JP2773682B2 - Applicable feedback device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention relates to an information retrieval system,
The present invention relates to a matching feedback device that performs a new search based on a matching determination as to whether or not a search result matches a searcher's request, and outputs new search results in order of matching.

[0002]

2. Description of the Related Art When searching for records in a database using a conventional information search system, a searcher creates a search formula and performs a search. If the search results are not satisfactory, it is common for the searcher to re-create the search formula or to modify the previous search formula partially and start the search again. It is difficult for a simple searcher to create a search expression, and in many cases a record that meets the request cannot be obtained even if the search is repeated.

[0003] Therefore, there has been proposed a matching feedback system in which a matching determination with respect to a searcher's result is reflected in the search formula and the search formula is automatically corrected or re-created.
For example, “Relevance Weighting
of Search Terms "(SE Robe)
rtson, Karen Spark Jones
Written by the Journal of the American
Society for Information
Science, vol. 26, p. 129-146,
(1976) (hereinafter referred to as Reference 1), a searcher inputs a record as a search result as a conforming record or a nonconforming record, examines the conforming / nonconforming record, and finds a certain word in the conforming record. Calculate the probability that the record is a matching record when the word appears in a record from that value. The value of the probability can be considered as a weight indicating the ability to search for a matching record of the word, and is calculated as follows.

[0004] For a word for which a search request is made from a set of records for which a searcher has previously determined that the word is relevant / non-conforming, a: the number of records in which the word appears in the relevant record b: the word in the relevant record C: the number of records in which the word appears in non-conforming records d: the number of records in which the word does not appear in non-conforming records, and the weight of the word is log ((a + 0. 5) (d + 0.5)) / ((c + 0.5) (b + 0.5)). Further, in Reference 1, as a result of comparing this weighting method with other weighting methods by experiments, it is shown that the retrieval efficiency is the highest.

[0005] Further, in the invention described in Japanese Patent Application Laid-Open No. 02-245971, "Information Search Processing Method and Apparatus" (hereinafter referred to as Document 2), it is necessary to determine whether or not a search result meets a searcher's request. , Extract words that appear in matching records and do not appear in non-conforming records, select valid words as search words from the extracted words, add the words as new search words to the search formula, and search It is carried out.

In the invention described in Japanese Patent Application Laid-Open No. 05-151271, "Information Retrieval Apparatus" (hereinafter referred to as Reference 3), a searcher inputs several conforming records and several non-conforming records of a record as a retrieval result, and A technique is described in which a word whose probability of appearing in a record is higher than the probability of appearing in a matching record and a non-matching record is selected as a search word, and a new search formula is generated.

[0007]

However, the methods described in the above-mentioned documents 1 to 3 have the following problems.

First, in References 1 to 3, the selection criterion for selecting a word in a matching record as a search word is ambiguous. For example, in the database, "Title", "Abstract",
When it is composed of "text" items, words appearing in the title are considered to be more important than words appearing in the abstract, but these methods take this point into consideration. It has not been.

[0009] In particular, in Document 2, the search term is intended for words in the text indicating the contents of the document, such as titles and abstracts. , A thesaurus, etc., which are specific to the words in the text, but the author, the affiliation of the author, and the journal that contains it if it is a journal article Names can be considered, but these elements are not considered in selection and weighting of search words.

[0010] Further, in References 1 to 3, the case where a nonconforming record is not input is not considered. If there is no nonconforming record, all the words in the conforming record will be selected as search words, but words that appear frequently in the database, for example, "information" in the informatics database
A word that appears in general, such as, is also selected as a search word.

[0011] In Document 3, the searcher must input a conforming record and a nonconforming record.
When only matching records are input, the search term cannot be selected, and the weight of the search term cannot be calculated.

As described above, the conventional adaptive feedback method has problems to be solved, such as selection of a search word and handling when there is no incompatible record.

An object of the present invention is to solve the above-mentioned problems,
The search term is not limited to text items such as titles and abstracts, but extends to authors, author affiliations, etc., and even when no nonconforming records are entered, selection of search terms,
It is an object of the present invention to provide a weighted adaptive feedback device.

