JP2003016105A - Device for calculating degree value of association - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a degree of association value calculating device that improve the accuracy of the calculated degree of association value by improving the degree of association of a keyword corresponding to a proper name to calculate the degree of association value in calculating a value representing the degree of association of one or a plurality of keywords with one or a plurality of documents. SOLUTION: A proper name keyword specifying means 2 specifies a keyword corresponding to a proper name on the basis of a set proper name keyword specification condition, and degree of association calculating means 3 and 7 improve the degree of association of the keyword specified by the proper name keyword specifying means 2 with each keyword about a document group to calculate the degree of association value of a keyword group with the document group.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a degree-of-association value calculating device for calculating a value indicating the degree of relevance of one or more keywords and one or more documents, and in particular, the degree of relevance of keywords corresponding to unique names. And a technique for calculating a relevance value by increasing the value.

[0002]

2. Description of the Related Art For example, a document search device for searching a document related to a document (seed document) held by a user has been studied. As an example, in the "recording medium recording the related document search device and the related document search program" described in Japanese Patent Laid-Open No. 10-260972, a document to be searched by a related word with a degree of relevance extracted from a seed document Search and
The degree of relevance of a document is calculated and output based on the degree of relevance of a related word included in the retrieved document.
Here, the method of calculating the degree of association of related words is determined by the frequency of appearance of words (related words) included in the seed document and the search target document.

However, when the degree of relevance is determined based only on the frequency of appearance of words, for example, a value having a low degree of relevance of a related word that appears to be of high importance to the user is used. However, there is a problem that an appropriate degree of association is not always calculated. In particular, a proper name such as a person's name, place name, company name (company name), etc. is usually a word that characterizes a document and is considered to have a large amount of information, but when the degree of association is calculated based only on the frequency of occurrence. Did not always obtain a high degree of association.

Further, for example, a document group classification device has been studied which automatically calculates the distance between documents in a document group and automatically classifies the document group based on the calculation result. If the distance between documents is short, the degree of association is high, and if the distance between documents is long, the degree of association is low. It can be considered that the distance between documents corresponds to the degree of association between documents.

In such a document group classification device, as a method for obtaining the degree of association between two documents, for example, by comparing with a dictionary or by a generally known n-gr.
The words included in each document are cut out by using the am method, the frequency of appearance of the cut out words is counted by a method defined by each person, and the value obtained by multiplying the counted value of each word between two documents is calculated. The sum of the words is 2
The degree of relevance between two documents is used.

[0006] However, in such a method, for example, a word without a feature becomes a target for counting the appearance frequency for calculating the degree of association, like a word with a feature, and the target word Since the range is widened, there are problems that the accuracy of classifying documents deteriorates and that the time required to calculate the degree of association increases.

[0007]

As described above, in the conventional document search apparatus or the like, when calculating a value indicating the degree of association of one or more keywords and one or more documents, for example, all of the keywords Since the relevance value is calculated by equally calculating the appearance frequency and the like for the word, there is a problem that a highly accurate relevance value cannot be obtained.

The present invention has been made in view of such conventional circumstances. When calculating a value indicating the degree of association between one or a plurality of keywords and one or a plurality of documents, the relation between the keywords corresponding to the proper names is calculated. An object of the present invention is to provide an association degree value calculation device or the like that realizes, for example, increasing the accuracy of the calculated association degree value by increasing the degree and calculating the association degree value.

[0009]

In order to achieve the above object, in the relevance value calculating apparatus according to the present invention, a keyword group composed of one or a plurality of keywords and a document group composed of one or a plurality of documents. As a value representing the degree of relevance for each keyword included in the keyword group and the relevance value for each keyword related to the document group (relevance value for each keyword) for all keywords, the following value is calculated. Then, the relevance value is calculated.
That is, the unique name keyword specifying means specifies the keyword corresponding to the unique name based on the set unique name keyword specifying condition, and the relevance value calculating means specifies the keyword specified by the unique name keyword specifying means and the keyword related to the document group. The degree of relevance for each is increased to calculate the degree-of-relevance value for the keyword group and the document group.

Therefore, by increasing the degree of relevance for each keyword related to a document group and a keyword identified as a unique name, for example, a keyword that characterizes a document and is associated with a unique name that is considered to have a high amount of information is identified. The relevance value for each keyword can be calculated using the keyword-related relevance value for each document group as a value having a high relevance value, and thus the accuracy of the calculated relevance value can be increased.

Here, the number of keywords included in the keyword group may be various, may be 1, or may be plural (two or more). Similarly, the number of documents included in the document group may be various, may be 1, or may be plural (two or more). As described above, in this specification, with respect to words and phrases using the expression “group” such as a keyword group and a document group,
The case where the “group” includes only one element such as a keyword or a document is also included.

As the keyword, words having various parts of speech may be used, and in the present invention, the relevance value is calculated in consideration of the importance of the keyword corresponding to the proper name. Various documents may be used as the document. Note that a document usually includes a plurality of words and phrases having various parts of speech. Further, as the document, for example, a document to which a keyword is added as an index can be used.

As the value representing the degree of association between the keyword group and the document group, for example, a value considered to represent the degree of association between the keyword group and the document group is used. It should be noted that normally, a setting in which a higher relevance value has a higher relevance is considered to be used, but conversely, a setting in which a lower relevance value has a higher relevance may be used.

Further, as the relevance value of each keyword included in the keyword group and each keyword related to the document group,
It is possible to calculate for each keyword. For example, a value indicating whether or not each keyword appears in a document group using a “1” value or a “0” value, or each keyword is A value representing the number of appearances in the document group, a value representing the frequency of appearance, or the like can be used.

In addition, for example, when a keyword group is extracted from a certain document group, each keyword included in the keyword group may be set to a value according to the appearance status in the document group. . In this way, when a value according to the appearance status of each keyword in a certain document group is set for each keyword, the degree of association between each keyword and another document group is calculated in consideration of the value. As a result, the relevance value for each keyword between the certain document group and the other document group can be calculated.

To calculate the sum of the relevance value of each keyword for all keywords, when the keyword group includes a plurality of keywords, the relevance value of each keyword is summed for all the keywords. When the keyword group includes only one keyword, the relevance value for each keyword calculated for the one keyword corresponds to the sum of all the keywords.

Various words and phrases may be used as the proper name, for example, a person's name, place name, company name, date, time, date and time, product name, etc. can be used. In a preferred embodiment, as the proper name referred to in the present invention, for example, a set of words or phrases generally classified as proper nouns can be used as it is, or a part of the set can be used.

Further, various conditions may be used as the set unique name keyword specifying condition, and various methods are used to specify the keyword corresponding to the unique name based on the unique name keyword specifying condition. You may be asked. The unique name keyword specifying condition may be set in advance in the memory, for example, the contents set in the memory may be rewritable by the user, or, for example, when calculating the relevance value. A mode in which the unique name keyword specifying condition is set by the user may be used.

Specifically, for example, a method of setting a word to be detected as a proper name as a proper name keyword specifying condition in a memory in the form of electronic data and specifying a keyword matching the phrase as a proper name. Or, for example, a dictionary having information about a proper name as a proper name keyword specifying condition is set in a memory in the form of electronic data, and a keyword matching a phrase of the proper name based on the contents of the dictionary corresponds to the proper name. Or a word or phrase such as “stock” as a proper name keyword specifying condition is set in the memory in the form of electronic data and matches the word or phrase with a suffix or prefix such as “stock”. How to specify a keyword as being equivalent to a proper name, or electronic terms such as a place name as a proper name keyword specifying condition Set in the memory in the form, or the like can be used how to identify the keywords that match words like the place names containing terms including the head and tail as corresponding to a specific name.

The analysis of the words included in the document can be performed by using, for example, a generally known morphological analysis. Further, the correspondence between the document and the word / phrase included in the document may be performed in various modes, and as an example,
It is possible to use a mode in which the identification information of one or a plurality of documents including each word is stored in association with each word and a set of attribute information such as a part of speech of each word.

Further, in the relevance value calculating device according to the present invention, as an example, the relevance value calculating means sets the keyword specified by the proper name keyword specifying means and the relevance value for each keyword related to the document group as the calculated value. By comparing and correcting to a value indicating that the degree of association is high, the degree of association for each keyword related to the keyword and the document group is increased.

