JP3502537B2 - Index deriving device and method, and computer-readable medium recording index deriving program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an index derivation apparatus and method capable of suppressing the number of derivations when deriving a new search index from a previously given search index, and a computer recording an index derivation program. The present invention relates to a readable medium.

[0002]

2. Description of the Related Art The inventors derive a new index from a search index in a database (original database) in which one term is preliminarily assigned to each record as a search index, and use this as a new index. A technique that can improve the search efficiency of a database by adding it as a special search index has been proposed (Japanese Patent Application No. 8-331039).

In the above technique, a morpheme analysis is performed in which terms (original index terms) assigned as a search index to each record of the original database are divided into word units, and each word obtained as a result of the analysis is started. The term (derivative index term) is derived from the original index term by extracting the words after the beginning of each step, and the stepwise record derivation is performed, and among the words obtained as a result of the morphological analysis, the original index term Determining unnecessary indexes by checking the frequency of occurrence of the word at the end of the original index in the original index term in the entire database and deleting the terms that exactly match these high-frequency words from the new search index. Was trying to suppress.

[0004]

However, in such a technique, (1) a term composed of a plurality of words is not deleted if even one of them has a low appearance frequency, and it is not deleted at the time of search. Unnecessary indexes will also remain,
(2) Since the frequently appearing index is deleted, synonyms (when the same word is used with different meanings) are deleted without distinction, and even the necessary index may be deleted. There was a problem that there was.

It is an object of the present invention to delete a term composed of a plurality of words if the frequency is high, to determine whether or not to delete synonyms and synonyms, and to eliminate unnecessary indexes. An object is to provide an index derivation device and method capable of effectively preventing derivation, and a computer-readable medium recording an index derivation program.

[0006]

In the present invention, in order to solve the above-mentioned problems, a morpheme for performing morpheme analysis in which a term (original index term) assigned as a search index to each record of an original database is divided into word units. And a stepwise record derivation means for creating a term starting from each word obtained as a result of the analysis (derivative index term) by extracting words after each starting word from the original index term. , the index derivation device in a database generating apparatus for generating at least one search target database storing the record with the derived index term as the index for the new search of the created derived index term, stepped records derived unit Derived index at
Source index term is
Derived from information indicating whether it was derived from the index term
Derivation source information addition means to be given, term aggregation means for checking the appearance frequency of the derived index terms in the entire database, and words positioned in front of terms with high appearance frequency
(Forward position word) is the original index obtained from the derivation source information.
Front position word acquiring means for acquiring from the word
Depending on the content of the morpheme information of the position word
Dex terms should be removed from the search index
Decide whether or not to delete the derived index terms before
Determinants for unnecessary records to be deleted from the search index
It is characterized in that it is provided with an unnecessary record deleting means composed of steps .

According to the above configuration, it is possible to delete a record having a high appearance frequency in a unit larger than a word, and thus a term composed of a plurality of words can be deleted if the frequency is high , and a plurality of words can be deleted. Consists of
According to the origin of the word as a whole, that is, distinguishing homonyms
You can decide whether to delete it .

[0008]

[0009]

At this time, it should be deleted by referring to the unnecessary record determination rule group having the morpheme information as a key and expressing the deletion condition by the content of the morpheme information of the forward position word of the derived index term. The derived index term may be determined.

[First Embodiment ] FIG. 1 shows a first embodiment (however, not included in the scope of the claims) of an index derivation device of the present invention, in which 1 is an original. A database, 2 is a search index, 3 is a search target database, and 10 is an index derivation device.

The original database 1 is an index derivation target database in which one term is added to each record as a search index (search key). Search index 2 is the original database 1
Is a target index generated by processing the index derivation device 10. Search target database 3
Is a database in which the search index 2 is added to the original database 1.

The index derivation device 10 comprises a data reading unit 11, a morpheme analysis unit 12, a stepwise record derivation unit 13, a term aggregation unit 14, a frequent term table 15, an unnecessary record deletion unit 16 and an information output unit 17. There is.