[0014]

Means for Solving the Problems The present invention, the first invention,
A database search system in which a searcher creates a search formula, searches a database using the search formula, and outputs a search result, wherein the search result is a match-determined record for which the searcher has determined whether the search is compatible or not. In the conformance feedback device for creating a file and re-searching the database by creating a search expression again with the conformance-determined record file, in the conformance-determined record file, a search result that has been determined to be suitable The word to which the word belongs and the item to which the word belongs are taken out, and the occurrence of each word is created in the judgment-determined appearance record table in which the word appears in the item of the conformity-determined record file in what proportion. Searching the database again with reference to a record table creating unit and the occurrence record table for each judgment A search word selection unit for selecting a search word and a search item for search, a search expression generation unit for creating a new search expression from the search word, the search item and the determined record file, and executing a search from the search expression And a search execution unit for obtaining a search result.

[0015] In a second aspect based on the first aspect, a search term weight calculator for calculating the weight of the search term selected by the search term selector with reference to the occurrence record table for each determination. In the search result obtained by the search expression execution unit, extract how much the search term selected by the search word selection unit exists, and refer to the search word weight calculation unit for the weight of each search word. A record that outputs a search result in order of relevance, which is a search result sorted in the order of relevance, with a value obtained by adding all the weights of the search terms present in the search result as the relevance of the search result. And a fitness calculating unit.

Further, a third invention is a database search system in which a searcher creates a search formula, searches a database using the search formula, and outputs a search result, wherein the search result is suitable for the searcher. Create a conformity-determined record file in which the determination of nonconformity has been made, in the conformance feedback device to create a search formula again by the conformance-determined record file and search the database again, in the conformance-determined record file, The words appearing in the search results determined to be compatible and the items to which the words belong were taken out, and the percentage of the words appearing in the items of the matched determined record file was described. A judgment-based occurrence record table creating unit for creating a judgment-based occurrence record table, and weights of the respective items in the database A search term selection unit for selecting a search term and a search item for re-searching the database with reference to the item knowledge holding the number, the judgment record table or the weight coefficient of the item knowledge, and the search term And a search formula generation unit for generating a new search formula from the search item and the determined record file; and a search execution unit for executing a search from the search formula to obtain a search result.

In a fourth aspect based on the third aspect, the search term weight is calculated by referring to the judgment-based occurrence record table and the item knowledge, the weight of the search term selected by the search term selection section. A calculating unit that extracts how many search terms selected by the search term selecting unit are present in the search results obtained by the search formula executing unit, and calculates a weight for each of the search terms by the search term weight calculation. The value obtained by adding all the weights of the search terms present in the search result is referred to as the relevance of the search result, and the relevance order search result is a search result sorted in the relevance order. And a record matching degree calculation unit that outputs the record matching degree.

According to a fifth aspect of the present invention, in the third and fourth aspects, when the search word selection section selects a search word,
Determine whether a non-conforming record exists in the conformance-determined record file, if there is a non-conforming record, select a search term by referring to the judgment-based occurrence record table, and if there is no non-conforming record, check the item knowledge. It is characterized in that a search word is selected by referring to.

[0019]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the configuration of the adaptive feedback device according to the present invention. FIG. 2 is a table showing an example of the appearance status of a certain word that has been determined to be suitable. FIG. 3 is a diagram showing an example of a processing flow of the search term selection unit 104 of FIG. FIG. 4 shows the search term weight calculator 10 of FIG.
FIG. 11 is a diagram showing an example of the processing flow of No. 6; FIG. 5 and FIG. 8 are examples of the judgment-specific appearance record table 103 in FIG. 6 and 10 show examples of the weight calculated by the search word weight calculator 106 in FIG. 7 and 11 show the search expression generation unit 1 of FIG.
It is an example of the search formula generated in 08. FIG. 9 is an example of the item knowledge 105 of FIG.

In FIG. 1, a conformity-determined record file 101 is a file in which a searcher has once performed a search, and has determined whether each record obtained as a result of the search is suitable or not.

The judgment-specific appearance record table creating unit 102 reads the conformity-determined record file 101, and extracts words appearing in the conformance record. At this time, what should be selected as an item from which a word is extracted is determined in advance by a database administrator or a database search system administrator.

The item selected here may be any item expressing the characteristics of each record. For example, in the case of a journal article database, titles, authors, author affiliations, journal names, abstracts, keywords, etc.
Further, in the case of a database relating to a company, there are a company name, an executive name, a business type, a series, and the like, and items such as a record ID of each record are inappropriate.

When selecting an item, texts such as titles and abstracts must be used to remove unnecessary words once using natural language analysis, etc., and authors, affiliations, journal names, etc. In this way, items whose values are to be extracted as one word are selected. When extracting words actually appearing in the matching record, words are extracted from a predetermined item by a method suitable for the item.