With such a configuration, the relevance value can be calculated in consideration of keywords other than the keyword specified by the proper name keyword specifying means. In this case, the keyword specified by the proper name keyword specifying means can be calculated. The accuracy of the calculated relevance value can be improved by correcting the relevance value for each keyword to a value indicating that the relevance value is higher than the calculated value.

Here, the term "keyword relevance value associated with the keyword and document group identified by the unique name keyword identifying means is compared with the calculated value to correct the value to indicate that the degree of relevance is high. If a higher degree value indicates a higher degree of relevance, the keyword-related relevance value for the keywords identified as corresponding to the proper name when not corrected is set to a value larger than the keyword-related degree value. Say to change. Note that various degrees may be used as the degree of correcting the keyword-related relevance value, for example, a mode in which a magnification for correcting the keyword-related degree value is set in advance, or such a magnification Can be set by the user.

Further, in the relevance value calculating device according to the present invention, as another example, the relevance value calculating means sets the keyword specified by the proper name keyword specifying means and the relevance value for each keyword relating to the document group to non-zero. On the other hand, by setting the relevance value for each of the other keywords to zero, the relevance for each keyword related to the keyword and the document group is increased.

With such a configuration, when the relevance value for the keyword group and the document group is calculated, only the relevance value for each keyword specified by the proper name keyword specifying means for each keyword is non-zero (that is, zero). Value), and only the non-zero keyword-related relevance value is calculated. Therefore, for example, the accuracy of the relevance value related to the keyword group and the document group is increased to calculate the relevance value. It is possible to reduce the amount of calculation and time required for.

Further, in the degree-of-association value calculation apparatus according to the present invention, the category information storage means stores information on one or a plurality of categories composed of a plurality of unique names of the same kind,
The category designation receiving means receives the designation of the category, and the proper name keyword specifying means, based on the category information stored in the category information storing means, the keyword corresponding to the proper name included in the category received by the category designation receiving means. Specify.

Therefore, by increasing the degree of relevance for each keyword related to the keyword and the document group that are identified as corresponding to the proper name included in the designated category, the keyword is included in the designated category in response to the user's request, for example. The degree of relevance of the keyword corresponding to the proper name can be increased to calculate the degree-of-relevance value, and thus the accuracy of the degree-of-relevance value calculated by reflecting the user's request can be increased.

Further, a plurality of keywords such as "Mitsui" included in a plurality of categories such as a person name category and a company name category, and a plurality of keywords such as "Hitachi" such as a place name category and a company name category. For the keywords included in the category, all the relevance of the same keyword included in the plurality of categories may be increased, but if the category is specified as described above, the It is possible to prevent the degree of relevance from being increased for keywords included in categories other than the specified category, even with the keywords (words).

That is, even if the same word or phrase is used in a plurality of categories as a person's name or a company name such as "Mitsui", it is possible to identify it as a word or phrase different for each category. In addition, embodiment examples (1) and (2) for supporting such category identification are shown. In the embodiment example (1), when the same phrase as the proper name (for example, proper noun) once determined appears again, the category of the same phrase is regarded as the same category as the previous one.
As a specific example, if there is a sentence "... according to the contents that Kunitoshi Mitsui announced the other day ..., Mitsui puts out its own plan ...", the second time "Mitsui" appears in the middle of the sentence.
(“Mitsui” of “Mitsui is unique…”) does not know whether it is a personal name or a company name, but for example, the first “Mitsui” (“Mitsui” of “Kunitoshi Mitsui…”) is the category of personal name. If it is analyzed that it belongs, by storing this in the memory, when you are confused about the category of "Mitsui" for the second time, refer to the previous analysis result and belong to the category of personal name. It is possible to specify a category as a thing. In the embodiment example (2), the category is specified by the appearance pattern of the word. As a specific example,
As for "Mitsui Bank", "Mitsui" is followed by "Bank", so it is judged to be the company name.

It is to be noted that which category the same phrase belongs to is determined by, for example, previously adding information on the category to which each phrase belongs, or by analyzing the meaning of each phrase and the like. It can be determined by estimating the category to which the word belongs.

Here, various categories may be used as the category composed of a plurality of similar unique names. For example, a person's name, a person's name of the same kind, a company name, a company name of the same kind, a product name, a kind of the same kind. Various categories such as product name, domestic place name, same region place name, same country place name, flower name, animal name can be used. As the same type of person name, for example, a historical person's name, the name of a researcher, or the like can be used, and as the same type of company name, for example, the name of a food manufacturer, the name of an automobile manufacturer, or the like can be used. The same applies to the other cases.

Specifically, for example, "A", which is the name (name) of a plurality of different persons who are similar in terms of the person's name,
Categories including “B”, “C”, ... (where A, B, and C are person names), and the names of a plurality of different companies that are similar in terms of the company name of the copying machine manufacturer, for example, “Company A” ,
Categories including “Company B”, “Company C”, ... (Company names of A, B, and C are companies) and a plurality of different copier products that are similar in terms of the product name of the copier, for example. Name "A
, Etc. (machine A, machine B, machine C are product names) and the like can be used.

The number of categories may be various, may be 1, or may be plural (two or more). In addition, as the information regarding the category, for example, information that identifies each of a plurality of proper names included in each category, information that identifies that proper names included in the same category belong to the same category, or the like is used. .

Further, as the category information storage means, a memory or the like for storing information in the form of electronic data can be used. Further, as the category designation receiving means, for example, when the category designation is received from the user, a keyboard or a mouse operated by the user to receive the input for designating the category can be used.

Further, for example, when the user inputs a keyword when calculating the degree-of-association value, it is determined that the category including the phrase corresponding to the keyword is designated and the category designation receiving means receives the designation. Aspects can also be used.

Further, in the association degree value calculating device according to the present invention, as one configuration example, the category designation receiving means displays and outputs information requesting the designation of the category to the user,
The designation is accepted by the input from the user. Here, as the information requesting the designation of the category, for example, information that prompts the user to designate the category is used. Moreover, as the display output, for example, a mode of displaying and outputting on a display screen or the like is used.

Further, in the degree-of-association calculation device according to the present invention, the category information storage means stores information on one or a plurality of categories composed of a plurality of unique names of the same kind,
The correction degree designation receiving means receives the designation of the degree of correcting the keyword relatedness value for each category, and the association degree value calculating means uses the proper name keyword specifying means based on the category information stored in the category information storage means. The keyword-related relevance value for each of the identified keyword and document group is corrected using the correction degree received by the correction degree designation receiving means for the category including the keyword.

Therefore, by using the correction degree designated for each category, the degree of relevance of each keyword related to the keyword and the document group identified as corresponding to the proper name included in each category is higher than the calculated value. By correcting to a value indicating, for example, the relevance value of the keyword corresponding to the proper name included in each category can be increased with a correction degree according to the user's request, and the relevance value can be calculated. The accuracy of the relevance value calculated by reflecting the user's request can be increased.

Here, the degree of correction of the keyword-related degree value for each category is, for example, a certain category A.
For example, the correction is performed by multiplying the pre-correction keyword-related relevance value by a, and for the other category B, the correction is performed by multiplying the pre-correction keyword-related relevance value by b. b times) can be used.

Further, as the correction degree designation receiving means,
For example, in the case of accepting the designation of the correction degree from the user, it is possible to use a keyboard, a mouse or the like which is operated by the user and receives an input from the user for designating the correction degree.

Further, in the association degree value calculating device according to the present invention, as one configuration example, the correction degree designation receiving means outputs the information requesting the designation of the correction degree for each category to the user and outputs the designated information. Is accepted by the input from the user. Here, as the information requesting the designation of the correction degree, for example, information that prompts the user to specify the correction degree is used. Moreover, as the display output, for example, a mode of displaying and outputting on a display screen or the like is used.

Further, in the degree-of-association value calculating apparatus according to the present invention, the proper name keyword information display / output means displays and outputs to the user the information regarding the keyword specified by the proper name keyword specifying means. Therefore, since the information about the keyword, in which the relevance value for each keyword is corrected to increase the relevance for each keyword, is displayed and output to the user, the user grasps which keyword has the higher relevance. be able to. Here, as the information on the keyword, for example, the phrase of the keyword, the degree of correction, or the like can be used.