The data reading section 11 reads information from the original database 1 record by record. The morphological analysis unit 12 divides the information read by the data reading unit 11, that is, the term (original index term) given as a search index into each record of the original database into word units. The stepwise record derivation unit 13 creates the derived index terms (stepwise records) that start with each word obtained by the morpheme analysis unit 12 by extracting the words after each starting word from the original index term.

The term totaling unit 14 determines the occurrence frequency of each derived index term (staircase record) obtained by the staircase record deriving unit 13 in the entire database in terms of the character face or a pair of the character face and the morpheme information. As a result, a frequently-used term table 15 in which the appearance frequencies of the terms are accumulated is created. The unnecessary record deletion unit 16 refers to the frequently-used term table 15, and finds a derived index term that completely matches a frequently appearing term or a derived index term that has a face and a morpheme information that completely matches a frequently appearing pair. Delete from the search index. Information output unit 17
Outputs the search index.

The index derivation device 10 uses the CP
With hardware such as U, memory, external storage device,
It is realized by software (program) having the procedure shown in the operation flowchart of FIG.

The operation of this apparatus will be described below with reference to FIG.

(Step s1) The data reading section 11 reads one data record (original index term) from the original database 1. Further, the read data record is sent to the morphological analysis unit 12.

(Step s2) The morpheme analysis unit 12 performs morpheme analysis on the extracted data record, divides it into constituent words, and adds morpheme information such as part of speech to each word.
Further, this result is sent to the stepwise record derivation unit 13.

(Step s3) Stepwise record derivation unit 1
In 3, the number of words of the data record currently being processed is counted from the sent morphological analysis result, and the value is set to n. Further, 0 is set to the number k of deleted words as an initial value.

(Step s4) Next, based on the sent data record and the morphological analysis result, a record (derivative index term) having the remaining information obtained by removing k records from the beginning is derived and stored in the memory. accumulate.

(Step s5) The number of deleted words k is n-1.
The processing is divided depending on whether or not it matches (one less than the total number of words). If they are not equal, the process proceeds to step s6, and if they are equal, the process proceeds to step s7.

(Step s6) Add 1 to the number of deleted words k (k = k + 1), and return to step s4.

(Step s7) Processing is divided depending on whether all data records in the original database 1 have been processed. When there is an unprocessed data record, the control is transferred to the data reading unit 11 and the process proceeds to step s8. When all the data records have been processed, the control is changed to the term totaling unit 14
Move to step s9.

(Step s8) The data reading unit 11 reads the next data record from the original database 1, and further reads the read data record from the morpheme analysis unit 1.
2 and returns to step s2.

[0026] (Step s9) the term counting unit 14, after the processing of the entire original database 1, a derivative record stored in the memory in step s4, pairs of characters Men'a Rui its textually and morphological information
In aggregate.

[0027] (Step s10) further term counting unit 14, at least typeface based on the counting result, two Kumia Rui textually frequency of occurrence, morphological information, a frequently-used word table 15 with three sets of frequency create.

(Step s11) Unnecessary record deleting unit 1
In step 6, the frequently-used term table 15 is referred to, the frequently-used term is deleted from the derivative records stored in the memory as an unnecessary record, and the control is transferred to the information output unit 17. Information output unit 17
Then, the derived record remaining in the memory is output to the search index 2.

By the above processing, only the frequently-used terms that match the morpheme information are deleted from the search index, so that the necessary index can be left correctly. Further, by performing term aggregation on a term basis rather than on a word basis, even a term consisting of two or more words can be deleted as an unnecessary record if the term frequently appears in the derived record.

[Second Embodiment ] FIG. 3 shows a second embodiment of the index derivation device of the present invention. In the figure, the same components as those in FIG. 1 are represented by the same reference numerals. That is, 1 is an original database, 2 is a search index, 3 is a search target database, and 20 is an index derivation device.

The index derivation device 20 includes a data reading unit 11, a morpheme analysis unit 12, a term totaling unit 14, a frequent term table 15, an information output unit 17, and a staircase record derivation unit 2.
1, a derivation source information addition unit 22, an unnecessary record deletion unit 23, and an unnecessary record determination rule group 24.

