JP2003281181A - Document retrieval device, document retrieval method, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document retrieval device, enabling improvement in retrieval accuracy by controlling the weight of a term (a retrieval character string or a retrieval word) not to be influenced by the degree of document frequency, a document retrieval method, a program and a recording medium. <P>SOLUTION: The document retrieval device is provided with a document ranking part 15, in which a document conforming to a retrieval condition input by a retrieval condition input part 12 is retrieved from a set of documents as retrieval objects stored in a document database 14, the weight of every term of the retrieval condition is calculated using an adjustment parameter for adjusting variation of weight depending on the number of documents to be retrieved, the number of documents where the concerned term appears, and a difference in the number of the documents, the document score of the document conforming to the retrieval condition is calculated according to the weight to rank the document compatible with the retrieval condition. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document search device, a document search method, a program and a recording medium for searching a document that matches given search conditions.

[0002]

2. Description of the Related Art Conventionally, this type of document retrieval apparatus assigns a weight according to the degree of importance to each word in a keyword based on the word statistical information in a document database, and the retrieval condition for each document to be retrieved is based on this weight. Equipped with a document ranking unit that calculates the document score (degree of matching) for each, and calculates the document score by adjusting the weight of each search character string or each search word (hereinafter also referred to as term) in the search condition input by the user (Japanese Patent Laid-Open No. 2000-134588).

Now, a document search method by the above-described document search device will be described. In this document search method, the term weight is calculated by the following equation (7).

[Equation 10] This weight calculation formula (7) is based on a probabilistic model in information retrieval, and is an improvement of the invention of Robertson et al. Regarding the previous invention, see "SE Robertson and S. Walker,
"On relevance weights with little relevance infor
mation ", Proc. of 20th ACM SIGIR Conference, pp. 1
6--24, 1997 ”. In addition, as for those by Ogawa et al. Of the present inventor, "Y. Ogawa, H. Mano,
M. Narita and S. Honma, "Structuring and expanding
queries in the probabilistic model ", Proc. of 8th
TREC, pp. --548, 2000 ”.

When the weight of each term is determined by the above equation (7), the document score of each document is calculated based on how many terms each document contains, and the following equation (8)
Ask in.

[Equation 11]

Further, after a search using the search condition input by the user is performed, words related to the search condition input by the user are selected from terms appearing in the matching document, and the original search condition is selected. Is added as a new search condition, and the search is performed again with this new search condition. When the related words are added in this way, the weighting at the time of re-search is obtained by the following calculation formula (9) using, for example, the frequency of appearance in the conforming document and the non-conforming document.

[Equation 12] This calculation formula (9) is also derived by Ogawa et al., Who is the inventor of the present application, similarly to the above calculation formula (7).

Further, in order to select a word related to the search condition input by the user, a relevance evaluation value (TS
V) is calculated by the following calculation formula (10) using the frequency information of terms in the conforming document and the nonconforming document.

[Equation 13]

[0007]

However, in such a conventional document retrieval apparatus, since the weight is calculated by using the above calculation formula (7), the weight is reduced as the document frequency (n) becomes smaller. Becomes extremely large, a term with a small document frequency (n) gets an unreasonably high weight, and there is a problem that the accuracy in the ranking search deteriorates. Further, in the conventional document search apparatus, since the above-mentioned calculation formula (9) is used for the weight calculation at the time of re-search, the weight becomes extremely large as the document frequency (n) becomes small as in the above. Therefore, there is a problem that a term with a small document frequency (n) gets an unreasonably high weight.

The present invention has been made in order to solve such a problem, and it is possible to improve the search accuracy by controlling the term weight so as not to be influenced by the size of the document frequency. An apparatus, a document search method, a program, and a recording medium are provided.

[0009]

According to a first aspect of the present invention, there is provided a document search device, a document storage unit for storing a set of documents to be searched, and a search condition input unit for inputting a given search condition. Documents that match the search condition input by the search condition inputting unit are searched from the set of search target documents stored in the document storing unit, and the number of search target documents and the corresponding term appear for each term in the search condition. The number of documents to be used and the adjustment parameter for adjusting the variation of the weight due to the difference in the number of documents are used to calculate the weight, and the weight is used to calculate the document score of the document that matches the search condition, and to match the search condition. A document ranking means for ranking documents is provided. With this configuration, it is possible to perform the ranking search by controlling the weight of the term (search character string or search word) so as not to be affected by the magnitude of the document frequency. Therefore, the document desired by the user can be efficiently searched.