Further, the relevance degree value calculating device according to the present invention as described above can be applied to various devices, for example, a related document retrieving device for retrieving documents related to a keyword group or the like. The present invention can be applied to a document categorizing device that categorizes documents into related keyword groups, a document clustering device that clusters a plurality of documents based on a relevance value between two documents, and the like.

For example, the related document search apparatus according to the present invention extracts a keyword group composed of one or a plurality of keywords from a document group composed of one or a plurality of documents (seed documents) (seed document group). , Multiple documents related to the extracted keyword group (search target documents)
A document group consisting of
Percentage value of the number of documents in which the keyword appears in the seed document group with respect to the total number of documents included in the seed document group for each keyword included in the extracted keyword group (first proportion value)
Is divided by the ratio value (second ratio value) of the number of documents in which the keyword appears in the search target document group to the total number of documents included in the search target document group, and is calculated as the appearance ratio value for each keyword Then, the sum of the appearance ratio values of the keywords that appear in the document among the keywords included in the extracted keyword group for each searched document is calculated as a value indicating the degree of association between the keyword group and the document, When outputting information related to each document retrieved in descending order of relevance value, the relevance value between the keyword group and the document is calculated as follows. That is, the proper name keyword specifying means specifies the keyword corresponding to the proper name based on the set proper name keyword specifying condition, and the relevance value calculating means determines the appearance ratio value of the keyword specified by the proper name keyword specifying means. The relevance value between the keyword group and the document is calculated by multiplying by a predetermined number.

Various methods may be used to extract the keyword group from the seed document group. In addition, various methods may be used as a method of searching a document related to the extracted keyword group from the search target document group. Specifically, how to search for documents related to the keyword using keywords, how to extract keywords from one or more documents and search for related documents using the keywords, and how to use keywords A method of searching a document and using another keyword included in the searched document to search for a document related to the other keyword can be used.

As a specific example, the total number of documents included in the seed document group is N, the total number of documents included in the search target document group is M, and a certain keyword included in the extracted keyword group is When the number of documents in which the keyword appears in the seed document group is n and the number of documents in which the keyword appears in the search target document group is m, the first ratio value is (n / N), the second ratio value becomes (m / M), and the appearance ratio value of the keyword becomes a value obtained by dividing the first ratio value by the second ratio value (n / N). / (M / M).

As a specific example, among the keywords included in the extracted keyword group, there are z (z is an integer of 1 or more) keywords that appear in a certain document retrieved, and these z keywords for the document are included. When the respective appearance ratio values of the above are Pi (i = 1 to z), the degree of association value between the keyword group and the document is represented by ΣPi.
Note that Σ represents the total sum from i = 1 to i = z.

When the appearance ratio value of (n / N) / (m / M) is used, it can be considered that the larger the appearance ratio value is, the higher the degree of association is. Various information may be used as the information regarding each searched document, and for example, information on the title of each document may be used. Various methods may be used to output the information. For example, a method of displaying and outputting the information on a display screen or a method of printing and outputting the information by a printer can be used.

The predetermined multiple times used for multiplying the appearance ratio value of the keyword specified by the unique name keyword specifying means by a predetermined number may be set in advance in a memory or the like, or by the user. May be specified.

Further, in the document categorizing apparatus according to the present invention, a plurality of keyword groups composed of one or a plurality of keywords and a document group composed of one or a plurality of documents are included in the keyword group for each keyword group. The value indicating the relevance of each keyword between each keyword and the document group is summed up for all the keywords, and calculated as the relevance value between the keyword group and the document group, and the calculated relevance value is the highest. When categorizing the document group into a keyword group having a value indicating the degree of association, the degree of association value between the keyword group and the document group is calculated as follows.
That is, the proper name keyword specifying means specifies the keyword corresponding to the proper name based on the set proper name keyword specifying condition, and the relevance value calculating means determines the keyword and the document group specified by the proper name keyword specifying means. The degree of relevance for each keyword is increased to calculate the degree of relevance between the keyword group and the document group.

Here, as an example of general categorization, procedure examples (1) to (4) of processing for allocating a certain document to a certain category will be shown. That is, in the procedure (1), N (N is a plurality) categories C1 to CN are prepared as a preliminary preparation. Further, one document S1 to SN representing each category C1 to CN is prepared for each category C1 to CN. In addition, each of the prepared documents S1 to SN is divided into word units, and a set W (Si) of words forming the document is generated for each document Si. Where i = 1 to N
Indicates.

Next, in step (2), the document D to be sorted into categories (that is, the documents to be categorized) D is divided on a word-by-word basis, and a set W of words constituting the document D is divided.
(D) is generated. Next, in the procedure (3), the word set W (S1) to W (SN) of each of the documents S1 to SN representing the categories C1 to CN is added to the word set W.
The number of words that overlap with (D), that is, each word set W
(S1) to W (SN) and the word set W
Number of words that also appear in (D) K (D, S1) to K (D, S
N) is calculated.

Then, in step (4), each of the documents S1 to S
Number of words K (D, S1) to K (D, found for N
The document D is classified (categorized) into a category corresponding to the largest number of words in (SN). That is, when the maximum number of words is K (D, Si), the document D is categorized into the category Ci.

Further, as specific examples of the categorizing method focusing on the proper name, embodiment examples (1) to (4) in the case where the present invention is applied to the above categorizing will be shown. In the embodiment example (1), in the above procedure (3), the word set W (D) and the documents S1 to S1 representing the categories C1 to CN are used.
The number of words K (D, S1) to K (D, SN) that appear redundantly in the word sets W (S1) to W (SN) forming the SN.
When checking the above, if the duplicated word corresponds to a proper name (for example, proper noun), the number of words K (D, Si) is increased by the number of proper names. As a specific example, if 10 words of the document Si overlap with the document D, and 3 of these 10 words correspond to a proper name, the word number K (D, Si ) = 10 + 3 = 13
And

In the embodiment example (2), the document Si representing a certain category Ci may be composed of a plurality of documents in the procedure (1). In the embodiment example (3),
In the above procedure (1), a set of words representing a certain category Ci may be used instead of the document Si representing the certain category Ci.

In the embodiment (4), it is assumed that, in the procedure (4) after the procedure (3), the duplicated words may be weighted in consideration of their appearance frequency and the like. As a specific example, the document S in which the word wi is included in the document D to be sorted and the word wi is N documents S1 to S1.
When the document is included in y documents in SN, the feature level F (wi) of the word wi is set to N / y, and the word w
When i corresponds to a proper name, for example, the feature degree F (wi)
= 10 × (N / y), the characteristic degree F (w
i) is multiplied by a predetermined number (here, 10 times). Note that y =
In the case of 0, the characteristic degree F (wi) = 0. And
The above-described characteristic degree F (wi) is calculated for all the words wi included in the document D, and the similarity R between the document D and the document Si is calculated.
Calculate (D, Si) = ΣF (wk). Here, Σ represents the total sum from k = 1 to k = K, and w1 to wK represent words that appear in duplicate in the document D and the document Si. In this way, the similarity R (D, Si) for all the documents Si (i = 1 to N) is obtained, and the document D is classified (categorized) into the category of the document having the largest similarity. ) Do. That is, when the value of the similarity R (D, Sk) is the largest, the document D is categorized into the category Ck.

Further, in the document clustering apparatus according to the present invention, 2 included in a document group composed of a plurality of documents.
With regard to one document, the value representing the degree of association of each keyword of these two documents for each keyword included in the keyword group composed of one or more keywords is the sum of all the keywords When the documents included in the document group are clustered based on the degree-of-association value, the degree-of-association value of two documents is calculated as follows. That is, the unique name keyword specifying means specifies the keyword corresponding to the unique name based on the set unique name keyword specifying condition, and the relevance value calculating means determines two documents about the keyword specified by the unique name keyword specifying means. The degree of relevance of each of the keywords is increased to calculate the degree of relevance of these two documents.

As a method of clustering a plurality of documents included in a document group, as an example, the relevance value of two documents in the plurality of documents is calculated for all the combinations of documents. It is possible to use a method in which the two documents for which the value indicating the highest degree of association is calculated are put together into one at the position of the center of gravity thereof and repeatedly executed.