The stepwise record derivation unit 21 extracts the derived index terms (stepwise records) starting from each word obtained by the morphological analysis unit 12 from the original index term and after each starting word. At the same time as creating, the derivation source information adding unit 22 is called, derivation source information indicating from which original index term the derivation index term is derived is attached to each derivation index term, and further derivation index term and its derivation source. Store information and in memory.

The derivation source information adding section 22 adds derivation source information to the derivation index term.

The unnecessary record deletion unit 23 acquires a word located in front of a term having a high frequency of occurrence (forward position word) from the original index term obtained from the derivation source information, and a frequent term table. 15 and unnecessary record determination rule group 24, it is determined whether or not the corresponding derived index term should be deleted from the search index according to the content of the morpheme information of the forward position word,
The unnecessary record determining unit 232 deletes the derived index term to be deleted from the search index.

The unnecessary record determination rule group 24 has a plurality of rules having morpheme information as a key and expressing deletion conditions by the content of the morpheme information of the forward position word of the derived index term.

The index derivation device 20 uses the CP
With hardware such as U, memory, external storage device,
It is realized by software (program) having the procedure shown in the operation flowcharts of FIGS.

The operation of this apparatus will be described below with reference to FIGS.

(Step s21) Data reading unit 11
Then, one data record is read from the original database 1. The morphological analysis unit 1 uses the read data record.
Send to 2.

(Step s22) The morpheme analysis unit 12 performs morpheme analysis on the extracted data record, divides it into constituent words, and adds morpheme information such as part of speech to each word. Further, this result is sent to the stepwise record derivation unit 21.

(Step s23) The stepwise record derivation unit 21 counts the number of words in the data record currently being processed from the sent morphological analysis result and sets the value to n. Further, 0 is set to the number k of deleted words as an initial value.

(Step s24) Further, the stepwise record derivation unit 21 derives a record having the remaining information after removing k records from the beginning, based on the sent data record and the morphological analysis result. In addition, the staircase record derivation unit 21 calls the derivation source information addition unit 22,
Derivation source information of the derived record is added to the derived record. Further, the stepwise record derivation unit 21 stores the derived record and its derivation source information in the memory.

(Step s25) The number of deleted words k is n-1.
The processing is divided depending on whether or not it matches (one less than the total number of words). If they are not equal, the process proceeds to step s26, and if they are equal, the process proceeds to step s27.

(Step s26) 1 is added to the number k of deleted words (k = k + 1), and the process returns to step s24.

(Step s27) The processing is divided depending on whether all the data records of the original database 1 have been processed.
If there is an unprocessed data record, control is transferred to the data reading section 11 and step s28, and if all data records have been processed, control is transferred to the term totaling section 14 and step s29 is proceeded to.

(Step s28) Data reading unit 11
Then, the next data record is read from the original database 1, and the read data record is read by the morphological analysis unit 1
2 and returns to step s22.

(Step s29) In the term totaling unit 14,
After the processing of the entire original database 1, the derived records accumulated in the memory are aggregated in step s24 as the character face or a pair of the character face and morpheme information.

(Step s30) Further, the term totaling unit 14
Then, based on the tabulation result, at least the character and the appearance frequency of 2
A frequently-used term table 15 having three groups of three groups of character or face, morpheme information, and appearance frequency is created.

(Step s31) Unnecessary record deleting unit 2
3, the frequent term table 15 and unnecessary record determination rule group 2
4, the frequently-used term is deleted from the derivative record stored in the memory as an unnecessary record, and the control is transferred to the information output unit 17. The information output unit 17 outputs the derived record remaining in the memory to the search index 2.

FIG. 5 is a detailed operation flowchart of step s31 (unnecessary record deletion processing) in FIG. 4, and the operation will be described below according to this.

(Step s41) Unnecessary record deleting unit 2
In the forward position word acquisition unit 231 constituting No. 3, first, one frequently-used term is read from the frequently-used term table 15.

(Step s42) Next, the forward position word acquisition unit 231 searches the derived record stored in the memory by the stepwise record derivation unit 21 (step s24) using the face of the frequently-used word as a key.