The document retrieval apparatus of the present invention according to claim 2 is
2. The document ranking means according to claim 1, wherein N is the number of documents to be searched, n is the number of documents in which the term appears, k ₁ (k ₁ > 0) and k for each of the terms.
₂ as the adjustment parameter, and

[Numerical equation 14] Is used to calculate the weight of the term. With this configuration, the term weight can be finely adjusted, and the search accuracy can be improved.

The document retrieval apparatus of the present invention according to claim 3 is
2. The document ranking means according to claim 1, wherein N is the number of documents to be searched, n is the number of documents in which the term appears, k ₁ (k ₁ > 0) and k for each of the terms.
₃ (k ₁ > k ₃ ≧ 0) is the adjustment parameter,

[Equation 15] Is used to calculate the weight of the term. With this configuration, the term weight can be finely adjusted, and the search accuracy can be improved.

The document retrieval apparatus of the present invention according to claim 4 is
The document ranking means according to claim 1,
N is the number of documents to be searched, and n is
Number of documents in which the term appears, k₁(K₁> 0), k₂,
k₃(K₁> K₃≧ 0) and k _FourAnd the adjustment parameter
Then the formula

[Equation 16] Is used to calculate the weight of the term. With this configuration, the term weight can be finely adjusted, and the search accuracy can be improved.

The document retrieval apparatus of the present invention according to claim 5 is
The k _{3 of the} adjustment parameter according to claim 3 or 4.
Has a configuration of α / N (0 <α <N). With this configuration, one of the adjustment parameters described in claim 3 or 4 can be determined according to the number of documents to be searched, so that the parameter adjustment becomes simple.

The document retrieval apparatus of the present invention according to claim 6 is
2. The ranking search result obtained by the document ranking means according to claim 1, wherein either a conforming document that conforms to the search condition or a non-conforming document that does not conform to the search condition, or one of the conforming document and the non-conforming document. The re-search instruction inputting means for inputting the re-searching instruction specifying the term included in and the search target for each of the specified terms when the re-searching instruction is input by the re-search instruction inputting means. Number of documents, number of documents in which the term appears, number of conforming documents, number of nonconforming documents,
Adjusting the number of relevant documents in which the term appears in the set of relevant documents, the number of non-compliant documents in which the term appears in the set of non-compliant documents, and the variation in weight due to the difference in the number of documents Weights should be recalculated using a plurality of adjustment parameters, and the weights should be used to add to the search condition from the relevant document based on the frequency information of the terms in the relevant document and the non-compliant document for each term. A relevance evaluation value for selecting a related word is calculated, and the specified terms are ranked according to this relevance evaluation value, and one of them is made a related word, and a new search condition is created in addition to the search condition. And a term ranking means for performing the same. With this configuration, when the user is not satisfied with the search result, it is possible to issue a re-search instruction without re-inputting the initial search condition, which saves the user's time and effort in re-search and improves operability. .

According to a seventh aspect of the present invention, there is provided a document retrieval device,
7. The document ranking means according to claim 6, wherein, when calculating the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, and k ₁ (k ₁ > 0) and k ₂ are the adjustment parameters, and the equation (1)

[Numerical formula 17] And the term ranking means recalculates the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, R is the number of relevant documents, S Is the number of the non-conforming documents, r is the number of the conforming documents in which the term appears, s is the number of the non-conforming documents in which the term appears, k ₅ (k ₅ ≧ 0), k ₆ (K ₆
≧ 0) as the adjustment parameter, and formula (4)

[Equation 18] It has a configuration to recalculate using. With this configuration, the term weight in the re-search can be finely adjusted, and the search accuracy can be further improved.

A document retrieval apparatus according to the present invention of claim 8 is
7. The document ranking means according to claim 6, wherein, when calculating the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, and k ₁ (k ₁ > 0) and k ₃ (k ₁ > k ₃ ≧
0) as the adjustment parameter and equation (2)

[Formula 19] And the term ranking means recalculates the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, R is the number of relevant documents, S Is the number of the non-conforming documents, r is the number of the conforming documents in which the term appears, s is the number of the non-conforming documents in which the term appears, k ₅ (k ₅ ≧ 0), k ₆ (K ₆
≧ 0) as the adjustment parameter, and the formula (5)

[Equation 20] It has a configuration to recalculate using. With this configuration, the term weight in the re-search can be finely adjusted, and the search accuracy can be further improved.