Here, as an example of general clustering, procedure examples (1) to (4) of processing for allocating a certain document to a certain category will be shown. That is, in the procedure (1),
Regarding each document D1 to DN included in the document set (document group) composed of N documents D1 to DN, each document D1
A vector having the words included in DN as elements is generated. At this time, the value of each element is set to 1 when a certain word is included, and is set to 0 when the certain word is not included.

As a specific example, the vector of the document D1 =
{1, 0, 1, 0, 0, 0, 1, ..., 0}, vector of document D2 = {0, 0, 1, 0, 0, 1, 0, ..., 1},
Vector of document D3 = {1,1,0,0,1,1,0,
..., 0} is obtained. Here, each vector is composed of the value (1 or 0) of M (M is 1 or more) elements, and the value of each element is whether or not each of the M words appears in each document. Is represented.

Next, in the procedure (2), the vector of the document Di and the document Dj for i and j that take values from 1 to N are used.
The inner product (Di, Dj) with the vector is calculated, and the pair of i and j that maximizes the value of the calculated inner product (Di, Dj) is obtained. However, the inner product (Di, Dj) is calculated for a pair of i and j having different values. The inner product (D
As i, Dj), for example, as is generally known, the sum of products of the element values of two vectors is used.

Next, in step (3), the above inner product (D
i, Dj) is regarded as the similarity between the two documents Di and Dj, and the similarity between the documents of all the combinations has the largest similarity (that is, the inner product) between the two documents. Seek the center of gravity between. That is, when the similarity (Dp, Dq) between the document Dp and the document Dq is the maximum, the center of gravity between the document Dp and the document Dq is taken. The vector representing the center of gravity is set as the vector of the document Dpq obtained by clustering the document Dp and the document Dq.

Next, in step (4), the two clustered documents (for example, the document Dp and the document Dq) are removed from the entire document (document set), and instead, the clustered result document Dpq is removed. Add to document set. Then, in the first clustering procedure, the total number of documents included in the document set is decreased by 1 (N-1).
By repeating the clustering procedure in the same manner as described above, the total number of documents included in the document set is decreased by one. Then, when the k-th clustering process is completed and the total number of documents (N−k) included in the document set is, for example, the number of clusters designated in advance or less, the process is terminated and the number of clusters is set to be the number of clusters or less. If not, the above steps (2), (3) and (4) are repeated.

Further, as specific examples of the clustering method focusing on the unique name, the embodiment examples (1) to (3) in the case where the present invention is applied to the above clustering will be shown.
In the embodiment example (1), in the above procedure (2), when a pair of i and j whose inner product has the maximum value is obtained for i and j having values from 1 to N, the vector of the document Di and the document Dj (Di, Dj) '= as the inner product of the vectors
The value (Di, Dj) + k is calculated. Where k
Is the number of unique names matching the occurrence of an element (element value of 1) in the values of the elements forming these two vectors, that is, the common name that appears in both document Di and document Dj. Is the number of unique names of.

In the embodiment (2), in the above procedure (1), it is determined whether each word is included in each document.
Alternatively, instead of constructing the vector represented by the value of the element of 0, the vector representing the number of times each word appears in each document, the frequency, etc. is constructed as the value of the element. In the embodiment example (3), when the embodiment example (2) is applied, the value of the element corresponding to the proper name in the vector is calculated by multiplying by a predetermined number. Increase the value that is played. In the above-described exemplary embodiments, it is possible to perform clustering with weighting on proper names.

The present invention also provides a method for calculating the degree-of-association value as shown above. It should be noted that such a method according to the present invention is executed in, for example, a computer including a CPU, a memory and the like. For example, in the degree-of-association value calculation method according to the present invention, the value is included in the keyword group as a value indicating the degree of relevance related to a keyword group composed of one or more keywords and a document group composed of one or more documents. When calculating the sum of the relevance values for each keyword and each keyword related to the document group for all keywords, for example, specify the keyword corresponding to the unique name based on the unique name keyword specifying condition set in the memory, The degree of association for each keyword related to the specified keyword and document group is increased to calculate the degree of association value for the keyword group and document group.

Further, in the related document search method according to the present invention, a keyword group composed of one or a plurality of keywords is extracted from a seed document group composed of one or a plurality of documents and is related to the extracted keyword group. A document is searched from a search target document group composed of a plurality of search target documents, and for each keyword included in the extracted keyword group, the keyword is included in the seed document group with respect to the total number of documents included in the seed document group. A value obtained by dividing the ratio value of the number of appearing documents by the ratio value of the number of documents in which the keyword appears in the search target document group to the total number of documents included in the search target document group is set as the appearance ratio value for each keyword. The value obtained by summing the appearance ratio values of the keywords that appear in the document among the keywords included in the keyword group that is calculated and retrieved for each retrieved document is the key. When a value representing the degree of association between a word group and the document is output and information about each document retrieved in descending order of the degree of association is output by, for example, an information output device, for example, a unique name keyword specifying condition set in a memory The keyword corresponding to the proper name is specified based on the above, and the appearance ratio value of the specified keyword is multiplied by a predetermined number to calculate the relevance value between the keyword group and the document.

Further, in the document categorizing method according to the present invention, a plurality of keyword groups composed of one or a plurality of keywords and a document group composed of one or a plurality of documents are included in each keyword group in the keyword group. The value indicating the relevance of each keyword between each keyword and the document group is summed up for all the keywords, and calculated as the relevance value between the keyword group and the document group, and the calculated relevance value is the highest. When categorizing the document group into a keyword group that is a value indicating the degree of association, for example, a keyword corresponding to the unique name is specified based on the unique name keyword specifying condition set in the memory, and the specified keyword and document group The degree of relevance for each keyword is increased to calculate the degree of relevance between the keyword group and the document group.

Further, in the document clustering method according to the present invention, 2 included in a document group composed of a plurality of documents.
With regard to one document, the value representing the degree of association of each keyword of these two documents for each keyword included in the keyword group composed of one or more keywords is the sum of all the keywords When a document included in the document group is clustered based on the relevance value, the keyword corresponding to the unique name is specified based on the unique name keyword specifying condition set in the memory, for example. The degree of relevance of each of the two documents for the generated keyword is increased, and the relevance value of these two documents is calculated.

The present invention also provides a program or the like for executing the processing for calculating the degree-of-association value as described above. For example, the program according to the present invention uses the keywords included in the keyword group and the keywords included in the keyword group as a value indicating the degree of association regarding the keyword group including one or more keywords and the document group including one or more documents. When causing a computer to execute the process of calculating the sum of the relevance values for each keyword related to a document group for all keywords, for example, the keyword corresponding to the unique name is specified based on the unique name keyword specifying condition set in the memory. The computer is caused to execute the processing for performing the above processing and the processing for increasing the degree of association for each keyword related to the specified keyword and the document group and calculating the degree of association for the keyword group and the document group.

Further, the program according to the present invention is a process of extracting a keyword group composed of one or a plurality of keywords from a seed document group composed of one or a plurality of documents, and a document related to the extracted keyword group. To search the target document group consisting of a plurality of search target documents for each keyword included in the extracted keyword group, and the keyword in the seed document group for the total number of documents included in the seed document group The value obtained by dividing the ratio value of the number of documents in which the word appears by the ratio value of the number of documents in which the keyword appears in the search target document group to the total number of documents included in the search target document group, for each keyword And the appearance ratio value of the keywords appearing in the document among the keywords included in the extracted keyword group for each searched document. Causing the computer to execute a process of calculating the value obtained as a value indicating the degree of association between the keyword group and the document, and a process of outputting information regarding each document retrieved in the descending order of the degree of association by, for example, an information output function At this time, for example, the process of identifying the keyword corresponding to the proper name based on the proper name keyword specifying condition set in the memory, multiplying the appearance ratio value of the specified keyword by a predetermined number, and the degree of association between the keyword group and the document. The computer is caused to execute a process of calculating a value.

