JP3385913B2 - Related word presentation device and medium recording related word presentation program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a related word presenting apparatus and a medium for recording a related word presenting program, and more particularly to a related word presenting apparatus for presenting words related to a search condition and a presentation of words related to the search condition. The present invention relates to a medium in which a related word presentation program to be executed by a computer is recorded.

[0002]

2. Description of the Related Art In a search system, a search method using a keyword is generally used. When an arbitrary keyword (search word) is input to the search system as a search condition, all documents that include the search word in the document content are obtained as search results. However, when a large amount of document information is searched for in such a search system, the search results are inevitably large, and a great deal of labor is spent to judge the suitability. Although there are individual differences in this effort, there is a physical / psychological limit called a disadvantage. When the search result exceeds the limit, it tends to take action to narrow down the search result in order to reduce the effort required to judge the suitability. This indicates that the amount of search results tends to fall below the disadvantage point rather than obtain matching search results. Therefore, you will be excessively using the logical product operator to narrow down the search target,
As a result, a search for useful documents that could otherwise be searched will be omitted. This tendency is especially noticeable in large-scale full-text database searches.

Therefore, as proposed in Japanese Patent Laid-Open No. 2-297290, a method of presenting a related word of a search word to a user by using a related word dictionary is used. The user selects the appropriate words from the related terms of the search terms included in the search expression, combines them with the logical product operator, and narrows down the search results to avoid the blind narrowing down described above. be able to.

For example, when the search word input by the user is "bearing", "ball bearing", "micro ball bearing", "liquid bearing", "gas lubricated bearing", and "magnetic bearing" are related terms of "bearing" from the related term dictionary. Is acquired and presented to the user. Here, if the user wants to know about "magnetic bearing", "magnetic bearing" is used as a new search term (or "magnetic" is combined with "AND" by the logical product operator) to obtain the search result. Can be appropriately narrowed down.

That is, it is possible to accurately narrow down the search result by selecting a word according to the search intention from the words presented as the related words of the search word and combining them with the logical product operator. Become.

[0006]

However, in the ordinary keyword search, only the documents that exactly match the search word are used as the search results, so the search results are often narrowed down more than necessary There is a problem that the user cannot easily know whether the results are narrowed down more than necessary.

In order for the user to confirm that the narrowing down is not excessive (important search results are not leaked), the contents of both the search result before narrowing down and the search result after narrowing down are compared and compared. There must be. Such confirmation work is actually impossible considering the large number of search results. Therefore, in the above-described related art, it is impossible to know what kind of search result is obtained by the narrowing down (how the search result is changed by the narrowing down). Therefore,
Even if the search result obtained by the narrowing down does not meet the search intention (there is a large number of omissions in the search), the search result is adopted as the final result.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide a related word presentation device capable of easily confirming the influence of a search result by changing a search condition.

Another object of the present invention is to provide a medium in which a related word presenting program for operating a computer is recorded so that the influence of a search result by changing a search condition can be easily confirmed. That is.

[0010]

According to the present invention, in order to solve the above-mentioned problems, in a related word presentation device for presenting words related to a search condition, a document storage means for storing a plurality of documents and a plurality of input plural words. Search condition receiving means for receiving the search condition, document search means for acquiring a document set matching the respective search conditions received by the search condition receiving means from the document storage means, and a document set acquired by the document search means. Each word existing in the is a related word candidate, a first value which is the number of documents acquired by the document searching unit, and a document including each related word candidate in the document set acquired by the document searching unit. Second value for each related word candidate and the third value for each related word candidate that is the number of documents containing each related word candidate among the documents stored in the document storage means. And get and first A fourth value, which is the product or sum of the third value and the third value, is calculated for each related word candidate, and based on the ratio between the second value and the fourth value, the search received by the search condition receiving means. Relevance calculating means for calculating the degree of relevance between the condition and each related word candidate for each search condition, and each calculated by the relevance calculating means according to each search condition received by the search condition receiving means. Related word calculation means for comparing related degrees of related word candidates and determining related words based on a change in the value of the related degree of each related word candidate, and related words determined by the related word calculation means are displayed on a display device. The related word display device is provided.

According to this related word presentation device, the plurality of input search conditions are received by the search condition receiving means, and the document search means obtains the document set conforming to each search condition from the document storage means. Next, the degree-of-association calculating unit calculates the degree of association between the search condition received by the search-condition receiving unit and each related-word candidate for each search condition. Then, the related word calculation means compares the related degrees of the related word candidates calculated by the related degree calculation means according to the respective search conditions received by the search condition receiving means, and changes in the value of the related degree of each word. The related word is determined based on this. The determined related word is displayed on the display device by the related word display means. Thus, the difference in the search results obtained from each of the plurality of search conditions can be easily confirmed by the change in the degree of association of the words extracted from the search results.

Further, in a medium in which a related word presentation program for causing a computer to present a word related to a search condition is recorded, a document storage means for storing a plurality of documents, and a plurality of input search conditions are received. Relating the search condition receiving means, the document searching means for acquiring from the document storing means a document set matching each search condition received by the search condition receiving means, and the respective words existing in the document set acquired by the document searching means As a word candidate, a first value that is the number of documents acquired by the document search unit, and a related word candidate that is the number of documents that include each related word candidate in the document set acquired by the document search unit. And a third value for each related word candidate, which is the number of documents containing each related word candidate among the documents stored in the document storage unit, And the value of Is the product or the sum of the third value 4
Is calculated for each related word candidate, and the degree of association between the search condition received by the search condition receiving means and each related word candidate is calculated based on the ratio between the second value and the fourth value. Relevance calculating means for each search condition, and comparing the relevance of each related word candidate calculated by the relevance calculating means according to each search condition received by the search condition receiving means, Related word presenting means for causing a computer to function as related word calculating means for determining a related word based on a change in the value of the degree of association and related word displaying means for displaying the related word determined by the related word calculating means on a display device. There is provided a medium having a program for recording recorded therein.