A document retrieval apparatus of the present invention according to claim 9 is
7. The document ranking means according to claim 6, wherein, when calculating the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, and k ₁ (k ₁ > 0), k ₂ , k ₃ (k ₁ > k ₃ ≧
0) and k ₄ are used as the adjustment parameters, and equation (3)

[Equation 21] And the term ranking means recalculates the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, R is the number of relevant documents, S Is the number of the non-conforming documents, r is the number of the conforming documents in which the term appears, s is the number of the non-conforming documents in which the term appears, k ₅ (k ₅ ≧ 0), k ₆ (K ₆
≧ 0) as the adjustment parameter, and the formula (6)

[Equation 22] It has a configuration to recalculate using. With this configuration, the term weight in the re-search can be finely adjusted, and the search accuracy can be further improved.

The document search method of the present invention according to claim 10 is
A first step of searching for a document that matches the search condition input by the search condition inputting device from the set of documents to be searched stored in the document storing device, and a search for each term in the search condition by the calculating device. A second step of calculating the weight using the number of target documents, the number of documents in which the term appears, and a plurality of adjustment parameters for adjusting the variation of the weight due to the difference in the number of documents; A third step of calculating a document score for each document satisfying the search condition, and a fourth step of ranking the documents meeting the search condition by the calculating means using the document score. There is. By this method, it becomes possible to perform ranking search by controlling the term weight so as not to be affected by the size of the document frequency. Therefore, the document desired by the user can be efficiently searched.

A program of the present invention according to claim 11 is the first step of retrieving a document that matches an input retrieval condition from a stored document set of retrieval objects in a computer; For each term, the second step of calculating the weight using the number of search target documents, the number of documents in which the term appears, and a plurality of adjustment parameters for adjusting the variation of the weight due to the difference in the number of documents; The method has a third step of calculating a document score for each document that meets the search condition, and a fourth step of ranking documents that meet the search condition by the document score. With this program, it is possible to perform ranking search by controlling the weight of terms so as not to be influenced by the size of the document frequency. Therefore, the document desired by the user can be efficiently searched. Also, by acquiring the program via the network, the update of this program becomes easy.

According to a twelfth aspect of the present invention, there is provided a recording medium according to the first aspect of the present invention, wherein a first step of searching for a document that matches an input search condition from a stored document set as a search target, A second step of calculating the weight for each term of the search target document number, the number of documents in which the term appears, and a plurality of adjustment parameters for adjusting the variation of the weight due to the difference in the document number; A configuration for recording a program for executing a third step of calculating a document score for each document satisfying the search condition by and a fourth step of ranking documents satisfying the search condition by the document score have. With this configuration, it is possible to perform a ranking search by controlling the term weight so as not to be influenced by the size of the document frequency. Therefore, the document desired by the user can be efficiently searched. Further, the program can be easily updated by moving or exchanging the portable recording medium.

[0021]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. [First Embodiment] FIG. 1 shows a hardware configuration of a document search apparatus according to a first embodiment of the present invention. In FIG. 1, a document search device 100 includes a CPU that controls the entire device.
2 and ROM, RAM for storing programs and necessary data for realizing various functions under the control of this CPU 2.
And a hard disk 4 for storing documents to be searched, search conditions, search results, etc.
And an input unit 5 for inputting necessary instructions and data with a pointing device such as a keyboard or mouse, and a CR
An output unit 6 composed of a T or a liquid crystal display, a flexible disk drive (hereinafter also referred to as FDD) 7 for writing (updating) and reading data to and from a flexible disk (hereinafter also referred to as FD),
Compact disc read-only memory (hereinafter,
CD-ROM that reads data from a CD-ROM)
A communication unit 10 for controlling connection between the drive 8 and a communication line via a communication interface, and for exchanging signals and data with another communication device connected via the communication line.
And each unit 2 to 8 and 10 are connected by a bus 9.

FIG. 2 shows a functional configuration of the document search device according to the first embodiment of the present invention. In FIG. 2, a document registration unit 11 registers the document itself input via the input unit 5 or a recording medium such as a flexible disk in the document database 14, and this function is realized by the CPU 2, the memory 3 and the like. It The search condition input section 12
The user can enter a term (search string,
A search condition including a search word or the like is input, and this function is realized by the input unit 5 or the like. The document ranking unit 15 selects a set of documents satisfying the search condition input by the search condition input unit 12, calculates a document score for each selected document, and sorts the selected documents in descending order of the document score. This function is realized by the CPU 2, the memory 3 and the like. The search result output unit 13
It outputs (displays) the documents searched for in the ranking, and this function is realized by the output unit 6 and the like. The document database 14 is for registering a document file that records the document itself and an index used for retrieval, and this function is realized by the hard disk 4 or the like.