Next, for each word, the number of records appearing in the conforming / non-conforming record for each item is counted.
Hereinafter, the numbers are a, b, c, and d as shown in FIG. a: The number of records in which the word appears in the item in the matching record b: The number of records in which the word does not appear in the item in the matching record c: The word appears in the item in the non-matching record Number of records d: Number of records that are nonconforming records and the word does not appear in the item. Finally, the appearance-by-judgment record creation unit 102 describes the numbers a, b, c, and d of the extracted words. A judgment-based appearance record table 103 is created. The appearance record table for each judgment will be described later in more detail.

Further, in the present embodiment, the same weighting method as in Reference 1 is employed, as shown in the experiment introduced in Reference 1, in which the performance is better than other weighting methods. . In addition, when weighting is required,
This is very simple because there is no need for information on the frequency such as how many times the word has appeared in the non-conforming record, and no more information. With the above in mind,
The weighting method in this embodiment is the same as that in Reference 1.

The search term selection unit 104 selects a search item and a search term for creating a new search formula using the appearance record table 103 for each judgment and the item knowledge 105. Further, the processing differs depending on whether or not there is a nonconforming record in the conformance-determined record file as follows.

First, when there is no nonconforming record, an item having a weight coefficient equal to or more than a threshold value is set as a search item by referring to the item knowledge 105, and a word appearing in the search item is set as a search word.

The item knowledge 105 describes items such as a title, an abstract, an author, and the like constituting the database, and a weight coefficient indicating how useful the item is in searching for a matching record. This weighting factor may be arbitrarily input by the database administrator, or may be calculated in consideration of the type of word appearing in the item.

When there is a non-conforming record, the probability that all words and items in the judgment-specific appearance record table 103 appear in the conforming record, that is, (the number of records appearing in the conforming record) / (conforming Number of records) = a / (a + b) (A) and the probability of occurrence in all records for which conformity has been determined, that is, (number of records appearing in conforming or nonconforming records) / (number of conforming records + nonconforming) (The number of records) = (a + c) / (a + b + c + d) (B)

If the probability of appearing in a matching record is higher than the probability of appearing in a matching record and a non-matching record, it is considered that there is a greater possibility that a matching record can be searched for, so that a / (a + b)> (a + c) / (A + b + c + d) (C) Search for combinations of words and items that satisfy the following, and select and output them as new search words and search items.

The search term weight calculator 106 is used for the search term selector 1.
The weight of the search term selected in 04 is calculated. First, the search term weight calculator 106 obtains the values of a, b, c, and d in the search item of each search term in the appearance record table 103 for each judgment. This weight indicates the likelihood that each search term can search for a new matching record. Therefore, it is considered that there is a high possibility that words that appear many times in matching records but do not appear in non-matching records will search for new matching records. That is, the values of a and d are as large as possible,
Since it is desirable to use a search word in which the values of b and c are as small as possible,
The following weighting methods are conceivable. Which weighting formula is used may be selected by a database administrator or a searcher. (A + 0.5) (d + 0.5) (1) ((a + 0.5) (d + 0.5)) / ((c + 0.5) (b + 0.5)) (2) log (( a + 0.5) (d + 0.5)) / ((c + 0.5) (b + 0.5)) (3) In this case, if there is no nonconforming record, c = 0 and d =
0, and the weights of all the search words are equal in the expression (1). Further, when there is only one matching record, b = 0, so that the weights of all the search words are equal regardless of any of the expressions (1), (2), and (3).

Therefore, as shown in the processing flow of FIG. 4, when there is no nonconforming record and only one conforming record exists in the conformity-determined record file 101, the value of a is set to the item knowledge. The weight is multiplied by the weight coefficient in 105. If there is no nonconforming record and there are two or more conforming records, there may be a word where b ≠ 0,
The weight is calculated by either equation (2) or equation (3), and the weight is calculated by multiplying the weight by the weight coefficient of the search item.

The conformity-determined record file 10
If there is a nonconforming record in 1, (1),
The weight may be calculated by either of the formulas (2) and (3). The features of the above formulas are as follows, and which weighting formula is used depends on the database manager, the search system manager, Alternatively, a searcher or the like may freely select.

Equation (1) is obtained by simply multiplying a and b. a + b = the number of matching records, c + d = the number of non-matching records, and these values are the same for all search terms. Therefore, the larger the value of a, the smaller the value of b, and the larger the value of d, the smaller the value of c. Therefore, even if a value obtained by simply multiplying a and b is larger, the larger this value is, the more the matching record becomes. It can be said that it has appeared.