(Step s43) The processing is divided depending on whether or not there is a derived record that matches the face of the frequently-used term. When it exists, it progresses to step s44, and when it does not exist, it returns to step s41 in order to process the next frequent term.

(Step s44) All the derived records that match the characters of the frequently used terms are read together with their derivation source information.

(Step s45) One of the derived records read in step s44 is processed.

(Step s46) Forward position word acquisition unit 2
At 31, the morpheme information on the forward-positioned word is acquired from the derivation source information of the derivation record as the processing target, and the control is transferred to the unnecessary record determination unit 232.

(Step s47) Unnecessary record determining unit 2
At 32, the unnecessary record determination rule group 24 is searched by the morpheme information of the frequently-used term currently being processed.

(Step s48) Processing is divided depending on whether or not there is a matching unnecessary record determination rule. If it exists, the process proceeds to step s49, and if it does not exist, the process proceeds to step s51 to process the next derived record.
Proceed to.

(Step s49) Whether or not the morpheme information on the forward-positioned word obtained from the derivation source information of the derivation record currently being processed matches any of the unnecessary record determination rules obtained at step s47. That is, the processing is divided depending on whether or not the record is a record to be deleted. If the records match, that is, the record should be deleted, the process proceeds to step s50. If the records do not match, that is, the record should not be deleted, the process proceeds to step s51.

(Step 50) Unnecessary record determination unit 23
In 2, the derived record that matches the unnecessary record determination rule group 24 is deleted from the memory.

(Step s51) The processing is divided depending on whether all the derived records have been processed. If so, the process proceeds to step s52, and if not, the control is transferred to the forward position word acquisition unit 231 to process the next derived record, and the process returns to step s45.

(Step s52) Processing is divided depending on whether or not all the frequently-used terms have been processed. If so, control is transferred to the information output unit 17, and the process proceeds to step s53. If not, control is transferred to the forward position word acquisition unit 231 to process the next frequently-used term, and step s41. Return to.

(Step s53) In the information output section 17,
The derived record remaining in the memory by the above processing is output to the search index 2.

By the above processing, it is determined whether or not the frequently-used term appears in the original data record after what word (description of the unnecessary record determination rule group 24) to delete it from the search index. , You can leave exactly what you need as an index. Further, by performing term aggregation on a term basis rather than a word basis, even a term composed of two or more words can be deleted as an unnecessary record if the term frequently appears in the derived record.

[Specific Example 1 ] Next , the operation of the apparatus according to the first embodiment will be described with a specific example. FIG. 6 shows the actual processing, and FIG. 7 shows an example of the frequently-used term table 15.

The data reading section 11 reads one data record from the original database 1. Here, it is assumed that the "snack center" is loaded. The data reading unit 11 sends the read data record "snack center" to the morphological analysis unit 12 (step s1).

The morpheme analysis unit 12 performs morpheme analysis on the extracted data record "snack center", divides it into constituent words, and adds morpheme information such as part of speech to each word (morpheme analysis in FIG. 6). Further, this result is sent to the stepwise record derivation unit 13 (step s2).

The stepwise record derivation unit 13 counts the number of words in the data record currently being processed from the sent morphological analysis result and sets the value to n. Since "snack center" is composed of two words, n = 2. Further, the number of deleted words k is set to 0 as an initial value (step s3).

Next, based on the sent data record "snack center" and the morphological analysis result, the remaining information after removing k = 0 from the beginning, that is, "snack center (crown name, proper noun)" A record having "" is derived and stored in the memory (step s4).

Since the number of deleted words k = 0 and n = 2, n-1 = 1, and k does not match n-1 (step s5). Therefore, 1 is added to the deleted word number k. K = 1
(Step s6).

Next, based on the sent data record "snack center" and the morphological analysis result, the remaining information obtained by removing k = 1 from the beginning, that is, a record having "center (proper noun)" Is derived and stored in the memory (step s4).