Next, the operation of the document search apparatus 100 according to this embodiment will be described with reference to FIG. First, in step S101, the document ranking unit 15 selects a document satisfying the search condition from the document database 14. Here, if the search condition is composed of a single term, a document including the single term is selected, and if the search condition is a combination of a plurality of terms (OR), any of the plurality of terms is selected. Documents containing or are selected. For other operators such as AND, the same processing as that of a well-known general document search device is performed.

Next, in step S102, the above-mentioned calculation formula (1) is used.

[Equation 23] Is used to calculate the weight for each term in the search condition.

Next, in step S103, a document score is calculated for each document selected in step S101 using the weight calculated in step S102. here,
The document score is calculated using the above formula (8).

[Equation 24] It may be calculated using, for example.

Finally, in step S104, the documents selected in step S101 are sorted in descending order of the document scores calculated in step S103. The document ranking result thus obtained is output by the search result output unit 13.

The weight calculation in step S102 will be described more specifically. Here, the search condition is “Amazon + Tropical”, 1000 documents are registered in the document database (N = 1000), the number of documents including “Amazon” = 4, and “Tropical” is included. The case where the number of documents = 10 is shown. These values can be obtained from the document database 14. At this time, assuming that the adjustment parameters of the equation (1) are k ₁ = 1 and k ₂ = 6 and the base of log is 2, the weight of “Amazon” is n = 4. Therefore, the weight is “log (1000 / 4 + 6) = log (256) = 8 ”. Also, since the weight of "tropical" is n = 100, the weight is "log".
(1000/100 + 6) = log (16) = 4 ”.

On the other hand, the above calculation formula (7)

[Equation 25] When the adjustment parameter is k ₁ = 1 when
The weight of “Amazon” is n = 4, so the weight is “log
(1000/4 + 1) = log (251) = 7.97 ... ”. Also,
Since the weight of "tropical" is n = 100, the weight is "log (100
0/100 + 1) = log (11) = 3.45 ... ”.

Therefore, the case of using the formula (1) is "Amazon" and the "tropical" than the case of using the formula (7).
It can be seen that the weight ratio of is small. Here, the larger the adjustment parameter k ₂ in Expression (1), the less the difference in n influences the weight.

The weight calculation formula used in step S102 is not limited to the formula (1), but the above formula (2) is used.

[Equation 26] Or formula (3)

[Equation 27] May be For example, under the same situation as described above, if equation (2) is used and the adjustment parameters are k ₁ = 1 and k ₃ = 0.1, the weight of “Amazon” is n = 4, and therefore the weight is “log”. ((1000 + 4) / (100 + 4)) = log (9.65 ...) = 3.27
… ” Also, since the weight of "tropical" is n = 100, the weight is "log ((1000 + 100) / (100 + 100)) = log (5.5).
= 2.45 ... ”.

On the other hand, if the adjustment parameters are k ₁ = 1, k ₂ = 6, k ₃ = 0.1, k ₄ = 1 using the equation (3), the weight of "Amazon" is n = 4. The weight is `` log ((100
0 + 24) / (100 + 4)) = log (9.85 ...) = 3.30 ... ”.
Also, since the weight of "tropical" is n = 100, the weight is "l".
og ((1000 + 600) / (100 + 100)) = log (8) = 3 ”.

Therefore, it is understood that the weight ratio of "Amazon" and "tropical" is smaller than that of the equation (7) by using the equation (2) or the equation (3). Regarding the effect of the above-mentioned adjustment parameters, the larger the adjustment parameters k ₃ and k ₄ , the less likely the difference in n influences the weight. The adjustment parameter k ₄ is empirically determined to be α / N.
(Α is a small positive value such as 0.5).

As described above, the document search device 100 according to the first embodiment of the present invention includes the document database 14 (included in the document storage means) for storing the document set to be searched,
From the search condition input unit 12 (included in the search condition input means) for inputting the given search condition and the search target document set accumulated in the document database 14 to the search condition input by the search condition input unit 12 A matching document is searched for, and for each term in the search condition, the number of documents to be searched (N), the number of documents in which the term appears (n), and the variation in weight due to the difference in this document number (n) are adjusted. A weight is calculated using the adjustment parameter, a document score of a document that matches the search condition is calculated using this weight, and a document ranking unit 15 (included in the document ranking means) that ranks documents that match the search condition. Is provided, it is possible to efficiently find the user's desired document by controlling the term weight and performing a ranking search so that it is not affected by the size of the document frequency. Rukoto can.