When the computer executes the related word presentation program recorded on this medium, the document storage means for storing a plurality of documents, the search condition receiving means for receiving a plurality of input search conditions, and the search A document search unit that acquires a document set that matches each search condition received by the condition reception unit from the document storage unit, a first value that is the number of documents acquired by the document search unit, and the document search unit A second value for each word, which is the number of documents including each of the words existing in the acquired document set, and a document including each word among the documents stored in the document storage unit. And a third value for each word that is the number of
The fourth value, which is the product or sum of the value of and the third value, is calculated for each word, and the search received by the search condition receiving means is based on the ratio of the second value and the fourth value. The degree of relevance between the condition and each word, the degree-of-relevance calculation means for calculating each retrieval condition, and the degree of relevance of each word calculated by the degree-of-association calculation means according to each retrieval condition received by the retrieval condition receiving means. A related word calculating means for comparing the related degrees and determining a related word based on a change in the value of the related degree of each word; and a related word displaying means for displaying the related word determined by the related word calculating means on a display device. , Are realized by a computer.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the principle of the present invention. The document storage means 1 stores a plurality of documents. The search condition receiving means 2 receives a plurality of search conditions “Sd” and “Sl” input by the user using an input device such as a keyboard. The document search means 3 uses the document sets “Xd” and “Xl” that match the search conditions received by the search condition receiving means 2.
From the document storage means 1.

The degree-of-association calculation means 4 is first a document search means.
Related word candidates for each word existing in the document set acquired by 3
And The number of documents acquired by the document search means 3
A certain first value and the document set acquired by the document search means 3
Is the number of documents containing each related word candidate existing in
The second value for each collocation candidate and stored in the document storage means
Number of documents that include each related word candidate
Obtain a third value for each related word candidate, and obtain the first value as
The fourth value which is the product or sum of the third value and the related value candidate
Calculated for each and based on the ratio of the second and fourth values
The search condition received by the search condition receiving means 2 and each relation
The degree of association with word candidates is calculated for each search condition. An example
For example, the extended mutual information “MI₀(Sd, Wd
n) ”“ MI ₀(Sl, Wln) "(" Wdn "is" X
"Wln", which is an identifier of a related word candidate included in "d".
Is an identifier of a related word candidate included in “X1”).
And calculate the value as the degree of association. The value of the extended mutual information is
The larger the value, the higher the degree of association between the search condition and the related word candidate.
Indicate that

The related word calculation means 5 is the retrieval condition receiving means 2
Compares the relevance of each related word candidate calculated by the relevance calculating means 4 in accordance with each of the search conditions “Sd” and “Sl” received by, and relates based on the change in the value of the relevance of each related word candidate. Determine the word. For example, the threshold value “Tu” of the word with increased relevance
And the threshold value “Ti” of the degree-of-association word is preset.
Then, for all related word candidates included in “Wdn” and “Wln”, the degree of association with one search condition “Sl” is subtracted from the degree of association with the other search condition “Sd”. To do. If the result of the subtraction is larger than the threshold value "Tu", it is regarded as a word with increased relevance, and if it is smaller than the threshold value "Ti", it is regarded as a word with a low relevance. These related degree increasing words and related degree decreasing words are related terms.

The related word display means 6 displays the related word determined by the related word calculation means 5 on the display device. According to such a related word presentation device, if the user inputs the search conditions before and after narrowing down in the search process, the degree of relevance of each word with respect to those search conditions is calculated, and the change in the degree of relevance is calculated. A suitable related word (for example, a word having a large change in the degree of association) is presented to the user in accordance with the above. As a result, it is possible to know how the characteristics of the search result set have changed by changing the search expression, without having to thoroughly read the document set obtained as the search result.
Specifically, if the degree of relevance of words that match the search intent significantly decreases (or if the degree of relevance of words that do not meet the search intent significantly increases), changing the search formula was not appropriate ( , It is necessary to re-correct the search formula so as to obtain an appropriate search result), and it is possible to avoid erroneous narrowing down that does not meet the user's search intention. Furthermore, by referring to the displayed related words, it is possible to efficiently re-correct the search expression for narrowing down.

The functions of the constituent elements of the above-described principle configurations can be realized by executing a program in which a command of each processing function is described by a computer. In that case, the program is stored in a computer-readable recording medium. As the recording medium, a semiconductor memory device,
A magnetic recording device, an optical disk, or the like can be used.

By the way, in the related word calculating means of the present invention,
Mutual information, Dice-coefficient, and t-sc, which are statistic amounts originally used as similarity between words.
By expanding ore, the similarity between the search expression and the word can be calculated and the similarity can be used as the relevance. As an example of using mutual information, Dice-coefficient and t-score for similarity calculation between words, "Haruno, Yamazaki: Bilingual alignment using dictionaries and statistics, Research Report of IPSJ Natural Language Processing Research Group, 96
-NL-112, pp. 23-30 (1996) ",
"Omori, Tsutsumi, Nakanishi: Preparation of bilingual word dictionary using statistical information, Proceedings of the Second Annual Conference of the Linguistic Processing Society, pp. 4
9-52 (1996) "and the like.

