to the extracted feature pattern is a ratio of a document belonging to a category corresponding to the feature pattern among documents of all categories having a keyword set including the feature pattern. Item 2. The document classification device according to Item 1. 4. The document classification apparatus according to claim 1, wherein, in a set of predetermined document information stored in the storage unit, conditional entropy for a feature pattern of a category is weighted to be assigned to the feature pattern. . 5. The weight given to the feature pattern obtained by multiplying the weight given to the extracted feature pattern by the number of keywords constituting the feature pattern. Document classification device according to the paragraph. 6. A storage unit for classifying and storing document information having a document number and a keyword set having a keyword characterizing the content of a document corresponding to the document number into a corresponding category, and storing each category of the storage unit. Extracting means for extracting, for each category, characteristic patterns having keywords having a high rate of being included in the keyword set of the document information stored in the category, which are information for characterizing, and not yet classified A feature pattern collection unit that collects, for each category, a feature pattern including a part or all of the keyword set of the document information; and a probability that the document information that has not been classified yet should appropriately belong to the category. Calculate and select the category with the highest probability, and still go to the selected category A document selecting device for storing document information that has not been classified. 7. The selection means, wherein, among the feature patterns collected by the feature pattern collection means, a category from which a feature pattern having a high weight is extracted or the document information which has not been classified yet belongs to the category. 7. The document classification apparatus according to claim 1, wherein a category from which a characteristic pattern having a high power probability is extracted is selected, and information of the category selected by the selection unit is displayed. 8. The document classification device according to claim 7, further comprising a document information search unit for searching whether or not the input characteristic pattern exists in the document information stored in the storage unit.