Further, the program according to the present invention relates to a plurality of keyword groups composed of one or a plurality of keywords and a document group composed of one or a plurality of documents. The process of calculating the value of the relevance value of each keyword between the keyword group and the document group for all keywords as the relevance value between the keyword group and the document group, and the calculated relevance value is the highest. When causing a computer to perform a process of categorizing a document group into a keyword group having a value indicating a degree of association, for example, a process of specifying a keyword corresponding to a unique name based on a unique name keyword specifying condition set in a memory, Calculating the relevance value between the keyword group and the document group by increasing the degree of relevance for each keyword between the specified keyword and the document group The process and to be executed on the computer.

Further, the program according to the present invention relates to two documents included in a document group composed of a plurality of documents, and these two keywords for each keyword included in the keyword group composed of one or a plurality of keywords. A process of calculating a sum of values representing the degree of association of each keyword of a document for all keywords as a degree-of-association value of these two documents, and clustering the documents included in the relevant document group based on the degree-of-association value When executing the process and the computer, for example, a process of specifying a keyword corresponding to a proper name based on a proper name keyword specifying condition set in a memory, and a degree of relevance for each keyword of two documents regarding the specified keyword And the process of calculating the relevance value of these two documents by the computer. .

[0074]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment according to the present invention will be described with reference to the drawings. First, a related document search device according to the first embodiment of the present invention will be described. FIG. 1 shows a configuration example of a related document search device to which the present invention is applied. The related document search device includes a search request receiving unit 1 that receives a search request.
And a unique name holding unit 2 that holds information for identifying a unique name, information on a category to which the unique name belongs, and the like in a memory,
A degree-of-association calculation unit 3 that calculates a value indicating the degree of association of each word,
Retrieval target document word appearance frequency retaining unit 4 that retains the occurrence frequency of each word in the document that is the retrieval target, retrieval target database (DB) 5 that retains the document that is the retrieval target in memory, and retrieval target database 5 There is provided a search unit 6 for searching the documents stored in the document, a document relevance calculating unit 7 for calculating a value indicating the relevance of the documents, and a search result presenting unit 8 for presenting the search results.

An example of the operation performed by the related document search device of this example will be shown. First, the user, as a search request for a search, retrieves the document content information of the seed document set (seed document group) and the information designating the category of the unique name from the search request receiving unit 1.
, And the search request receiving unit 1 receives these pieces of information. Here, in this example, it is assumed that the seed document set is composed of four documents in which concert information of a certain singer group is described. Further, in this example, it is assumed that the user specifies the category of “personal name” as the category of the unique name.

Next, the degree-of-association calculation unit 3 extracts words (keyword groups) that are keywords from the seed document set and the appearance frequency of each word, and the appearance frequency of each word in the seed document set and the search target document. The relevance value of each word is calculated based on the appearance frequency of each word with respect to the entire search target document held in the word appearance frequency holding unit 4.

Specifically, in this example, from the seed document set,
"ABC record publication", "Budokan", "new song", "singer group DEF", "April 27", "concert"
Suppose that the word is extracted. “Singer group DEF” is the name of a certain singer group, and is “person name”.
It is assumed that they are classified in the category of the proper name.

It is also assumed that each of the above-mentioned words appears in the following number (n) of documents among the N (= 4) seed documents included in the seed document set. "ABC record publication" appears in one seed document (n = 1).
"Budokan" appears in two seed documents (n = 2).
The "new song" appears in three seed documents (n = 3).
“Singer group DEF” appears in four seed documents (n = 4). "April 27th" appears in one seed document (n = 1). "Concert" appears in three seed documents (n = 3).

It is also assumed that the total number M of search target documents is 1000 (M = 1000) and that each of the above-mentioned words appears in the following number (m) of documents in the search target document set. . The search target document word appearance frequency storage unit 4 stores a value of (m / M). "ABC record publication" appears in 10 search target documents (m =
10). "Budokan" appears in 5 documents to be searched (m = 5). The “new song” appears in 100 search target documents (m = 100). The "singer group DEF" appears in 200 search target documents (m = 20).
0). "April 27th" appears in three search target documents (m = 3). The "concert" appears in 300 search target documents (m = 300).

Then, the relevance value of each word is (n / N) /
When calculated as (m / M), the relevance value of each word described above is as follows. The relevance value for each word of "ABC record publication" is 25. The relevance value for each word of "Budokan" is 100. The relevance value for each word of "new song" is 7.5. The relevance value for each word of the "singer group DEF" is 5. The relevance value for each word of "April 27th" is (about) 83. The relevance value for each word of "concert" is 2.5.

Next, the proper name holding unit 2 determines whether or not each of the above-mentioned words belongs to the designated category "personal name" based on the information held in the proper name holding unit 2. . In this example, a case is shown in which a category is specified. However, for example, when no category is specified and the search request receiving unit 1 does not receive a category specification, weighting is applied to all words corresponding to unique names. do. In this example, it is determined that the word “singer group DEF” in the above words corresponds to a proper name belonging to the category “personal name”.

Next, the degree-of-association calculating unit 3 determines the "singer group DE" that is the proper name that matches the category designated by the user.
Adjust the relevance value for the word "F". In this example, the relevance value of the word "singer group DEF" that is determined to be a proper name belonging to the category "personal name" is multiplied by 100. Then, the relevance value of each word described above is adjusted to the following value. The relevance value for each word of "ABC record publication" is 25. The relevance value for each word of "Budokan" is 100. The relevance value for each word of "new song" is 7.5. The relevance value for each word of the "singer group DEF" is 500. The relevance value for each word of "April 27th" is (about) 83. The relevance value for each word of "concert" is 2.5.

Next, the retrieval unit 6 receives the above-mentioned words and the degree-of-relationship values obtained for each word from the degree-of-relationship calculation unit 3 as related-word information, and uses the received words as keywords, for example, in general. An OR search is performed on the search target database 5, whereby a document including at least two keywords is searched from a plurality of documents (search target document group) held in the search target database 5.

In this example, it is assumed that the following four documents 1 to 4 are searched and the following words are included in each document. Document 1 includes the words “Budokan” and “new song”. Document 2 includes the words "singer group DEF" and "concert". Document 3 says “ABC
The words "published record" and "new song" are included. Document 4 includes “ABC record publication”, “new song”, and “April 2
The word "7 days" is included.

Next, the document relevance calculating unit 7 calculates the document relevance values (keywords) of the documents 1 to 4 based on the relevance values of the related words (words) included in the retrieved documents 1 to 4. Calculate the degree of association between the group and each document 1-4. In this example, the sum of the relevance values of the related words included in the retrieved documents 1 to 4 is set as the document relevance value of each of the documents 1 to 4. Then, in this example, the document relevance values of the retrieved documents 1 to 4 are as follows. The document relevance value of document 1 is 100
It becomes + 7.5 = 107.5. The document relevance value of document 2 is 500 + 2.5 = 502.5. The document relevance value of document 3 is 25 + 7.5 = 32.5. The document relevance value of the document 4 is 25 + 7.5 + 83 = 115.5.

Next, the retrieval result presentation unit 8 receives the information about the retrieved documents 1 to 4 and the document relevance value obtained for each of the documents 1 to 4 from the document relevance calculation unit 7, and receives the document 1 Information about 4 to 4 are presented to the user in descending order of document relevance value. In this example, the title information of each of the documents 1 to 4 is arranged in order from the highest document relevance value and displayed and output on the display screen. In this example, the title information of four documents 1 to 4 is arranged in the following order. (Document with highest document relevance value) Document 2 (Document with second highest document relevance value) Document 4 (Document with highest third document relevance value) Document 1 (Fourth document relevance value) Highly expensive document) Document 3

As described above, in the related document retrieval apparatus of this example, when calculating the relevance value for each word, the relevance value of the word corresponding to the unique name belonging to the designated category "personal name" is calculated. By executing the correction to increase the weight, for example, the document 2 in which the word “singer group DEF” that belongs to the category appears can be notified to the user as the document having the highest document relevance value.

In this example, the user inputs the seed document set. However, for example, the user inputs a keyword and the related document search device searches for documents related to the keyword as a seed document set. Various configurations can also be used.