An extension method for applying mutual information and the like to the present invention will be described below. Word word1
And the mutual information (MI) between word2 is

[0021]

[Equation 1]

Is defined as However, the total number of search target documents is M, the number of documents including both word1 and word2 is a,
When the number of documents including only word1 is b and the number of documents including only word2 is c,

[0023]

[Equation 2]

[0024]

[Equation 3]

[0025]

[Equation 4]

[0026] On the other hand, in the present invention, the mutual information amount (MI ₀ ) between the search expression S and the word word is

[0027]

[Equation 5]

It is defined as However, the total number of search target documents includes M and words, and the number of documents obtained from the search formula S includes a ₀ , and the number of documents that do not include word among the documents obtained from the search formula S includes b ₀ and word. If the number of documents excluding the documents obtained from the search expression S among the documents is c ₀ ,

[0029]

[Equation 6]

[0030]

[Equation 7]

[0031]

[Equation 8]

It is Here, “a ₀ + b ₀ ” corresponds to the “first value” in the description of FIG. 1, and “a ₀ ” corresponds to the “second value”.
"Value of" and "a ₀ + c ₀ " corresponds to "third value". Therefore, equation (5) becomes

[0033]

[Equation 9]

As a result, the total number of search target documents M,
The calculation formula has “first value”, “second value”, and “third value” as variables. Dice-coefficie is used as a statistic for obtaining the similarity between words as well as the mutual information.
Mention may be made of nt and t-score. Di
ce-coefficient (DC) and t-sc
ore (TS) is

[0035]

[Equation 10]

[0036]

[Equation 11]

Is defined as Similar to the mutual information, these can be expanded as follows to calculate the similarity between the search formula and the word.

[0038]

[Equation 12]

[0039]

[Equation 13]

MI ₀ (S, word), DC ₀ (S, wor
The larger the value of both d) and TS ₀ (S, word), the higher the similarity between the search expression S and the word word. Thereafter, MI ₀ (S, word) is “extended mutual information”, DC ₀ (S, word) is “extended DC”, and TS ₀
(S, word) will be referred to as "extended TS". In addition,
Similar to the mutual information, the extended DC and the extended TS are

[0041]

[Equation 14]

[0042]

[Equation 15]

It can be expressed as As can be seen from the equation (14), the total number of search target documents M is not necessary when obtaining the extended DC. Next, an embodiment of the related document search device of the present invention will be specifically described.

FIG. 2 is a block diagram showing the configuration of the embodiment of the present invention. The document storage unit 11 is a storage device that stores the content of the digitized search target document in a pair with the document identifier added by the morpheme analysis unit 12.

The morphological analysis means 12 is the document storage means 11
After adding the document identifier to each document stored in
A morphological analysis process is performed on each document to extract an independent word (a word to be a keyword), which is stored as a pair with a corresponding document identifier.

The index structure generating means 13 determines an index structure based on the morphological analysis processing result of the morphological analyzing means 12.
Word-word identifier list 14a, word identifier-document identifier list 14b, document identifier-word identifier list 14c
To create.

The index structure storage means 14 is a word-word identifier list 14 created by the index structure generation means 13.
a, a word identifier-document identifier list 14b, and a document identifier-word identifier list 14c.

The word-word identifier list 14a is a list describing the correspondence between word character strings and word identifiers for specifying the word. The word identifier-document identifier list 14b is a list that describes, for each word identifier, a set of document identifiers of documents including the word character string indicated by the word identifier.

The document identifier-word identifier list 14c is
For each document identifier, it is a list describing a set of word identifiers of words included in the document indicated by the document identifier.

The search condition receiving means 21 is a user interface that receives a search condition (search formula) formed by connecting words by a logical sum operator or a logical product operator a plurality of times from an input device such as a keyboard. Is.

The document search means 22 is the search condition receiving means 2
The document identifiers of all the documents that match the search condition input in 1 are acquired by referring to the word-word identifier list 14a and the word identifier-document identifier list 14b, and the acquired document identifier set is stored. Also, the degree-of-association calculation means 25
With respect to, the number of identifiers in the stored document identifier set is passed, and the total number of documents including the word corresponding to the word identifier given from the degree-of-association calculating unit 25 is passed.

The in-document word searching means 23 obtains the identifier set of words included in each document of the document set satisfying the search condition obtained from the document searching means 22 by referring to the document identifier-word identifier list 14c. , Connect them 1
One word identifier set.

The word appearance number calculation means 24 calculates the number of appearances of each word identifier in the word identifier set obtained from the in-document word search means 23, and creates a pair of the word identifier and the appearance number as a list. .

The degree-of-association calculating means 25 calculates the appearance frequency of each word identifier calculated by the word appearance number calculating means 24,
Based on the three values of the total number of document identifiers that match the search condition obtained from the document search unit 22 and the number of documents that include the word corresponding to the word identifier obtained from the document search unit 22, The expanded mutual information between the input search condition and the word corresponding to each word identifier is calculated.

The degree-of-association storage means 26 is used as the degree-of-association calculation means 2.
Each word and its extended mutual information calculated by 5 are stored together with the corresponding search condition. The search condition designation means 27 displays the search conditions (search conditions input to the search condition receiving means 21) stored in the degree-of-association storage means 26, from which the search condition receiving means 21 is finally input. It is a user interface capable of designating a search condition to be compared with the search condition.