Equation (2) is obtained by further dividing the value of equation (1) by the product of b and c. As in the equation (1), a + b = the number of matching records and c + d = the number of non-matching records, and these values are the same for all search terms. Therefore, the larger the values of a and d, the smaller the values of b and c. . As the values of b and c are smaller, the value of the denominator is smaller, and as a result, the weight is larger.

Equation (3) is obtained by taking the logarithm (base 10) of the value of equation (2). By taking the logarithm, the value differentiated by the equation (2) is a value larger than 10, for example, 2.301 and 2.477 for 200 and 300, respectively, and the ratio of the difference is small and smaller than 10. For values, for example 2 and 3, 0.301 and 0.47
7 and the ratio of the difference is almost maintained. That is,
The feature that the weight of the large value by far from the value of the expression (2) disappears and falls within the range of about 0 to 3, and the ratio of the value difference in the range of less than 10 in the value of the expression (2) is almost maintained. There is.

On the other hand, the search expression generation unit 107 receives the search word and the search item selected in the search word selection, and generates a search expression. In the search formula, when there are a plurality of search words in the same search item, they are connected by OR, and different items are also connected by OR.

Also, when generating a search expression, the reconciliation-determined record file 101 is referred to, and a record once a rejection is determined by the searcher is not included in the result.
Generate a search expression.

The search execution unit 108 performs a search using the search generated by the search expression generation unit 107, and obtains a search result.

The record matching degree calculation unit 109 sets the matching degree of each record of the search result obtained by the search execution unit 108 as the total sum of the weights of the search words included in the record. More specifically, if the search item selected by the search word selection unit 104 includes the same search word, the search word weight calculation unit 106
Is calculated by adding the weights obtained in (1).

Here, an example will be described. For the purpose of explanation, the search target is a database of journal articles related to library and information science, and the database items include "title", "abstract", "author", "affiliation", and "magazine name". This is an example of a library information science database. In addition, in the relevance-determined record file 101, the searcher sets 10 relevance records and 10
It is assumed that a total of 20 records are described.

The appearance-by-judgment record table creation unit 102 retrieves, from the ten conformance records in the conformity-determined record file 101, an item determined in advance by a database administrator or a database retrieval system administrator. To extract words by item.
In this example, the items to be extracted are “title”, “abstract”, “author”, “affiliation”, and “magazine name”. For the items "author", "affiliation", and "magazine name", the value of the item itself is extracted as a word.

Since "title" and "abstract" are generally text items, unnecessary words are deleted using natural language analysis, and the remaining words are extracted. The natural language analysis according to the present embodiment is performed, for example, in “Basic technology of natural language processing” (Hirogo Nomura, published by the Institute of Electronics, Information and Communication Engineers, 1988)
The natural language analysis as described in Chapter 1 and Chapter 2 of the year) is performed, and unnecessary words are used here. The endings, auxiliary verbs, adjunct particles, final particles, accessory particles, case particles, and parallel particles are used. I do.

For the extracted words, the number of records corresponding to a, b, c, and d in FIG. 2 is counted for a total of 20 conforming records and nonconforming records.
The determination-specific appearance record table 103 described in (a) is created. According to FIG. 5A, the word "university library" appears in eight out of ten matching records in the title item, and does not appear in ten title items in the non-matching record. 9 in the conformance record in the abstract item
And two non-conforming records. The author "Taro Yamada" appears in five out of ten conforming records, and does not appear in ten non-conforming records.

The search term selection unit 104 follows the flow of the processing shown in FIG.
In order to select a search term and a search item from, an appearance record table for each judgment shown in FIG. 5B is created.

The processing will be described in detail. Since there is a non-conforming record in the conformity-determined record file 101, the probability that each word appears in the conforming record and the probability that it appears in all the records for which conformity has been determined are calculated by the above-described equations (A) and (B). Is calculated. For example, in the word “university library” in the title item of FIG. 5, the probability of appearing in a matching record is 8/10 = 0.8, the probability of appearing in all records is (8 + 0) /20=0.4, and The probability of appearance is higher. Therefore, “university library” in the title item is selected as a search word.