At this time, the number of deleted words k = 1 and n =
Since n is 2, n-1 = 1, and k matches n-1 (step s5). Therefore, the processing of this data record "snack center" is completed.

By the processing up to this point, "snack center (crown name, proper noun)" is obtained as the stepwise record derivation result.
And "center (proper noun)" are stored in the memory together with morpheme information (step-like record derivation in FIG. 6).

The processing is divided depending on whether all the data records in the original database 1 have been processed. Here, it is assumed that there is an unprocessed data record (step s7).

The data reading unit 11 reads the next one data record from the original database 1 (step s
8). Here, it is assumed that the "Kanto Incombustible Building Center" is loaded. In the same manner as above (steps s2 to s
8), as a result of stepwise record derivation, "Kanto Incombustible Building Center (place name, common noun, sahen noun, suffix)",
Four types of "non-combustible building center (common noun, sahen noun, suffix)", "building center (sahennoun, suffix)", "center (suffix)" are stored in the memory together with morpheme information.

Up to this point, assuming that all data records in the original database 1 have been processed, the control is performed by the term totaling unit 1
The process moves to 4 (step s7).

In the term totaling unit 14, the original database 1
After the entire processing, the derived records stored in the memory in step s4, its shape Men'a Rui aggregate with corresponding) to the pair of the character faces and morpheme information (step s9). In addition the term totaling unit 14, counting result on the basis of at least typeface, two Kumia Rui frequency textually, morpheme information, to create a frequently-used word table 15 with three sets of frequency (step s10) . Here, it is assumed that the contents of the frequently-used term table 15 shown in FIG. 7 are obtained.

At this time, assuming that the frequency of 3000 or more is a frequent term, the unnecessary record deleting unit 16 refers to the frequent term table 15 of FIG. 7 to refer to the frequent terms "center (suffix)", "building center (sa "Noun, suffix)", "company (crown name)", "corporation (crown name)" are deleted from the derivative records stored in the memory as unnecessary records (step s11).

Here, the data record "snack center" has the same character face as "center (proper noun)", but the part of speech that is the morpheme information is "proper noun", and the unnecessary record "center" (Suffix) ”does not match and is not deleted. Therefore, for the data record "snack center", the two records "snack center" and "center" are directly derived records (unnecessary record deletion in FIG. 6).

As for the data record "Kanto Incombustible Building Center", two records "building center (sahenon noun, suffix)" and "center (suffix)" are
The unnecessary records "Architectural Center (Sahennoun, Suffix)" and "Center (Suffix)" are deleted because they match the part of speech that is the morpheme information, and as a result, "Kanto Incombustible Building Center" Fire-retardant building center "
And two records of "Flameproof Building Center" will be derived records.

Finally, the information output unit 17 outputs the derived records "snack center", "center", "Kanto incombustible building center", and "incombustible building center" remaining in the memory to the search index 2 (step s1).
1).

As is clear from the above description, according to the present device, the suffixes in "Center" and "Kanto Incombustible Building Center" which are the proper nouns even if the character face is the same as "Center" such as "Snack Center" It is possible to distinguish it from the "center" so that what is necessary as an index can be left correctly. In addition, by performing term aggregation on a term-by-term basis rather than on a word-by-word basis, even if the term consists of two words, such as "Building Center", if that term occurs frequently in the derived records, that term is deleted as an unnecessary record. be able to.

[Specific Example 2 ] Next , the operation of the apparatus according to the second embodiment will be described with a specific example. FIG. 8 shows the actual processing, and FIG. 9 shows an example of the unnecessary record determination rule group 24. For the frequently-used term table 15, the example of FIG. 7 is used as it is. Moreover, in the following description, the description of the same operation as that of the first specific example will be omitted.

In the stepwise record derivation unit 21, the concrete example 1
After the same operation as the above, for the data record "Kanto Incombustible Building Center", "Kanto Incombustible Building Center (place name, common noun, sahen noun, suffix)", "Incombustible Building Center (common noun, sahenun, suffix) Data "," Architectural Center Co., Ltd. ", and four records" Architectural Center (Sa noun, suffix) "and" Center (proper noun) "
In contrast, three records of "Architecture Center Co., Ltd. (crow name, sahen noun, suffix)", "Architectural Center (sahenoh noun, suffix)", "center (suffix)",
Each is stored as a derivative record in the memory together with the morpheme information (morpheme analysis / step-like record derivative in FIG. 8).