The related word calculation means 28 corresponds to the extended mutual information amount of each word corresponding to the search condition last input to the search condition receiving means 21 and each search condition specified by the search condition receiving means 21. By comparing with the extended mutual information of the word, the absolute value of the difference between the two values is obtained. Then, a word in which the difference between the absolute values of the mutual information exceeds a preset threshold is acquired as a related word.

The related word display means 29 is related word calculation means 2
9 is a user interface for outputting each related word calculated in 8. The search result display means 30 is a document search means 22.
This is a user interface for outputting a document set matching the search condition input to the search condition receiving means 21 obtained from the above with reference to the document storage means 11.

The function of each of the above components is a function realized by the computer executing a predetermined program module. A computer program for realizing these is recorded on a recording medium such as a semiconductor memory or a magnetic recording medium. However, the document storage means 11 and the index structure storage means 14
Is realized by controlling a storage device such as an actual HDD (hard disk device) with a predetermined program.

Each component of the related word presentation device of FIG. 2 has the following relationship with the component of FIG. The document storage means 11 and the index structure storage means 14 correspond to the document storage means 1. The search condition receiving means 21 corresponds to the search condition receiving means 2. The document search means 22 is the document search means 3
Corresponding to. The in-document word searching means 23, the word appearance number calculating means 24, and the relevance calculating means 25 correspond to the relevance calculating means 4. Relevance storage 26 and related word calculator 28
Corresponds to the related word calculation means 5. The related word display means 29 corresponds to the related word display means 6.

In this embodiment, it is necessary to execute the index structure generation process in advance before performing the related document search. Therefore, the index structure generation process will be described first.

As a premise of the index structure generation process, the morphological analysis result list must be generated. Figure 3
FIG. 6 is a diagram showing an example of a morphological analysis result list 12a stored in the morphological analysis means 12. Morphological analysis means 12
Assigns an identifier to each search target document stored in the document storage unit 11, performs a morphological analysis process on each document to extract an independent word, and stores it in a pair with a corresponding document identifier. However, when the same independent word is extracted a plurality of times from the same document, the second and subsequent extraction results are ignored, and independent words corresponding to one document identifier do not overlap.

Based on this morphological analysis result list 12a,
The index structure generation means 13 generates various index structures. Figure 4
6 shows an example of the index structure created by the index structure generating means 13 and stored in the index structure storing means 14. The data in FIGS. 4 to 6 are examples created based on the data in FIG.

FIG. 4 is a diagram showing an example of the word-word identifier list. The word-word identifier list 14a stores the extracted words and the identifiers assigned to the words as a set.

FIG. 5 is a diagram showing an example of a word identifier-document identifier list. Word identifier-document identifier list 14
In b, a word identifier and a document identifier (document identifier) including a word to which the word identifier is assigned are stored as a set.

FIG. 6 is a diagram showing an example of the document identifier-word identifier list. Document identifier-word identifier list 14
In c, a document identifier and a word identifier of a word included in the document to which the document identifier is assigned are stored as a set.

The index structure generation algorithm by the index structure generation means 13 is as follows. FIG. 7 is a flowchart showing an index structure generation algorithm. [S1] Generation processing of word-word identifier list 14a All words in the morpheme analysis result list stored in the morpheme analysis means 12 are sorted without duplication and in the order of the value of the word character string to create a list. . For each word, a natural number starting with 1 from the beginning of the list is assigned as a word identifier. [S2] Generation process of document identifier-word identifier list 14c Each word in the morpheme analysis result list stored in the morpheme analysis means 12 is replaced with the word identifier assigned in step S1, and the word identifier corresponding to each document identifier is replaced. Are sorted in ascending order. [S3] Generation processing of word identifier-document identifier list 14b The word identifiers are arranged in order from 1, and the document identifiers of the documents including the words corresponding to the respective word identifiers are set to the document identifier-word identifier list 14c created in step S2. It is referred to and extracted, and is stored as a pair with a word identifier.

The index structure is generated by the above algorithm. After the index structure generation process is performed, a search expression can be input to the search condition receiving unit 21. Then, when the user inputs a desired search expression using an input device such as a keyboard and issues a search start command, the related word presentation process is started.

FIG. 8 is a flow chart showing an algorithm for obtaining the degree of association from the search expression input to the search condition receiving means 21. Hereinafter, each step of FIG. 8 will be described. In the following description, the word-word identifier list 14a is L1, the word identifier-document identifier list 14b is L2, and the document identifier-word identifier list 14c.
Is described as L3. [S11] The search condition receiving means 21 receives a search expression in which words are combined by a logical product operator or a logical sum operator.
This search formula will be called S. [S12] The document search unit 22 acquires the document identifier of the document that matches S by referring to L1 and L2. The obtained document identifier set is called X, and the number of elements of the set X is N. [S13] If N = 0 in step S12, the process proceeds to step S14, and if not, the process proceeds to step S15. [S14] The document search unit 22 determines that there is no related document of S and ends the process. [S15] The in-document word searching means 23 acquires all word identifiers corresponding to each document identifier belonging to X by referring to L3. Let Y be the set of acquired word identifiers. [S16] The word appearance number calculation means 24 removes the duplication of the word identifiers belonging to Y and records the number of duplications of each word identifier. Newly remove the duplicated word identifier set Y
And the number of overlaps of the element Wn (n = 1, 2, ..., P) of Y is R (Wn). However, P is the number of elements of Y. [S17] With respect to all word identifiers Wn (n = 1, 2, ..., P) belonging to Y, the document search means 22 acquires the total number of document identifiers corresponding to Wn from L2. Let F (Wn) be the number of document identifiers corresponding to the element Wn of Y. [S18] The degree-of-association calculation means 25 sets the total number of search target documents to M for all word identifiers Wn (n = 1, 2, ..., P) belonging to Y,