Similarly, when calculating for “public library” in the title item, the probability of appearing in the matching record is 1 /
10 = 0.1, probability of appearing in all records is (1 + 3)
The probability of appearing in all records at /20=0.2 is higher and is not selected as a search term. With respect to the words extracted in this way, the expression (C) is calculated to create a search-specific appearance record table of FIG. 5B in which a search word and a search item are selected.
The search term and the search item selected here are stored in the search expression generation unit 10.
7 is passed to the search term weight calculation unit 106.

The search term weight calculator 106 calculates the search term weight in FIG.
The weight of the search term selected by the (search term selection unit 104) is calculated according to the processing flow of FIG. The processing content will be described in detail. Now, since there is a nonconforming record, the weight of the search word is calculated by using any of the above formulas (1), (2) and (3). For example, “University Library” in the title item is 8.5 × 10.5 =
89.25, and according to equation (2), (8.5 × 10.
5) / (2.5 × 0.5) = 71.4, and according to equation (3), log ((8.5 × 10.5) / (2.5 × 0.
5)) = 1.85. Similarly, when weights are calculated for other words, the result is as shown in FIG.

On the other hand, in the retrieval formula generation unit 107, FIG.
The search term and its search item are examined, and a search expression is generated so that records having the search term in the search item can be searched.
The search terms “university library” and “network” in the same search item, for example, the title item, are connected by an OR operator. Similarly, "Academic Library" and "Network" are connected in the abstract item, "Taro Yamada" and "Hanako Tanaka" are connected in OR by the author item, and the search items are also connected in OR. Furthermore, the record numbers included in the conformance-determined file are linked by the NOT operator so that a record that is identical to the conformity-determined file and the search result is not searched first. FIG. 7 is an example of a search expression generated by the search expression generation unit 107. This search formula is passed to the search execution unit 108, and a search is performed in the database.

The record matching degree calculation unit 109 sets the weight of the record obtained as a result of the search using the search formula shown in FIG. For example, the relevance of a record in which “university library” and “network” appear in the title item is 1.85 + 0.49 = 2.34 if the weighting formula of the search term is (3).
become. Thereafter, the similarity is calculated in the same manner, the records are sorted in the order of the similarity, and the fitness search result 110 is output.

Further, another example will be described. For the sake of explanation, it is assumed that the conformity-determined record file 101 contains only one conforming record. Receiving the conformance-determined record file 101, the judgment-specific appearance record table creation unit 102 extracts words existing in the conformance record, and determines the word-by-judgment occurrence record table 10 shown in FIG.
Create 3. Since there is no nonconforming record, the search term selection unit 104 searches the item knowledge 105 of FIG. 9 for an item whose weight coefficient is equal to or larger than the threshold value according to the processing flow when there is no nonconforming record of FIG. Let the appearing word be a search word. Here, if the threshold is set to 0.8, the search items are the title, abstract, author, and cited document, and those selected as search terms are those marked with a circle in FIG. 8B.

The search word weight calculator 106 calculates the weight of the search word shown in FIG. 8B according to the processing flow shown in FIG. Now, since there is no nonconforming record and there is only one conforming record, the weight is set to the value 1 of a and the item knowledge 10 in FIG.
The value is a value obtained by multiplying the weight coefficient described in No. 5.

For example, the weight of the “reference” in the abstract item is 1 × 0.8 = 0.8. FIG. 10 shows the search words and the weights thus calculated. Also, the search expression generation unit 1
Reference numeral 07 generates a search expression by connecting the search term and the search item of FIG. 8 selected by the search term selection unit 104 with an OR operator. At this time, if the record number of the conforming record described in the conforming record file 101 is # 21, NOT is set so that this record is not included in the search result.
Concatenate with operators. FIG. 11 shows an example of the search formula generated as described above.

The search execution unit 108 executes the search formula shown in FIG.
In step 9, the relevance of each record of the search result is calculated based on the weight of the search word in FIG. 10, and the relevance-ordered search result 110 sorted in the relevance order is output.

When the searcher performs a re-judgment on the output result, the information is added to the judgment-specific appearance record table 103, a new search is performed, and the process is continued until the searcher is satisfied.

[0057]

As described above, according to the present invention, a searcher can obtain search results newly arranged in the order of relevance only by determining relevance / non-relevance.

In the calculation of the degree of relevance, the weight of the search term is calculated for each item, such as the title and the abstract, in addition to the words of the author and the affiliation of the author. It is possible to gain a degree.

Further, by using the item knowledge, a search term can be selected and weighted even when no nonconforming record is input or when only one matching record is input.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration diagram according to the present invention.