Further, the document search method according to the first embodiment of the present invention uses the search condition input unit 12 (search condition) from the document set of the search target stored in the document database 14 (included in the document storage means). Step S101 of searching for a document that matches the search condition input by (included in the input means)
By the document ranking unit 15 (included in the calculation means), the number of search target documents (N), the number of documents in which the term appears (n), and this document number (n) Step S102 of calculating the weight using a plurality of adjustment parameters for adjusting the variation of the weight due to the difference
And a step S103 of calculating a document score for each document satisfying the search condition using the weight by the document ranking unit 15, and ranking the documents meeting the search condition by using the document score by the document ranking unit 15. Since step S104 is performed, the user can efficiently find the desired document by controlling the term weight so as not to be influenced by the size of the document frequency and performing the ranking search.

In the first embodiment, the case where the program for implementing the above-described document retrieval method (including FIG. 3) is stored in the memory 3 has been described. A computer-readable recording medium (CD-ROM, FD, magneto-optical disc (MO), mini-disc (MD), rewritable CD (RW-CD-RW) )
(Including) is attached to the document retrieval apparatus 100, the recording medium is read by the CD-ROM drive 8, the FDD 7 and the like to execute the program, and the same effect can be obtained.
With this configuration, the program can be easily updated by moving and exchanging the recording medium.

Further, in the first embodiment, the case in which the program for realizing the above-mentioned document retrieval method (including FIG. 3) is stored in the memory 3 has been described. Even if the program is downloaded from the external device on the network such as LAN to the memory 3 and executed by the communication unit 10 including the communication interface, the same effect can be obtained. With this method, the program can be easily updated via the network.

[Second Embodiment] FIG. 4 shows a second embodiment of the present invention.
3 shows a functional configuration of the document search device according to the embodiment. This is different from the first embodiment in that the re-search instruction input unit 16
And that the term ranking unit 17 is provided. Note that this embodiment has a hardware configuration that is substantially the same as that of the first embodiment, so that FIG. 1 is used and the same reference numerals are given to the same configurations and description thereof is omitted.

In FIG. 4, the re-search instruction input unit 16 inputs an instruction for re-search when the search result (hereinafter also referred to as initial search result) for the search condition input by the user is not satisfied. This function is realized by the input unit 5, the CPU 2, the memory 3, and the like. Further, in the re-search instruction, a document that the user thought to match the search condition in the initial search result (hereinafter, also referred to as a conforming document) and a document that the user did not match (hereinafter also referred to as a non-conforming document) Can be specified respectively.

The term ranking section 17 calculates a relevance evaluation value (TSV) for a term (a word such as a search character string or a search word) selected from the above-mentioned conforming document or non-conforming document or both of them. , In descending order, add an appropriate number of the selected terms as related words to the initial search condition (search condition first input by the user), and add new search condition (hereinafter, also referred to as extended search condition). ) Is created, and this function is realized by the CPU 2, the memory 3, and the like.

The document registration unit 11 registers the document itself, which is input via the input unit 5 or a recording medium such as a flexible disk, in the document database 14, and this function is realized by the CPU 2, the memory 3 and the like. To be done. The search condition input unit 12 is used by the user to input search conditions including terms (search character string, search word, etc.) using a keyboard or the like, and this function is realized by the input unit 5 or the like. The document ranking unit 15 selects a set of documents satisfying the search condition input by the search condition input unit 12, calculates a document score for each selected document, and sorts the selected documents in descending order of the document score. This function is realized by the CPU 2, the memory 3 and the like. The search result output unit 13 outputs (displays) the documents searched for in the ranking, and this function is realized by the output unit 6 and the like. The document database 14 is for registering a document file that records the document itself and an index used for retrieval, and this function is realized by the hard disk 4 or the like.

Next, the re-search operation of the document search apparatus 100 according to this embodiment will be described with reference to FIG. Note that the operation of the initial search is the same as that of the first embodiment, so description will be omitted.

First, in step S201, when a re-search instruction input unit 16 instructs and inputs a re-search instruction and a document from which a term is to be extracted, the term ranking unit 17 instructs the relevant document or Extract terms from non-conforming documents or both. Here, all words obtained by morphological analysis of the instructed document may be extracted as terms, or only independent words may be extracted as terms.

Next, in step S202, the term ranking unit 17 calculates the above equation (4) for each term extracted in step S201.

[Equation 28] Calculate the weight using.