[0069]

[Equation 16]

[0070]

## EQU16 ## Prob (Wn) = F (Wn) / M ... (17) is calculated, and these values are paired with Wn to form a list.
Also,

[0071]

## EQU18 ## Prob (S) = N / M ... (18) is calculated. [S19] The degree-of-association calculation means 25 calculates the extended mutual information MI ₀ (S, Wn) for each word identifier Wn (n = 1, 2, ..., P) belonging to Y according to the equation (5). The calculated value is used as the degree of association and the degree-of-association storage unit 26
Store in and exit. However, a word having a negative value for MI ₀ (S, Wn) is not stored in the degree-of-association storage unit 26.

By the steps described above, it becomes possible to obtain the degree of association of each word included in the document obtained by the search expression, corresponding to the input search expression. That is, in the degree-of-association storage unit 26, "search formula" for each search formula "list of words appearing in document set obtained by search formula"
"List of relevance (extended mutual information) corresponding to each word"
Will be stored.

With the information corresponding to a plurality of search expressions stored in the degree-of-association storage means 26, the user uses the search condition specifying means 27 to specify the search expressions to be compared and compared with the search results. Then, the related word calculation means 28 calculates a related word. The related word calculation algorithm executed by the related word calculation means 28 is as follows.

FIG. 9 is a flowchart showing the related word calculation algorithm. All the processing shown in this flowchart is executed by the related word calculation means 28. Note that the related term here is characteristically the difference between the search results obtained from both the search expression specified by the search condition specifying means 27 and the search expression last input to the search condition receiving means 21. It is a word to show. [S21] Each word (Wd1, Wd2, ..., W) corresponding to the search expression (Sd) specified by the search condition specifying means 27.
dn) and its degree of association (MI ₀ (Sd, Wd1), MI ₀
(Sd, Wd2), ..., MI ₀ (Sd, Wdn)) and each word (W1, W12, ..., Wl) corresponding to the search expression (Sl) last input to the search condition receiving means 21.
m) and its degree of relevance (MI ₀ (Sl, Wl1), MI
₀ (S1, W12), ..., MI ₀ (S1, Wlm)) is acquired from the degree-of-association storage unit 26. [S22] Each word (Wdi (1 ≦ i
≦ n))

[0075]

[Formula 19] If the obtained value is larger than a preset threshold Tu, Wdi is added to the relevance increasing word list together with the relevance difference, and if the obtained value is smaller than the threshold Tl (<Tu), Wdi is relevance. Add to the reduced relevance word list along with the difference.
Where Wdi is (Wl1, Wl2, ..., Wl
m 0) does not exist, MI ₀ (Sl, Wdi) = 0
And [S23] Each word (Wlj (1≤j
≦ m)) Relevance difference

[0076]

[Equation 20] If the obtained value is larger than a preset threshold value Tu, Wlj is added to the word list with increased relevance and the obtained value is smaller than the threshold value Tl. Add to reduced word list. Where W
When lj does not exist in (Wd1, Wd2, ..., Wdn), MI ₀ (Sd, Wlj) = 0 is set. However, if Wlj already exists in the list, it is not added to the list again. [S24] Each word in the word list with increased relevance is sorted in descending order of difference in relevance, and each word in word list with reduced relevance is sorted in ascending order of difference in relevance.

The related word calculation is performed as described above,
A relevance-increasing word list sorted in descending order of relevance difference and a relevance-decreasing word list sorted in descending order of relevance difference are generated. And each generated list is
It is displayed on the screen of the display device by the related word display means 29.

FIG. 10, FIG. 11, FIG. 12 and FIG. 13 show the user interface of this embodiment. FIG. 10 is a diagram showing a related word search screen. This related term search screen 40
Is divided into four sub-windows 41 to 44.

The sub window 41 is a window for inputting a search expression, and has a text input field 41a and a search button 41b. The user can input a search command by inputting a search expression in the text input field 41a using an input device such as a keyboard and pressing the search button 41b.

The sub window 42 is a window for displaying a retrieval formula history. In this sub-window 42, the search formulas input in the past in the text input field 41a are displayed in the input order. When the user specifies one of the search expressions displayed in the sub window 42, the search expression to be compared with the last input search expression is selected.

The sub-window 43 is a window for displaying related words. The sub-window 43 has a degree-of-association word display field 43a and a degree-of-association word display field 43b. Relevance reduction word display field 4
On 3a, the relevance-reduced word list obtained in step S24 is displayed. Relevance increasing word display field 4
In 3b, the word list with increased relevance obtained in step S24 is displayed.

The sub window 44 is a window for displaying a search result. In this sub-window 44, the document information matching the search formula input in the text input field 41a is displayed.

The text input field 41a in FIG. 10 is a user interface provided by the search condition receiving means 21, the subwindow 42 is a user interface provided by the search condition specifying means 27, and the subwindow 43 is a related word. The sub-window 44 is a user interface provided by the display unit 29, and the sub-window 44 is a user interface provided by the search result display unit 30.