In this example, the case where the category "personal name" is designated is shown. For example, even if the seed documents inputted by the user are the same, "company name" and "place name" are specified.
Or "date" or "time" or "date and time" or "product name"
By designating another category such as, the document relevance value placing importance on the proper name included in the designated category can be calculated. In this example, one category is designated. However, as described above, a mode in which the relevance value for words corresponding to all proper names is increased without designating a category, or a plurality of categories are used. It is also possible to use a mode of designating the category. Further, it is also possible to use a mode in which the degree of correction, such as how much the degree of relevance value for each word before correction is corrected, is specified by the user for each category.

FIG. 2 shows an example of screen information for the user to input a user profile and specify a category. The screen information is, for example, the search request receiving unit 1.
Is displayed and output to the user. In this screen information, which category of the "company", "personal name", "government", "time", and "place" the user is interested in, that is, which category corresponds to a proper name The user is asked if the word relevance value should be corrected, and in the example of the figure, the categories of "company" and "personal name" are specified by the user. In addition, in this screen information, the user is asked a scaling factor such as how much the degree of relevance of a word belonging to a category designated by the user is increased. In the example of the figure, “double” is set for “company”. Therefore, the "personal name" is tripled.

Here, in this example, the search request receiving unit 1 displays the screen information as shown in FIG. 2, for example, to the user and displays the screen information to the user to accept the category specification by the user's input. The category information storage unit is configured by a function of storing the information about the category in which the unique name holding unit 2 is composed of a plurality of unique names of the same kind, and the unique name holding unit 2 is set. The unique name keyword specifying means is configured by the function of specifying the keyword corresponding to the unique name included in the category specified based on the unique name keyword specifying condition. The (relevance value for each keyword) is compared with the calculated value and corrected to a large value (a value indicating that the relevance is high), and the document relevance calculation unit 7 determines the document relevance. Relevance value calculating means is constituted by the function of calculating a.

Further, in this example, the search request accepting section 1 outputs the screen information as shown in FIG. 2 to the user and inputs the correction factor (correction degree) for each category according to the input from the user. A correction degree designation receiving means is configured by the function of receiving the designation, and the search result presentation unit 8
However, the corrected keyword information display output means is constituted by a function of displaying and outputting information such as a word or phrase of the keyword for which the relevance value for each keyword is corrected to the user.

Further, in this example, the keyword relevance value for each keyword between the keywords (words) other than the keyword corresponding to the proper name included in the designated category and each of the documents 1 to 4 is set to non-zero, and the document relevance value is set. Although the case of calculation is shown, for example, by configuring such that the relevance value for each keyword is set to zero to calculate the document relevance value, the amount of calculation and time required for calculating the document relevance value are reduced. You can also

Next, clustering by the document group classification device (document clustering device) according to the second embodiment of the present invention will be described. In this example, a document is a collection of one or more sentences described in, for example, a natural language.
It is such that a collection of one or more sentences is the subject of classification. Specifically, a specific one that can be classified, such as newspaper articles classified into politics, economy, sports, etc.
Documents are those that contain and contain sentences.

FIG. 3 shows a configuration example of the document group classification device of this example. The document group classification device includes a document group input unit 11 for inputting a document group, and a document group input device 11 for inputting each document. Document group storage unit 12 for storing document group data including contents
A document group analysis unit 13 that analyzes the document group data by morphological analysis or the like to extract keywords, and a document group classification unit 14 that classifies the document group based on the analysis result.
And a classification result storage unit 15 for storing the classification result.

Here, the document group storage unit 12 and the classification result storage unit 15 are composed of, for example, a hard disk or a semiconductor memory for storing information. Also, the document group analysis unit 13
The document group classifying unit 14 is, for example, a memory that stores a program or a CP that operates according to the description content of the program.
It has a U (Central Processing Unit). In addition,
The above memory and CPU can be shared by a plurality of processing units.

An example of the operation performed by the document group classification device of this example will be shown. First, the document group input unit 11 inputs information on a document group to be classified by a user operating a keyboard or the like, or from a hard disk in the device or an external network. Then, the document group input unit 11 stores the document group data including the contents of the respective input documents in the document group storage unit 12 in a format in which each document can be identified. Specifically, for example, each document is given a document number or the like and stored in the document group storage unit 12, and each document is managed by the document number.

Next, the document group analysis unit 13 reads the document data from the document group storage unit 12 and performs a natural language analysis on each document data corresponding to each document. The grammatical information indicating the word, its appearance position, its part of speech, the relationship between words, etc. is extracted from the. Then, the document group analysis unit 13 extracts, for example, only words having part-of-speech information corresponding to proper nouns, and for each extracted word, obtains the appearance frequency in the document including the word. In addition, as the value of the appearance frequency,
For example, information of a "1" value or a "0" value indicating whether or not a word appears in a document, or information on the number of times or frequency of occurrence of a word in a document can be used.

In this way, when the appearance frequency information of each word corresponding to a proper noun is obtained for all documents, next,
The document group classification unit 14 selects, for example, one arbitrary document,
Find the distance between the selected document and all other documents. In this example, as shown below, the distance between the two documents is such that the larger the value, the closer the distance is. This corresponds to, for example, the relevance value between the two documents. Then it can be considered.

As a specific example, there are t (t is a number of 1 or more) words for obtaining the distance between two documents, and the value of the appearance frequency of the i-th word in the document Dr is ar (i). , The value of the appearance frequency of the i-th word in the document Ds is a
Assuming s (i), the distance Sim (Dr, Ds) between the two documents of the document Dr and the document Ds is expressed by, for example, Expression 1.

[0101]

[Equation 1]

Here, with reference to FIG.
An example of clustering a document group based on the distance between documents is shown. For convenience of description, it is assumed that the document group includes four documents A, B, C, and D. For example, it is assumed that the result of obtaining the distance between two documents for all combinations of selecting two documents from all documents A to D is represented as a matrix in FIG. Then, referring to this matrix, the distance is short (in this example, the value is large)
Clustering can be performed by sequentially creating a hierarchical tree from two documents.

First, in the example of FIG. 10A, since the distance between the document A and the document D is the shortest, as shown in the figure, the document A
And the document D are associated with each other. Then, by combining the documents A and D, the matrix described above is changed as shown in FIG. 7B, and in the example of FIG. 7B, the distance between the group of documents A and D and the document C is changed. Is closest to each other, the group of documents A and D is associated with the document C as illustrated.
Then, the above matrix is changed as shown in FIG. 6C, and the remaining document B and the group of documents A, C, and D are associated with each other as illustrated. Then, a hierarchical tree as shown in FIG. 7C can be obtained as a clustering result.

As described above, in the document group classification apparatus of this example, the element of the keyword included in the document is extracted by the natural language analysis, and the part of speech or the category information attached to the element is extracted to obtain the element. In the situation where only the elements corresponding to proper nouns are counted based on the accompanying part-of-speech information, the number of appearances of each element is counted based on the counting rule that stipulates how to calculate the appearance frequency, and the count result By clustering the document groups by determining the distance between two documents based on, for example, a method that considers the amount of information of a large number of document groups and the information amount of the word itself based on its part-of-speech It can be classified (clustering).

In this example, the distance between the two documents is calculated based only on the word corresponding to the proper noun, but the importance of the word corresponding to the proper noun is increased by weighting each word, for example. Such a configuration is also possible, and in such a configuration, for example, based on the part-of-speech information attached to the element, an element corresponding to the proper noun is given a greater weight than elements corresponding to other parts of speech, Something like calculating the distance between two documents is performed.

Specifically, as the one corresponding to the above expression 1, for example, the weighting coefficient for the i-th word is Ki.
Then, it is possible to obtain the distance between the two documents by the arithmetic expression shown in Expression 2.

[0107]

[Equation 2]

Note that, for example, the elements to be counted are only those corresponding to proper nouns, or the elements corresponding to parts of speech other than proper nouns are also counted as weights for proper nouns to other parts of speech. It is also possible to have a configuration in which the user can arbitrarily select whether to make it larger than that.

Various methods may be used to extract the keywords (elements), for example, a method of extracting a phrase corresponding to a proper noun registered in the dictionary in advance, for example, "corporation". It is possible to use a method of extracting a word or phrase considered as a proper noun based on an appearance pattern peculiar to the proper noun such as containing a word or phrase such as. Further, such a dictionary can be made editable by the user so that the user can arbitrarily add a new word to the dictionary.