At this time, pointers indicating the derivation source information of each derivation record are also accumulated at the same time (indicated by an arrow in the stepwise record derivation result in FIG. 8) (step s24).
Further, the frequently-used term table 15 is obtained as shown in FIG. 7 in the same manner as in the concrete example 1 (steps s29, s30).

Next, unnecessary record deletion processing (step s
31) will be described in detail.

The forward position word acquisition section 231 constituting the unnecessary record deletion section 23 first reads one frequently-used term from the frequently-used term table 15. Here, assuming that the frequency of 3000 or more is a frequently-used term, the frequently-used term table 15 is as shown in FIG. 7, and thus "center (suffix)" is read (step s41).

Next, the forward position word acquisition unit 231 searches for the derived record accumulated in the stepwise record derivation unit 21 (step s24) using the character face "center" of the frequently-used term as a key (step s42). From Figure 8, "Center (suffix)" derived from "Architectural Center Co., Ltd."
A "center (suffix)" derived from "Kanto Incombustible Building Center" is obtained (steps s43, s44).

First, the "center (suffix)" derived from the "Architecture Center Co., Ltd." is targeted for processing (step s45).

The forward position word acquisition unit 231 acquires the morpheme information “architecture (sahen noun)” related to the forward position word from the derivation source information of the derived record “center (suffix)” which is the processing target, and does not require control. Record determination unit 232
(Step s46).

The unnecessary record determination unit 232 searches the unnecessary record determination rule group 24 with the part of speech (suffix) which is the morpheme information of the frequently-used term "center (suffix)" currently being processed (step s47). Since the unnecessary record determination rule group 24 is as shown in FIG. 9, the unnecessary record determination rule (suffix, part of speech is not "crown name") is obtained (step s47).

At step s47, the part of speech (sahen noun) which is the morpheme information on the forward position word "architecture (sahen noun)" obtained from the derivation source information of the derived record "center (suffix)" currently processed It coincides with the obtained unnecessary record determination rule (suffix, part of speech is not "crown name") (steps s48, s49). Therefore, the unnecessary record determination unit 232 deletes the derived record “center (suffix)” derived from “Architecture Center Co., Ltd.” that matches the unnecessary record determination rule from the memory (step s50).

Similarly, by processing the "center (suffix)" derived from "Kanto Incombustible Building Center", this derived record is also deleted from the memory.

Up to this point, it is assumed that all derived records relating to the frequently-used term "center (suffix)" have been processed (step s51). Further, since there is an unprocessed frequently-used term, the control is moved to the forward position word acquisition unit 231, and the processing is moved to (steps s52, s41).

Since the frequently-used term table 15 is as shown in FIG. 7,
"Architectural center (sahen noun, suffix)" is read (step s41).

Next, the forward position word acquisition unit 231 searches for the derived record accumulated in the stepwise record derivation unit 21 (step s24) with the character face "building center" of the frequently used term as a key (step s42). From FIG. 8, the "Architectural Center (Sahen noun, suffix)" derived from the "Architectural Center Co., Ltd." and the "Architectural Center (Sahen noun, suffix)" derived from the "Kanto Incombustible Architecture Center" are obtained (step s43, s44).

First, "building center (sahenon noun, suffix)" derived from "building center corporation" is processed (step s45).

The forward position word acquisition unit 231 acquires the morpheme information "corporation (corporation name)" related to the forward position word from the derivation source information of the derived record "building center (Sahen noun, suffix)" which is the processing target. Then, the control is transferred to the unnecessary record determining unit 232 (step s46).

In the unnecessary record determination unit 232, the unnecessary record determination rule group 24 is defined by the part of speech (sa variant noun) which is the morpheme information of the word located at the beginning of the frequently-used term “building center (sa variant noun, suffix)” currently being processed. Is searched (step s47). Since the unnecessary record determination rule group 24 is as shown in FIG. 9, the unnecessary record determination rule (Sahen noun, part-of-speech is not "crown name") is obtained (step s47).