Here, for example, consider a case where the user's search intention is “I want to know about an earthquake-resistant building effective during an earthquake”. In this case, the user may
Specify the search formula "earthquake or earthquake or earthquake" from among.
Then, "seismic resistance" is combined with this search formula by a logical product operator to form a new search formula.

FIG. 11 shows "(earthquake or earthquake or vibration) a.
It is a figure which shows a display screen at the time of inputting "nd seismic resistance" as a search type. Here, the selected search formula “earthquake or earthquake or vibration” in the sub-window 42 is highlighted. Here, by pressing the search button 41b, the processes of steps S11 to S19 (shown in FIG. 8) and steps S21 to S24 (shown in FIG. 9) are executed, and the obtained related words are displayed in the sub window 43. It

FIG. 12 shows "(earthquake or earthquake or vibration) a.
It is a figure which shows the display result of the related term when "nd seismic resistance" is input as a search formula. Sub-window 43 as shown
The content of the related degree decreasing word list and the content of the related degree increasing word list are displayed in each of the related degree decreasing word display field 43a and the related degree increasing word display field 43b.

By referring to the related words displayed in this manner, the user changes the search expression "earthquake or earthquake or vibration" to "(earthquake or earthquake or vibration) and seismic resistance" There are two problems that can be discovered. (1) Information about words similar to “seismic resistance” such as “seismic isolation” and “seismic control” is leaked. (2) A large number of search results are included for earthquake-proof methods for non-buildings such as "furniture,""bookcases," and "chaise chests."

In order to remedy the above-mentioned problems, the user may say, for example, "(earthquake or earthquake or vibration) and (earthquake resistance o
Change to a new search formula such as r seismic isolation or seismic control) and (not (furniture or bookshelf or chest of drawers)).

FIG. 13 shows "(earthquake or earthquake or vibration) a.
nd (seismic resistance or seismic isolation or damping) and (not (furniture o
It is a figure which shows the display result of the related term when "r bookshelf or chest") "is input as a search formula. In the sub-window 42, the search formula "earthquake or earthquake or quake" remains selected. By entering such a search formula, compared to the case where "(earthquake or earthquake or earthquake) and seismic resistance" is entered as the search formula, "seismic isolation" and "seismic control" disappear from the degree of relevance reduction, “Furniture”, “bookshelf”, and “chaise chest” are missing from the list of words with increased relevance.

As described above, by knowing the contents of the word with the reduced degree of association and the word with the increased degree of relevance, it is possible to quickly create a search expression in accordance with the search intention. In the above embodiment, the degree of relevance of each word is obtained for each of the two search conditions, and then the related term indicating the feature of the change in the search result set is determined based on the difference in the degree of relevance. However, it is also possible to determine the related word by obtaining the difference between the document sets obtained from the two search conditions and obtaining the degree of association of the words included in only one document set. Even if the degree of association is calculated after obtaining the difference between the document sets, the same effect as in the above-described embodiment can be obtained.

Such an embodiment will be described below.
Since it can be realized by a system having a configuration similar to that of the embodiment shown in FIG. 2 of this embodiment, description will be given using the reference numerals of the respective constituent elements of FIG. However, the search condition specifying means 2
The search condition designated by 7 is passed to the document search means 22 instead of the relevance degree storage means 26.

FIG. 14 is a flowchart showing a procedure for determining a related word based on the difference between document sets. [S31] The search condition receiving means 21 receives a search expression in which words are combined by a logical product operator or a logical sum operator. [S32] The document search means 22 is replaced by the search condition designation means 27.
The document identifiers of the documents that match the search condition specified by are acquired by referring to L1 (word-word identifier list) and L2 (word identifier-document identifier list). Let A be the obtained document identifier set. The document search means 22
Further, the search condition receiving means 21 acquires the document identifier of the document that matches the search condition received by referring to L1 and L2. Let B be the obtained document identifier set. [S33] The document identifier set corresponding to the document set “Aand (notB)” is set as X in step S12 of FIG. 8, and the document searching unit 22, the in-document word searching unit 23, the word appearance number calculating unit 24, and the degree of association calculating unit 25 performs the same processing as steps S12 to S19 (shown in FIG. 8). As a result, the document set “Aand (not
The degree of association of each word included in “B)” is calculated and stored in the degree-of-association storage unit 26. [S34] The related word calculation unit 28 adds words having a degree of relevance obtained in step S33 that is greater than a preset threshold value T to the reduced degree-of-relevance word list and sorts them in descending order of degree of relevance. [S35] The document identifier set corresponding to the document set "Band (notA)" is set as X in step S12 of FIG. 25 performs the same processing as steps S12 to S19 (shown in FIG. 8). As a result, the document set “Band (not
The degree of association of each word included in “A)” is calculated and stored in the degree-of-association storage unit 26. [S36] The related word calculation unit 28 adds words having a degree of relevance obtained in step S35 that is greater than a preset threshold value T to the word list with increased degree of relevance and sorts them in descending order of relevance. [S37] The related word display means 29 causes the related word calculation means 28.
The contents of the related degree decreasing word list and the related degree increasing word list sorted by are displayed on the screen of the display device.

In this way, the word with increased relevance and the word with decreased relevance can be presented to the user. In this embodiment, the related words are presented from the words that have been omitted from the search results due to the change of the search expression and the newly appeared words.