Here, in this example, a group of t elements (keywords) used for calculating the distance between two documents by the above Expression 1 or Expression 2 corresponds to a keyword group. A document group is formed, and a degree-of-association value relating to the keyword group and the document group is calculated as the distance between these two documents. More specifically, two documents Dr, Ds
A document group is composed of the keywords, and the keyword included in the keyword group and the relevance value for each keyword related to the document group are {ar (i) .as (i)} in the above Expression 1 or {Ki. ar (i) · as (i)}. Further, as another way of understanding, for example, the document Ds can be regarded as a document group, and ar (i) can be regarded as weighting information depending on the document Dr added to the i-th keyword. Also in this case, {ar (i) · as (i)} in the above expression 1 or {Ki · ar (i) · as (i)} in the above expression 2 for each keyword is the relevance value for each keyword. Equivalent to.

Also, in this example, the proper name keyword specifying means is constituted by the function of the document group analysis unit 13 to specify the keyword (element) corresponding to the proper noun based on the set proper noun keyword specifying condition. , The document group analysis unit 13 and the document group classification unit 14 set the relevance value for each keyword of two documents only for a keyword identified as a proper noun to a non-zero value, or for such a keyword. The relevance value of each of the two documents is weighted, compared with the calculated value, and corrected to a large value (a value indicating that the relevance is high) to obtain the relevance value of the two documents. A relevance value calculation means is configured by the function of calculating.

Next, an example of categorization performed by the document group classification device (document categorization device) according to the third embodiment of the present invention will be described. Note that the document classification apparatus of this example may have the same configuration as that shown in FIG. 3, for example, and detailed description thereof will be omitted in this example.

An example of categorization performed by the document classification device of this example will be described with reference to FIG. As shown in the figure, it is assumed that, for example, a plurality of keyword groups are prepared in advance for various contents, and that each keyword group is associated with a different classification 1, 2, 3, ... Further, it is assumed that there are documents 1, 2, 3, ... Which are to be categorized.

In this case, for example, taking document 1 as an example, first, by using a method of weighting keywords corresponding to proper names, each keyword group corresponding to document 1 and each classification 1, 2, 3, ... The degree of relevance value with is calculated, the category having the highest calculated degree of relevance is detected, and the document 1 is categorized into the detected category.

Specifically, in the example of the figure, document 1 and classification 1
The degree of relevance with the keyword group of is document 1 and other classification 2,
3 shows the case in which the degree of relevance with the keyword group is the largest, and in this case, the document 1 is categorized into the category 1. In the example of FIG. 5, similarly, the document 2 is categorized into the category 1 and the document 3 is categorized into the category 2.

As described above, even when the document group is categorized into a plurality of categories, the keywords corresponding to the proper names are weighted, as in the first and second embodiments. As a result, it is possible to improve the accuracy of the degree-of-association value and reduce the amount of calculation and time required to calculate the degree-of-association value.

Here, the configuration of the degree-of-association value calculating device and the like according to the present invention is not necessarily limited to the ones described above, and various configurations may be used. Further, the application fields of the present invention are not necessarily limited to those shown above, and the present invention can be applied to various fields.

Further, as various kinds of processing performed in the degree-of-association value calculating apparatus and the like according to the present invention, for example, a control program in which a processor is stored in a ROM (Read Only Memory) in hardware resources including a processor and a memory The configuration controlled by executing the above may be used, or each functional unit for executing the process may be configured as an independent hardware circuit. The present invention also relates to a floppy (registered trademark) disk or a CD (Compact Disk) storing the above control program.
sc) -ROM or the like, which can be grasped as a computer-readable recording medium or the program (itself), and the processing according to the present invention by inputting the control program into the computer from the recording medium and causing the processor to execute the control program. Can be accomplished.

[0119]

As described above, according to the degree-of-association value calculating apparatus and the like according to the present invention, the degree of relevance relating to a keyword group composed of one or a plurality of keywords and a document group composed of one or a plurality of documents. A unique name keyword that is set when calculating the sum of the relevance value for each keyword included in the keyword group and the relevance value for each keyword related to the document group (relevance value for each keyword) It is calculated because the keyword corresponding to the unique name is specified based on the specific condition and the degree of relevance for each keyword related to the specified keyword and the document group is increased to calculate the degree of relevance related to the keyword group and the document group. The accuracy of the relevance value can be increased.

Further, according to the degree-of-association value calculating apparatus and the like according to the present invention, information on one or a plurality of categories composed of a plurality of unique names of the same type is stored, the category designation is accepted, and the accepted category is accepted. Since the keyword corresponding to the unique name included in is specified, the accuracy of the relevance value calculated by reflecting the user's request can be increased.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration example of a related document search device according to a first exemplary embodiment of the present invention.

FIG. 2 is a diagram showing an example of a screen display for inputting a user profile and designating a category and a magnification.

FIG. 3 is a diagram showing a configuration example of a document group classification device according to a second exemplary embodiment of the present invention.

FIG. 4 is a diagram for explaining an example of clustering.

FIG. 5 is a diagram for explaining an example of categorization.

[Explanation of symbols]

1 .. Search request reception part 2 .. Unique name storage part 3.
-Relevance calculator, 4 ... Search target document word appearance frequency holding unit, 5 ... Search target database, 6 ... Search unit,
7 ... Document relevance calculation section, 8 ... Search result presentation section, 1
1 ... Document group input unit, 12 ... Document group storage unit, 13
..Document group analysis unit, 14 ... Document group classification unit, 15 ...
Classification result storage section,

   ─────────────────────────────────────────────────── ─── Continued front page    F-term (reference) 5B009 SA12                 5B075 ND03 NK02 NR12 PR06                 5B082 EA08

Claims (19)