Next, in step S203, the term ranking section 17 calculates a relevance evaluation value (TSV) for each term extracted in step S201 using the weight calculated in step S202. This relevance evaluation value is calculated using the above formula (10).

[Numerical formula 29] It may be calculated using, for example.

Next, in step S204, the term ranking unit 17 selects an appropriate number (for example, 10) of the terms in descending order of the degree of association evaluation value, and adds (terminated by OR) to the initial search condition to expand. Create search conditions.

Then, in step S205 (corresponding to S101 in FIG. 3), the document ranking unit 15 selects documents satisfying the search conditions from the document database 14 according to the first embodiment.

Next, in step S206 (corresponding to S103 in FIG. 3), the document ranking section 15 assigns the weight calculated in step S202 to each document selected in step S205 according to the first embodiment. Use to calculate the document score.

Then, in step S207 (corresponding to S104 in FIG. 3), the document ranking unit 15 sets the documents selected in step S205 according to the first embodiment in descending order of the document scores calculated in step S206. Sort to.

The weight calculation in step S202 will be described more specifically. In the present embodiment, the weight in the re-search is the above-mentioned equation (1).

[Numerical formula 30] Instead, it is calculated using equation (4). This is because the weight reflects the information of the conforming document and the non-conforming document included in the re-search instruction from the user. The number of documents to be searched, the document frequency, and the adjustment parameter are the same as those in the first embodiment. That is, the re-search instruction input unit operated by the user
By inputting an instruction from 16, the number of conforming documents R = 4, the number of nonconforming documents S = 4, the number of documents including “Amazon” in the conforming documents = 2, the number of documents including “tropical” = 4, and the nonconforming documents The number of documents including “Amazon” = 2, and the number of documents including “tropical” = 2. Further, the adjustment parameter appearing only in the equation (4) is k ₅ = k ₆ = 2.

Here, the weight of "Amazon" is n = 4,
Since r = 2 and s = 2, the weight is “0.5 * log (1000/9
96 + 6 * 4/996) + 0.5 * log (2.5 / 2.5) -0.5 * log (4
/996)-0.5*log(2.5/2.5)=0.5*log(256)=4 ". The weight of "tropical" is n = 100, r = 4, s =
Since it is 2, the weight is “0.5 * log (1000/900 + 6100/9
00) + 0.5 * log (4.5 / 0.5) -0.5 * log (100/900) -0.5 *
log (2.5 / 2.5) = 0.5 * log (16) + 0.5 * log (9) = 3.58
… ”

On the other hand, when the above equation (10) is used,
Since the weight of "Amazon" is n = 4, r = 2, and s = 2, the weight is "0.5 * log (1000/996 + 4/996) + 0.5 * log".
(2.5 / 2.5) -0.5 * log (4/996) -0.5 * log (2.5 / 2.5)
= 0.5 * log (251) = 3.98 ... ”. In addition, since the weight of “tropical” is n = 100, r = 4, and s = 2, the weight is “0.5 * log (1000/900 + 100/900) + 0.5 * log (4.5 / 0.
5) -0.5 * log (100/900) -0.5 * log (2.5 / 2.5) = 0.5 * l
og (11) + 0.5 * log (9) = 3.31 ... ”. Therefore,
It can be seen that the ratio of the weights for re-searching “Amazon” and “tropics” is smaller in the case of using the formula (4) than in the case of using the formula (10). The larger the above-mentioned adjustment parameter k ₂ , the less likely the difference in n influences the weight.

As described above, the document search device 100 according to the second embodiment of the present invention further includes the document ranking unit 15.
Regarding the ranking search result by (included in the document ranking means), one of the conforming document that conforms to the search condition and the non-conforming document that does not conform to the search condition, and the conforming document or the non-conforming document A re-search instruction input unit 16 (included in the re-search instruction input means) for inputting a re-search instruction in which the included terms are specified,
When a re-search instruction is input by the re-search instruction input unit 16, for each of the specified terms, the number of documents to be searched (N), the number of documents in which the term appears (n), and the matching documents Number (R), the number of non-conforming documents (S), the number of relevant documents in which the term appears in the set of relevant documents (r), the term appears in the set of non-conforming documents The number of non-conforming documents (s) and a plurality of adjustment parameters for adjusting the variation of the weight due to the difference in the number of documents are used to recalculate the weight, and the weight is used for each term,
A relevance evaluation value (TSV) for selecting a related word to be added to the search condition from the relevance document is calculated based on frequency information of the term in the relevance document and the non-conformance document, and the relevance evaluation value is calculated. With the term ranking unit 17 (included in the term ranking means) that ranks the designated terms by using any of them as a related term and creates new search criteria by adding to the search criteria, If the user is not satisfied with the initial search results,
It is possible to issue a re-search instruction without modifying the initial search conditions. Therefore, the time and effort of the user for setting the re-search can be saved and the operability can be improved.