[0094]

As described above, in the related word presentation device of the present invention, the degree of relevance of each related word candidate with respect to a plurality of search conditions is calculated, the related word is determined based on the change in the degree of relevance, and the Since the related words are presented to the user, the document set obtained as the search result is not carefully read,
By changing the search condition, it becomes possible to know how the characteristics of the search result set have changed.

Further, according to the medium in which the related word presentation program of the present invention is recorded, the computer executes the recorded related word presentation program so that the computer can obtain each related word candidate for a plurality of search conditions. It is possible to calculate the degree of relevance of, determine the related word based on the change in the degree of relevance, and present the related word to the user. Therefore, by using the related word presentation program recorded on this medium, it is possible to present related words so that it is possible to easily know how the characteristics of the search result set are changed by changing the search condition. , It is possible to make a computer do it.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a block diagram showing a configuration of an embodiment of the present invention.

FIG. 3 is a diagram showing an example of a morpheme analysis result list stored in a morpheme analysis unit.

FIG. 4 is a diagram showing an example of a word-word identifier list.

FIG. 5 is a diagram showing an example of a word identifier-document identifier list.

FIG. 6 is a diagram showing an example of a document identifier-word identifier list.

FIG. 7 is a flowchart showing an index structure generation algorithm.

FIG. 8 is a flowchart showing an algorithm for obtaining a degree of association from a search expression input to a search expression receiving means.

FIG. 9 is a flowchart showing a related word calculation algorithm.

FIG. 10 is a diagram showing a related word search screen.

FIG. 11 is a diagram showing a display screen when “(earthquake or earthquake or vibration) and seismic resistance” is entered as a search formula.

FIG. 12 is a diagram showing a display result of related words when “(earthquake or earthquake or earthquake) and seismic resistance” is entered as a search expression.

[Fig. 13] "(Earthquake or earthquake or earthquake) and (Earthquake resistance o
It is a figure which shows the display result of the related term when "r seismic isolation or seismic control) and (not (furniture or bookshelf or chest of drawers)" is input as a search expression.

FIG. 14 is a flowchart showing a procedure for determining a related word based on a difference between document sets.

[Explanation of symbols]

1 Document storage means 2 Search condition receiving means 3 Document search means 4 Relevance calculation means 5 Related word calculation means 6 Related word display means

Continuation of the front page (56) References JP-A-5-81327 (JP, A) Haruhiko Masahiko, 1 other person, bilingual alignment using dictionaries and statistics, IPSJ Natural Language Processing Research Group research report, 1996 3 March 14, 96-NL-112, p. 23-30 Mihoko Kitamura, et al., Automatic extraction of bilingual expressions using bilingual corpus, Journal of Information Processing Society of Japan, April 15, 1997, Vol. 38, No. 4, p. 727-735 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/30

Claims (12)