FIG. 2 is a diagram illustrating the appearance of words according to a matching determination;

FIG. 3 is a diagram showing a processing flow of a search term selection unit.

FIG. 4 is a diagram showing a processing flow of a search term weight calculation unit.

FIG. 5 is an example of an appearance record table for each determination.

FIG. 6 is an example of a weight calculated by a search word weight calculator.

FIG. 7 is an example of a search expression generated by a search expression generation unit.

FIG. 8 is an example of an appearance record table for each determination.

FIG. 9 is an example of item knowledge.

FIG. 10 is an example of a weight calculated by a search word weight calculator.

FIG. 11 is an example of a search formula generated by a search formula generation unit.

[Explanation of symbols]

 101 Record file for which conformity has been determined 102 Appearance record table creation unit for each judgment 103 Appearance record table for each judgment 104 Search term selection unit 105 Item knowledge 106 Search term weight calculation unit 107 Search expression generation unit 108 Search execution unit 109 Record conformance calculation unit 110 Matching order search results

Continuation of the front page (56) References JP-A-2-2495971 (JP, A) JP-A-3-294963 (JP, A) JP-A-4-281565 (JP, A) JP-A-6-318234 (JP) , A) JP-A-5-204975 (JP, A) JP-A-5-151271 (JP, A) “Relevance Weighting of Search Terms” (SE Robertson, Karen Spark Jones, Journal of the Journal) American Society for Information Science, vol. 26, p. 129-146, 1976) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06F 17/30

Claims (5)

(57) [Claims] 1. A database search system in which a searcher creates a search formula, searches a database using the search formula, and outputs a search result. In the database search system, it is determined whether or not the search result matches or does not match the searcher. A conformance feedback device that creates a conformity-determined record file, creates a search expression again using the conformance-determined record file, and searches the database again. The words appearing in the result and the item to which the word belongs are taken out, and a judgment-based appearance record table stating how much the word appears in the item of the conformance-determined record file is described. A judgment-based occurrence record table creation unit to be created, and the database with reference to the judgment-based occurrence record table A search word selection unit for selecting a search word and a search item for performing a search again; a search expression generation unit for creating a new search expression from the search word, the search item, and the determined record file; and a search from the search expression And a search execution unit that obtains a search result by executing the search. 2. A search word weight calculation unit for calculating a weight of a search word selected by the search word selection unit with reference to the appearance record table for each judgment, and a search result obtained by the search expression execution unit. Extracting the number of search words selected by the search word selection unit, inputting the weight of each search word with reference to the search word weight calculation unit, and presenting the weight in the search result. A record relevance calculating unit that outputs a relevance order search result, which is a search result sorted in the relevance level, with a value obtained by adding all weights for each search word as the relevance level of the search result. The adaptive feedback device according to claim 1. 3. A database search system in which a searcher creates a search formula, searches a database using the search formula, and outputs a search result, and determines whether the search result matches or does not match the searcher. A conformance feedback device that creates a conformity-determined record file, creates a search expression again using the conformance-determined record file, and searches the database again. The words appearing in the result and the item to which the word belongs are taken out, and a judgment-based appearance record table stating how much the word appears in the item of the conformance-determined record file is described. An occurrence record table creation unit for each judgment to be created, and an item information holding a weight coefficient of each item of the database. A search term selection unit for selecting a search term and a search item for re-searching the database with reference to the determination-based occurrence record table or the weight coefficient of the item knowledge; and the search term, the search item, and the A matching feedback device comprising: a search formula generation unit that generates a new search formula from a determined record file; and a search execution unit that performs a search from the search formula to obtain a search result. 4. A search term weight calculation unit for calculating a weight of the search term selected by the search term selection unit by referring to the appearance record table for each judgment and the item knowledge, and obtained by the search expression execution unit. In the search result obtained, extract how much the search term selected by the search term selection unit exists, input the weight of each search term by referring to the search word weight calculation unit, and input the weight to the search result. A record relevance calculating unit that outputs a relevance order search result, which is a search result sorted in the relevance level, with a value obtained by adding all weights for each of the existing search terms as the relevance level of the search result. The adaptive feedback device according to claim 3, characterized in that: 5. When the search term selection unit selects a search term, it determines whether or not there is a nonconforming record in the record file for which conformity has been determined. To select a search term,
The matching feedback device according to claim 3, wherein when there is no nonconforming record, a search term is selected with reference to the item knowledge.