Part-of-speech (title) which is the morpheme information about the forward position word "corporation" obtained from the derivation source information of the derivation record "building center" which is the current processing target.
Does not match the unnecessary record determination rule (sahen noun, part of speech is not "crown name") obtained in step s47.
Therefore, the derived record is not deleted from the memory (step s49).

Next, the processing of the derived record "Architectural Center (Sahen noun, suffix)" derived from "Architectural Center Co., Ltd." is performed. In the same way as above, as an unnecessary record determination rule (sahen noun, part of speech is not "crown name")
Is obtained (step s47).

The part-of-speech (ordinary noun), which is the morpheme information on the forward-position word "incombustible" obtained from the derivation source information of the derivation record "building center (Sahen noun, suffix)" currently being processed, is the step s47. It matches the obtained unnecessary record determination rule (sahen noun, part of speech is not "crown name") (steps s48, s49). Therefore,
The unnecessary record determination unit 232 deletes the derived record “building center (Sahen noun, suffix)” derived from “Kanto Incombustible Building Center” that matches the unnecessary record determination rule from the memory (step s50).

Up to this point, it is assumed that all the derived records relating to the frequently-used term "building center (Sahen noun, suffix)" have been processed (step s51). Further, since there is an unprocessed frequently-used term, the control is moved to the forward position word acquisition unit 231, and the processing is moved to (steps s52, s41).

Next, “Company (Colour name)” is searched from the frequently-used term table 15, but this frequently-used term does not exist in the derived record. Similarly, "stock corporation (crown name)" does not exist in the derived record.

Assuming that all the frequently-used terms have been processed by the above processing, control is transferred to the information output unit 17 (step s5).
2).

Finally, the information output unit 17 retrieves the derived records "Architectural Center Co., Ltd.", "Architectural Center", "Kanto Incombustible Architecture Center", and "Incombustible Architecture Center" that remain in the memory after the above processing. It outputs to the index 2 for (step s53).

As is clear from the above description, if this device is used, the morpheme information of the forward position word is (crown name) even if the character face is the same as "Architectural Center" such as "Architectural Center Co., Ltd." It is possible to distinguish between the "Architectural Center" and the "Architectural Center" derived from the "Kanto Fire-retardant Building Center", and it is possible to correctly leave the necessary index.

[0107]

As described above, according to the present invention,
(1) Since the frequency is aggregated not by word but by term, even a frequent term composed of a plurality of words can be deleted from the search index if the frequency is high,
(2) When deleting the index of high-frequency words, only those that match even the morpheme information or those that comply with the unnecessary record determination rule are deleted, so homonyms (the same word is used with a different meaning) In this case, it is possible to decide whether or not to delete, and it is possible to obtain the effect that there is no fear of deleting a necessary index.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a first embodiment of an index derivation device of the present invention.

2 is an operation flowchart of the apparatus of FIG.

FIG. 3 is a configuration diagram showing a second embodiment of an index derivation device of the present invention.

4 is an operation flowchart of the apparatus of FIG.

FIG. 5 is a detailed operation flowchart of unnecessary record deletion processing in FIG.

FIG. 6 is a diagram showing a state of actual processing according to the first embodiment.

FIG. 7 is a diagram showing an example of a frequently-used term table.

FIG. 8 is a diagram showing a state of actual processing in the second embodiment.

FIG. 9 is a diagram showing an example of an unnecessary record determination rule group.

[Explanation of symbols]

1 ... Original database, 2 ... Search index, 3 ... Search target database, 10, 20 ... Index derivation device, 11 ... Data reading unit, 12 ... Morphological analysis unit, 1
3, 21 ... Stepwise record derivation part, 14 ... Term aggregation part,
15 ... Frequent term table, 16, 23 ... Unnecessary record deletion section,
Reference numeral 17 ... Information output unit, 22 ... Derived information addition unit, 231, ... Forward position word acquisition unit, 232 ... Unnecessary record determination unit, 24
... Unnecessary record determination rule group.