(57) [Claims] 1. A related word presentation device for presenting words related to a search condition, a document storage unit for storing a plurality of documents, a search condition receiving unit for receiving a plurality of input search conditions, and a search condition receiving unit. A document search unit that acquires a document set that matches each search condition received by the unit from the document storage unit, and each word existing in the document set acquired by the document search unit as a related word candidate, and the document search unit A first value, which is the number of documents acquired by, and a second value for each related word candidate, which is the number of documents that include each related word candidate in the document set acquired by the document search unit, A third value for each related word candidate, which is the number of documents including each related word candidate among the documents stored in the document storage means, is acquired, and the first value and the third value are acquired. Relates a fourth value that is the product or sum of It is calculated for each candidate, and the degree of association between the search condition received by the search condition receiving means and each related word candidate is calculated for each search condition based on the ratio between the second value and the fourth value. The degree of relevance of each related word candidate is compared with the degree of relevance of each related word candidate calculated by the degree of relevance calculation means according to each search condition received by the search condition receiving means. A related word presenting device, comprising: a related word calculating means for determining a related word based on the change of the related word; and a related word displaying means for displaying the related word decided by the related word calculating means on a display device. 2. The degree-of-association calculating means sets the number of all documents stored in the document storage means as M, the first value as α, the second value for each related word candidate as β, and the related value. When the third value for each word candidate is γ, the value of the extended mutual information obtained by the following calculation formula, extended mutual information = log ₂ {(Mβ) / (αγ)} The related word presentation device according to claim 1, wherein the related condition presenting apparatus sets the degree of relevance between the search condition received by the receiving means and each related word candidate. 3. The degree-of-association calculating means sets the number of all documents stored in the document storage means to M, the first value is α, the second value for each related word candidate is β, and the related value is related. When the third value for each word candidate is γ, the search condition receiving means obtains the value of the extended TS obtained by the extended TS (t-score) = M {(Mβ-αγ) / (αγ)}. The related word presenting apparatus according to claim 1, wherein the related word presenting apparatus sets the degree of relevance between the received search condition and each related word candidate. 4. The degree-of-association calculation means sets the first value to α,
The second value for each related word candidate is β, and the third value for each related word candidate
When the value of γ is γ, the value of the extended DC obtained by the following formula, extended DC (Dice-coefficent) = 2β / (α + γ), is used as the search condition and each related word received by the search condition receiving means. The related word presenting apparatus according to claim 1, wherein the degree of association with a candidate is used. 5. The related word calculating means, when determining a related word based on two search conditions before and after narrowing down a document search, rather than the degree of relevance between the search condition before narrowing down and A word with a higher degree of relevance to the search condition after narrowing down is defined as a word with an increased degree of relevance, and is more related to the search condition after narrowing down than the degree of relevance to the search condition before narrowing down. 2. The related word presenting apparatus according to claim 1, wherein the related word candidate having a smaller degree is the related degree decreasing word, and the related degree increasing word and the related degree decreasing word are the related words. 6. A related word presentation device for presenting words related to a search condition, a document storage means for storing a plurality of documents, and a search condition reception for receiving a pair of search conditions before and after narrowing down a document search. Means, a document search means for acquiring from the document storage means a document set that satisfies one of the search conditions received by the search condition receiving means and does not satisfy the other search condition, and the document search means acquires the document set. Each word existing in the document set described above is set as a related word candidate, and a first value that is the number of documents acquired by the document searching unit and each related word existing in the document set acquired by the document searching unit. A second value for each related word candidate that is the number of documents including the candidate, and a related word that is the number of documents that include each related word candidate among the documents stored in the document storage unit. For each candidate The third value is acquired, and the fourth value, which is the product or sum of the first value and the third value, is calculated for each related word candidate, and the ratio between the second value and the fourth value is calculated. Based on the relevance calculation means for calculating the relevance between the pair of search conditions received by the search condition receiving means and each related word candidate, and the relevance obtained from the relevance calculation means is constant. Related word presenting means comprising: a related word calculating means having a related word candidate having a value or more as a related word; and a related word displaying means for displaying the related word determined by the related word calculating means on a display device. apparatus. 7. The related word calculation means, wherein the degree of relevance obtained based on a document set that satisfies the search condition before narrowing down and does not satisfy the search condition after narrowing down is a predetermined value or more. Relevance is defined as a related word candidate in which the degree of relevance is reduced and the degree of relevance obtained based on a document set that does not satisfy the search condition before narrowing down and satisfies the search condition after narrowing down is a predetermined value or more. The related word presentation device according to claim 6, wherein the related word is an increased word, and the related degree increasing word and the related degree decreasing word are the related words. 8. The degree-of-association calculating means sets the number of all documents stored in the document storage means to M, the first value is α, the second value for each related word candidate is β, and the related value is related. When the third value for each word candidate is γ, the value of the extended mutual information obtained by the following calculation formula, extended mutual information = log ₂ {(Mβ) / (αγ)} 7. The related word presentation device according to claim 6, wherein the degree of relevance between the search condition received by the receiving means and each related word candidate is used. 9. The degree-of-association calculation means sets the number of all documents stored in the document storage means as M, a first value as α, a second value for each related word candidate as β, and a related value. When the third value for each word candidate is γ, the search condition receiving means obtains the value of the extended TS obtained by the extended TS (t-score) = M {(Mβ-αγ) / (αγ)}. 7. The related word presenting apparatus according to claim 6, wherein the degree of relevance between the received search condition and each related word candidate is set. 10. The relevance calculating means calculates as follows when the first value is α, the second value for each related word candidate is β, and the third value for each related word candidate is γ. The value of the expanded DC obtained by the formula, expanded DC (Dice-coefficent) = 2β / (α + γ), is used as the degree of association between the search condition received by the search condition receiving means and each related word candidate. The related word presentation device according to claim 6. 11. A medium for storing a related word presentation program for causing a computer to present a word related to a search condition, a document storage means for storing a plurality of documents, and a plurality of input search conditions. Retrieval condition receiving means, document retrieval means for obtaining from the document storage means a document set matching each retrieval condition received by the retrieval condition receiving means, and relating each word existing in the document set obtained by the document searching means As a word candidate, a first value that is the number of documents acquired by the document search unit, and a related word candidate that is the number of documents that include each related word candidate in the document set acquired by the document search unit. And a third value for each related word candidate, which is the number of documents containing each related word candidate among the documents stored in the document storage unit, And the value of A fourth value, which is the product or sum of 3 and the value of 3, is calculated for each related word candidate, and based on the ratio between the second value and the fourth value, the search condition received by the search condition receiving means Relevance calculating means for calculating the degree of association with each related word candidate for each search condition, and each related word candidate calculated by the related degree calculating means according to each search condition received by the search condition receiving means Related word calculating means for comparing related degrees of the related word candidates and determining related words based on a change in the value of the related degree of each related word candidate, and related word display for displaying the related words determined by the related word calculating means on a display device. A medium in which a related word presentation program for causing a computer to function as means is recorded. 12. A medium storing a related word presenting program for causing a computer to present a word related to a search condition, a document storage means for storing a plurality of documents, before and after narrowing down a document search. Search condition receiving means for receiving a pair of search conditions, and a document search for acquiring from the document storing means a document set satisfying one of the search conditions received by the search condition receiving means and not satisfying the other search condition. A first value, which is the number of documents acquired by the document search unit, and a set of documents acquired by the document search unit, wherein each word existing in the document set acquired by the document search unit is a related word candidate. Second value for each related word candidate, which is the number of documents containing each related word candidate existing in the document, and each related word in the documents stored in the document storage means. And a third value of each related term candidate is the number of documents that contain the complement acquired first value and the third
A fourth value, which is the product or sum of the search conditions, is calculated for each related word candidate, and the pair of search conditions received by the search condition receiving means is calculated based on the ratio between the second value and the fourth value. And a related degree calculating unit that calculates a degree of association between each related word candidate, and a related word calculating unit that uses a related word candidate in which the degree of relevance obtained from the related degree calculating unit is a certain value or more, A related word display program for causing a computer to function as a related word display means for displaying the related word determined by the related word calculation means on a display device.