In the present embodiment, the case where the formula (4) is used as the weight calculation formula used in step S202 has been described, but the present invention is not limited to the formula (4), but also the formula (5).

[Equation 31] Or equation (6)

[Equation 32] The same effect can be obtained by using. That is, by using these equations (5) or (6), it is possible to reduce the influence of the document frequency (n) on the term weight. Further, as an effect of the adjustment parameters, the larger the k _3, k _4, the difference in document frequency is less likely to affect the weights.

In the second embodiment, the case where the program for realizing the above-described document search method (including FIG. 5) is stored in the memory 3 has been described. A computer-readable recording medium (CD-ROM, FD, magneto-optical disc (MO), mini-disc (MD), rewritable CD (RW-CD-RW) )
(Including) is attached to the document retrieval apparatus 100, the recording medium is read by the CD-ROM drive 8, the FDD 7 and the like to execute the program, and the same effect can be obtained.
With this configuration, the program can be easily updated by moving and exchanging the recording medium.

Further, in the second embodiment, the case where the program for realizing the above-mentioned document search method (including FIG. 5) is stored in the memory 3 has been described. Even if the program is downloaded from the external device on the network such as LAN to the memory 3 and executed by the communication unit 10 including the communication interface, the same effect can be obtained. With this method, the program can be easily updated via the network.

[0056]

As described above, according to the present invention, a document matching the input search condition is searched from the document set of the search target, and the number of search target documents is searched for each term in the search condition.
The weight is calculated using the number of documents in which the term appears, and the adjustment parameter for adjusting the variation of the weight due to the difference in the number of documents, and the weight is used to calculate the document score of the document satisfying the search condition, and the search is performed. A document search device, a document search method, a program, and a program that have an excellent effect of improving the search accuracy by controlling the term weight so as not to be affected by the size of the document frequency by ranking the documents that meet the conditions A recording medium can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing a hardware configuration of a document search device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a functional configuration of a document search device according to the first embodiment of the present invention.

FIG. 3 is a flowchart showing an operation of the document search device according to the first embodiment of the present invention.

FIG. 4 is a block diagram showing a functional configuration of a document search device according to a second embodiment of the present invention.

FIG. 5 is a flowchart showing an operation of the document search device according to the second exemplary embodiment of the present invention.

[Explanation of symbols]

2 CPU 3 memory 4 hard disk 5 Input section 6 Output section 7 FDD 8 CD-ROM drive 9 buses 10 Communications department 11 Document Registration Department 12 Search condition input section 13 Search result output section 14 Document database 15 Document Ranking Department 100 Document retrieval device (computer)

Claims (12)