─────────────────────────────────────────────────── ─── Continuation of front page (56) References JP-A-7-78182 (JP, A) JP-A-6-309366 (JP, A) Tatsuo Kamio, automatic keyword extraction in newspaper article database, information management, Japan, July 1, 1989, Vol. 32, No. 4, 283 to 293 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/30 JISST file (JOIS)

Claims (6)

(57) [Claims] 1. A morphological analysis means for performing a morphological analysis for dividing a term (original index term) assigned as a search index to each record of an original database into word units, and each word obtained as a result of the analysis. A stepwise record derivation means for creating a starting term (derivative index term) by extracting each starting word and subsequent words from the original index term, and deriving at least one of the created derivation index terms. In the index derivation device in the database creation device that accumulates records having index terms as a new search index and creates the search target database, the derived index terms are created by the stepwise record derivation means.
When creating the index
A group that gives derivation origin information that indicates whether it was derived from the term
Origin information providing means, term aggregation means for checking the appearance frequency of the derived index terms in the entire database, and words located in front of words with high appearance frequency (forward position
Word) from the original index term obtained from the derivation origin information
Front position word acquisition means to be acquired and shape of the front position word
Derived index term applicable depending on the content of the feature information
Decide whether to delete from the search index
The derived index terms to be deleted
An index derivation device comprising: an unnecessary record deletion unit configured to delete an unnecessary record from a dex . 2. A derived index term to be deleted with reference to an unnecessary record determination rule group, which has morpheme information as a key, and which represents a deletion condition according to the content of the morpheme information of the forward position word of the derived index term. index derivation apparatus according to claim 1, comprising the unnecessary record determining means for determining. 3. A term starting from each word obtained as a result of the morphological analysis in which a term (original index term) assigned to each record in the original database as a search index is divided into word units. (Derived index term) is created by extracting each starting word and subsequent words from the original index term, and at least one derived index term of the created derived index terms is newly searched. In the index derivation method in the database creation method for accumulating the records having as the index for the search and creating the search target database, when creating the derived index term , the derived index
Indicates from which original index term the term was derived
Derivation source information is added, the frequency of occurrence of the derived index term in the entire database is checked, and words located in front of terms with high frequency of occurrence (forward position single
Word) from the original index term obtained from the derivation origin information
Acquired, and a group corresponding to the content of the morpheme information of the forward position word
Delete raw index terms from the search index
An index derivation method characterized by determining whether or not to do so, and deleting the derived index term to be deleted from the search index. 4. A derived index term to be deleted with reference to an unnecessary record determination rule group, which has morpheme information as a key and which represents a deletion condition according to the content of the morpheme information of the preceding position word of the derived index term. 4. The index derivation method according to claim 3, wherein 5. A morphological analysis in which a term (original index term) assigned as a search index to each record of an original database is divided into word units, and a term starting from each word obtained as a result of the analysis (Derived index term) is created by extracting each starting word and subsequent words from the original index term, and at least one derived index term of the created derived index terms is newly searched. a computer-readable medium recording the index derivation program in database creation to create the search database storing the record with the use index, when the index derivation program that is read by the computer, this computer, derived When you create an index term, the derivative production index
Indicates from which original index term the term was derived
Derivation source information is added, the frequency of occurrence of the derived index term in the entire database is checked, and words located in front of terms with high frequency of occurrence (forward position single
Word) from the original index term obtained from the derivation origin information
Acquired, and a group corresponding to the content of the morpheme information of the forward position word
Delete raw index terms from the search index
A computer-readable medium having an index derivation program recorded thereon , which determines whether or not to do so, and executes an operation of deleting a derived index term to be deleted from the search index. 6. A derived index term to be deleted with reference to an unnecessary record determination rule group, which has morpheme information as a key and which expresses a deletion condition according to the content of the morpheme information of the forward position word of the derived index term. claim, characterized in that to perform the operation of determining the 5
A computer-readable medium on which the described index-derived program is recorded.