[Claims] 1. As a value representing the degree of association between a keyword group composed of one or a plurality of keywords and a document group composed of one or a plurality of documents, each keyword included in the keyword group and the keyword related to the document group A relevance value calculating device for calculating a value obtained by summing up relevance values of respective keywords for all keywords, and a unique name keyword specifying means for specifying a keyword corresponding to a unique name based on a set unique name keyword specifying condition. And a degree-of-association value calculating means for increasing the degree of relevance for each keyword related to the keyword and document group identified by the proper name keyword specifying means, and calculating a degree-of-relevance value related to the keyword group and document group. Relevance value calculating device. 2. The degree-of-association value calculation device according to claim 1, wherein the degree-of-association value calculation means compares the keyword identified by the proper name keyword identification means and the keyword-related degree of association for each document group with the calculated value. Then, the relevance value calculation device is characterized in that the relevance value for each keyword related to the keyword and the document group is increased by correcting the value to indicate that the relevance is high. 3. The degree-of-association value calculation device according to claim 1, wherein the degree-of-association value calculation means sets the degree-of-relevance value for each keyword related to the keyword and document group specified by the proper name keyword specifying means to non-zero. A degree-of-association value calculating device for increasing the degree of relevance for each keyword related to the keyword and the document group by setting the degree-of-association value for each other keyword to zero. 4. The category information storage device according to claim 1, wherein the degree-of-association value calculation device stores information about one or a plurality of categories that are composed of a plurality of unique names in the same category. Means and category specification receiving means for receiving the specification of the category,
The unique name keyword identifying means identifies the keyword corresponding to the unique name included in the category accepted by the category designation accepting means, based on the category information stored in the category information storage means. Relevance value calculation device. 5. The degree-of-association value calculation device according to claim 4, wherein the category designation receiving means displays and outputs information requesting designation of a category to the user, and receives the designation by input from the user. An association degree value calculating device characterized by: 6. The degree-of-association value calculation device according to claim 2, wherein the category information storage unit stores information about one or a plurality of categories composed of a plurality of unique names of the same kind, and a relation for each keyword for each category. Correction degree designation receiving means for receiving designation of the degree of correcting the degree value, and the related degree value calculating means is specified by the unique name keyword specifying means based on the category information stored in the category information storage means. A degree-of-association value calculation apparatus for correcting a degree-of-keyword-related degree-of-keyword for a keyword and a document group by using a degree of correction received by a degree-of-correction designation receiving unit for a category including the keyword. 7. The degree-of-association value calculation device according to claim 6, wherein the correction degree designation receiving means outputs, to the user, information requesting the designation of the correction degree for each category, and the designation is input from the user. An apparatus for calculating a degree-of-association value, which is received by inputting. 8. The relevance value calculating device according to claim 1, wherein the relevance value calculation device displays and outputs to the user information related to the keyword specified by the unique name keyword specifying means. A degree-of-association value calculation apparatus comprising information display output means. 9. A keyword group composed of one or a plurality of keywords is extracted from a seed document group composed of one or a plurality of documents, and a document related to the extracted keyword group is extracted from a plurality of documents to be searched. A search is performed from the composed search target document group, and the ratio of the number of documents in which the keyword appears in the seed document group to the total number of documents included in the seed document group for each keyword included in the extracted keyword group is searched. A value obtained by dividing the total number of documents included in the target document group by the ratio of the number of documents in which the keyword appears in the search target document group was calculated as the appearance ratio value for each keyword, and extracted for each searched document. Among the keywords included in the keyword group, the sum of the appearance ratio values of the keywords appearing in the document is used as a value indicating the degree of association between the keyword group and the document. A related document search device that outputs information related to each document searched in descending order of the degree of relevance value, and that identifies a keyword corresponding to a unique name based on a set unique name keyword specifying condition. The name keyword specifying means and the relevance value calculating means for calculating the relevance value between the keyword group and the document by multiplying the appearance ratio value of the keyword specified by the proper name keyword specifying means by a predetermined number. Characterized related document retrieval device. 10. Regarding a plurality of keyword groups composed of one or a plurality of keywords and a document group composed of one or a plurality of documents, each keyword included in the keyword group and a keyword of the document group A keyword group in which a value obtained by summing the values indicating the degree of relevance for each keyword is calculated as the degree of relevance between the keyword group and the document group, and the calculated degree of relevance is the value indicating the highest degree of relevance A document categorizing device that categorizes the document group into a unique name keyword specifying unit that specifies a keyword corresponding to a unique name based on the set unique name keyword specifying condition, and a unique name keyword specifying unit Relevance to increase the relevance of each keyword between a keyword and a document group to calculate a relevance value between a keyword and a document group Document categorization apparatus characterized by comprising: a calculating means. 11. With respect to two documents included in a document group composed of a plurality of documents, for each keyword included in a keyword group composed of one or a plurality of keywords, the degree of relevance of each of these two documents for each keyword. Is a document clustering apparatus that calculates a value obtained by summing up the values representing all the keywords as a relevance value of these two documents and clusters the documents included in the document group based on the relevance value. The unique name keyword specifying means for specifying the keyword corresponding to the unique name based on the unique name keyword specifying condition, and the degree of relevance of each of the two documents for the keyword specified by the unique name keyword specifying means are increased. And a relevance value calculation means for calculating a relevance value of two documents. Document clustering device. 12. A keyword related to a keyword group composed of one or a plurality of keywords and a keyword related to a document group composed of one or a plurality of documents as a value representing a degree of association with the document group composed of one or a plurality of documents. It is a relevance value calculation method that calculates the sum of relevance values for each keyword for all keywords, and identifies the keywords corresponding to the proper names based on the set unique name keyword identification conditions And a relevance value for each keyword related to the document group are calculated to calculate a relevance value related to the keyword group and the document group. 13. A keyword group composed of one or a plurality of keywords is extracted from a seed document group composed of one or a plurality of documents, and a document related to the extracted keyword group is extracted from a plurality of documents to be searched. A search is performed from the composed search target document group, and the ratio of the number of documents in which the keyword appears in the seed document group to the total number of documents included in the seed document group for each keyword included in the extracted keyword group is searched. A value obtained by dividing the total number of documents included in the target document group by the ratio of the number of documents in which the keyword appears in the search target document group was calculated as the appearance ratio value for each keyword, and extracted for each searched document. A value that represents the degree of association between the keyword group and the document, which is the sum of the appearance ratio values of the keywords that appear in the document among the keywords included in the keyword group. Is a related document search method that outputs information related to each document searched in descending order of relevance value, and identifies and identifies the keyword corresponding to the unique name based on the set unique name keyword specifying conditions. A related document search method, wherein a relevance value between a keyword group and a document is calculated by multiplying an appearance ratio value of the generated keyword by a predetermined number. 14. Regarding a plurality of keyword groups composed of one or a plurality of keywords and a document group composed of one or a plurality of documents, for each keyword group, each keyword included in the keyword group and the keyword of the document group A value obtained by summing the values indicating the relevance for each keyword for all the keywords is calculated as the relevance value between the keyword group and the document group, and the calculated relevance value is the value indicating the highest relevance. Is a document categorizing method for categorizing the relevant document group, in which the keyword corresponding to the unique name is specified based on the set unique name keyword specifying condition, and the degree of relevance for each keyword between the specified keyword and the document group is determined. A document categorizing method characterized by increasing the degree of association between a keyword group and a document group. 15. With respect to two documents included in a document group composed of a plurality of documents, for each keyword included in a keyword group composed of one or a plurality of keywords, the degree of relevance of each of these two documents for each keyword. Is a document clustering method for calculating a value obtained by summing the values representing the above as a relevance value of these two documents and clustering the documents included in the document group based on the relevance value. The keyword corresponding to the proper name is specified based on the unique name keyword specifying condition, and the degree of relevance of each of the two documents for the specified keyword is increased to calculate the relevance value of these two documents. A document clustering method characterized by. 16. A keyword related to a keyword group composed of one or a plurality of keywords and a document group composed of one or a plurality of documents, as a value indicating the degree of association, and a keyword related to the document group. A program that causes a computer to execute a process of calculating a sum of the relevance values for each keyword for all keywords, and a process of identifying a keyword corresponding to a unique name based on a set unique name keyword specifying condition, A program for causing a computer to execute a process of increasing a degree of association for each keyword related to a specified keyword and a document group and calculating a degree of association value related to the keyword group and a document group. 17. A process of extracting a keyword group composed of one or a plurality of keywords from a seed document group composed of one or a plurality of documents, and a document related to the extracted keyword group being a plurality of search targets. The process of searching from the search target document group composed of documents and the number of documents in which the keyword appears in the seed document group with respect to the total number of documents included in the seed document group for each keyword included in the extracted keyword group A process of calculating a value obtained by dividing the ratio value by the ratio value of the number of documents in which the keyword appears in the search target document group to the total number of documents included in the search target document group as the appearance ratio value for each keyword, and the search Among the keywords included in the extracted keyword group for each document, the sum of the appearance ratio values of the keywords that appear in the document is regarded as the relevant keyword group. A program for causing a computer to execute a process of calculating as a value representing the degree of association with the document and a process of outputting information regarding each document searched in the descending order of the degree of association, wherein a specific name keyword specified is set. Executed on the computer, the process of identifying the keyword corresponding to the proper name based on the condition, and the process of multiplying the appearance ratio value of the identified keyword by a predetermined number to calculate the degree of association between the keyword group and the document. A program characterized by: 18. Regarding a plurality of keyword groups formed of one or a plurality of keywords and a document group formed of one or a plurality of documents, each keyword included in the keyword group and a keyword of the document group A process of calculating the value of the relevance for each keyword for all keywords as a relevance value between the keyword group and the document group, and the calculated relevance value becomes a value indicating the highest relevance. A program for causing a computer to perform a process of categorizing the document group into a keyword group, the process of identifying a keyword corresponding to a unique name based on a set unique name keyword identifying condition, and the identified keyword and document A process of increasing the degree of association of each keyword with a group and calculating a degree-of-association value between the keyword group and the document group. A program characterized by being executed by. 19. With respect to two documents included in a document group composed of a plurality of documents, for each keyword included in a keyword group composed of one or a plurality of keywords, the degree of relevance of each of these two documents for each keyword. Causing a computer to execute a process of calculating a value obtained by summing up the values representing the above as a relevance value of these two documents and a process of clustering the documents included in the document group based on the relevance value. The program is a process for specifying a keyword corresponding to a proper name based on a set proper name keyword specifying condition, and the degree of relevance of each of the two documents for the specified keyword is increased to increase the two documents. A program for causing a computer to execute a process of calculating a relevance value of.