[Claims] 1. A search condition based on a document storage unit that stores a document set to be searched, a search condition input unit that inputs a given search condition, and a document set to be searched that is stored in the document storage unit. Documents that match the search conditions entered by the input means are searched, and for each term in the search conditions,
The weight is calculated using the number of documents to be searched, the number of documents in which the term appears, and the adjustment parameter that adjusts the variation in weight due to the difference in the number of documents, and the document score of the document that matches the search condition is calculated by this weight. A document retrieval device, comprising: a document ranking unit that calculates and ranks documents that meet the retrieval conditions. 2. The document ranking means, for each of the terms, N is the number of search target documents, n is the number of documents in which the term appears, and k ₁ (k ₁ > 0) and k ₂ are adjusted. As a parameter, the formula [Formula 1] The document search device according to claim 1, wherein the weight of the term is calculated by using. 3. The document ranking means is provided for the term
For each, N is the number of documents to be searched, and n is the target document.
The number of documents in which a game appears₁(K₁> 0) and k₃(K₁
> K ₃≧ 0) as the adjustment parameter, [Equation 2] Is characterized in that the weight of the term is calculated using
The document search device according to claim 1. 4. The document ranking means, for each of the terms, N is the number of documents to be searched, n is the number of documents in which the term appears, k ₁ (k ₁ > 0), k ₂ , k ₃ (K
₁ > k ₃ ≧ 0) and k ₄ are the adjustment parameters, and the equation The document search device according to claim 1, wherein the weight of the term is calculated by using. 5. The adjustment parameter k ₃ is α / N (0 <
The document retrieval device according to claim 3 or 4, wherein α <N). 6. The ranking search result by the document ranking means is either a conforming document that conforms to the search condition or a non-conforming document that does not conform to the search condition, or one of the conforming document and the non-conforming document. The re-search instruction inputting means for inputting the re-searching instruction specifying the term included in and the search target for each of the specified terms when the re-searching instruction is input by the re-search instruction inputting means. The number of documents, the number of documents in which the term appears, the number of relevant documents, the number of non-conforming documents, the number of relevant documents in which the term appears in the set of relevant documents, and the number of relevant documents in the set of non-compliant documents The number of non-conforming documents in which the term appears, and a plurality of adjustment parameters for adjusting the variation of the weight due to the difference in the number of documents are used to recalculate the weight, and the weight is used. For each term, a relevance evaluation value for selecting a related word to be added to the search condition from the relevance document is calculated based on frequency information of the term in the relevance document and the non-conformance document, and the relevance degree is calculated. 2. The term ranking means for ranking the designated terms according to an evaluation value to make any one of the related terms a related term and creating a new retrieval condition in addition to the retrieval condition is provided. Document retrieval device. 7. The document ranking means calculates the weight of the term, for each of the terms,
N is the number of documents to be searched, n is the number of documents in which the term appears, and k ₁ (k ₁ > 0) and k ₂ are the adjustment parameters. And the term ranking means recalculates the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, R is the number of relevant documents, S Is the number of the non-conforming documents, r is the number of the conforming documents in which the term appears, s is the number of the non-conforming documents in which the term appears, k ₅ (k ₅ ≧ 0), k ₆ (K ₆
≧ 0) as the adjustment parameter, and the equation The document retrieving apparatus according to claim 6, wherein the document is recalculated using. 8. The document ranking means, when calculating the weights of the terms, for each of the terms,
N is the number of documents to be searched, n is the number of documents in which the term appears, and k ₁ (k ₁ > 0) and k ₃ (k ₁ > k ₃ ≧ 0) are the adjustment parameters. And the term ranking means recalculates the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, R is the number of relevant documents, S Is the number of the non-conforming documents, r is the number of the conforming documents in which the term appears, s is the number of the non-conforming documents in which the term appears, k ₅ (k ₅ ≧ 0), k ₆ (K ₆
≧ 0) as the adjustment parameter, and the formula The document retrieving apparatus according to claim 6, wherein the document is recalculated using. 9. The document ranking means calculates the weights of the terms for each of the terms.
N is the number of documents to be searched, n is the number of documents in which the term appears, k ₁ (k ₁ > 0), k ₂ , k ₃ (k ₁ > k ₃ ≧ 0) and k ₄ are the adjustment parameters, Formula [Equation 8] And the term ranking means recalculates the weight of the term, N is the number of documents to be searched, n is the number of documents in which the term appears, R is the number of relevant documents, S Is the number of the non-conforming documents, r is the number of the conforming documents in which the term appears, s is the number of the non-conforming documents in which the term appears, k ₅ (k ₅ ≧ 0), k ₆ (K ₆
≧ 0) as the adjustment parameter, and the equation The document retrieving apparatus according to claim 6, wherein the document is recalculated using. 10. A first step of retrieving a document matching a retrieval condition input by a retrieval condition inputting unit from a set of retrieval target documents stored in a document storing unit, and a calculating unit in the retrieval condition. A second step of calculating a weight for each term using a plurality of search target documents, a number of documents in which the term appears, and a plurality of adjustment parameters for adjusting variations in weight due to a difference in the number of documents, and the calculating means. And a third step of calculating a document score for each document satisfying the search condition by using the weight, and a fourth step of ranking the documents meeting the search condition by the calculating means using the document score. A document search method comprising: 11. A first step of searching a document that matches a search condition input from a stored document set of search targets, the number of search target documents for each term in the search condition, The second step of calculating a weight using a plurality of adjustment parameters for adjusting the number of documents in which the term appears and the weight variation due to the difference in the number of documents, and for each document that matches the search condition by the weight. A program for executing a third step of calculating a document score, and a fourth step of ranking documents matching the search condition by the document score. A first step of searching a document that matches a search condition input from a stored document set of search targets, the number of search target documents for each term in the search condition, The second step of calculating a weight using a plurality of adjustment parameters for adjusting the number of documents in which the term appears and the weight variation due to the difference in the number of documents, and for each document that matches the search condition by the weight. A computer-readable recording medium recording a program for executing a third step of calculating a document score and a fourth step of ranking documents matching the search condition by the document score.