JP3475009B2 - Information retrieval device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information retrieval system such as a document filing system for retrieving information corresponding to a designated key from a huge amount of information. With the development of information processing devices in recent years, instead of storing a huge amount of various kinds of information as documents, a large-scale document filing system was constructed, and a huge amount of documents were converted to digital information in a database. It is designed to accumulate. In such a database, an index table created in advance corresponding to a keyword or the like is also large-scaled, and in particular, an index table compatible with full-text search is very large. Therefore, the seek time required when searching the index table is not negligible, and the corresponding data part is efficiently searched from the index table in order to reduce the processing time of the entire search process. Technology is needed.

[0002]

2. Description of the Related Art FIG. 7 shows a configuration example of a conventional information retrieval apparatus. In FIG. 7, the index creation unit 410 creates an index table 402 corresponding to various search methods for each document registered in the document database 401,
The configuration is provided for the processing of the search processing unit 403.

The index table 402 has a data area 404 for storing a data part corresponding to each key item.
And an index section 405 that stores pointer information indicating the storage location of the corresponding data section corresponding to each key item. Therefore, the search processing unit 403
By reading the corresponding data part from the data area 404 based on the pointer information of the index part 405, the data part corresponding to the key item designated as the search key can be obtained.

For example, when creating an index table for full-text search, the key detecting section 41 of the index creating section 410 is used.
1 sequentially detects key items from the beginning of the input text data, and in accordance with the detection result, the registration processing unit 412 causes the index unit 405 and the data area 4 to operate.
Registration processing to 04 is being performed. This key detection unit 411
As a method of detecting key items, there are N methods in addition to the method of detecting a word (free term) from input text based on a dictionary prepared in advance, the method of detecting a keyword specified by a user, and the like. There is a method called N-gram that mechanically extracts consecutive characters whose upper limit is.

Further, in response to the input of a key item, the registration processing unit 412 first refers to the index unit 405, and if the corresponding key item is registered, inputs it to the data unit indicated by the pointer information. The document number indicating the text may be added. On the other hand, when a key item not registered in the index unit 405 is detected, the registration processing unit 412
Registers the detected key item in the index section 405 as a new key item, and
A new data section may be created in to store the document number of the input text.

In this case, the data part consisting of the document number including the key item is arranged in the data area 4 in the order of appearance of the key item.
It will be placed at 04. Further, in this way, each key item and data part are stored in the index table 402.
In some cases, the index unit 405 and the data area 404 may be rearranged in the order of the dictionary after the registration.

For example, in an index table corresponding to a document database in which English texts are stored, considering a case where similar key items are consecutively designated, such as words having the same stem but different endings, In the data area 404 of the index table 402, the data part of each key item is arranged in dictionary order.

[0008]

As described above, in the conventional information retrieving apparatus, regardless of the possibility of being designated as a retrieval key, the data parts are arranged in the data area in the order of appearance of the key items or in the dictionary order. It is located at 404.

By the way, when the number of documents registered in the database is enormous, the document number for identifying each document is very large, so the data length of the data part corresponding to each key item is It's getting longer. Particularly, in the data section corresponding to a key item having a high frequency of appearance, a large number of document numbers including the key item are listed, so that the data length thereof is extremely long.

Therefore, particularly in the index table for full-text search, the size of the data area 404 is remarkably increased, and when the data portion corresponding to the specified search key is read from this data area 404, the search is performed. Since the range is extremely wide, it took a long time to perform seek processing by the disk device.

In particular, when the data parts corresponding to the key items are arranged in the order of appearance in the document, the data parts to be searched are widely distributed in the entire data area 404, so that the search range for the data parts is limited. It becomes bigger and the average seek time becomes longer. In this case, the search key specified by the user has nothing to do with the order of appearance at the time of registration, so the seek range for reading the corresponding data part is
Since the entire data area 404 is used, the processing capacity of the information retrieval apparatus is significantly reduced as the data area 404 is enlarged.

On the other hand, when the data parts corresponding to the key items are arranged in the dictionary order, the data parts corresponding to similar key items can be expected to be arranged close to each other in the data area 404. Therefore, as long as similar words are continuously specified as the search key, the seek range can be limited to a part of the data area 404, and the seek time becomes short. However, if a completely different search key is specified, the seek area corresponding to the capacity of the data area 404 is necessary because the entire data area 404 becomes the seek range, as in the case of the arrangement in the appearance order described above. I will end up.

It is an object of the present invention to provide an information search device capable of limiting the search range of the data part corresponding to the search key.

[0014]

FIG. 1 is a block diagram showing the principle of the information retrieval apparatus according to the first and second aspects.

The invention according to claim 1 may serve as a search key in a full-text search system that searches for a document including the search key from all the documents stored in the document database 101 in response to the input of the search key. A word detection unit 1 that detects all documents including each key item from the document database 101 by using all words as key items, and outputs information specifying the detected document as document information corresponding to each key item.
11, a document information storage unit 112 for holding document information corresponding to each key item, based on the document information stored in correspondence with each key item, total number of documents containing each key field
Counting means 113 for counting and outputting as an index corresponding to the key item, and classification means 114 for classifying each key item into a plurality of groups based on the index corresponding to each key item,
The document information holding unit 112 is based on the obtained classification result.
The document information stored in the document information storage unit 112 stores the document information stored in the
It is characterized in that it is provided with an arranging means 115 arranged in a storage medium, and a searching means 116 for searching and outputting corresponding document information from the document information holding means 112 in response to an input of a search key.

According to the first aspect of the invention, the document database 1 is operated by the operation of the word detecting means 111 and the counting means 113.
Document in which the key item appears in all documents stored in 01
The number of Ru can be obtained as a statistical indicator for each key field. Depending on the index thus obtained,
For example, by the classifying means 114 classifying each key item ,
For example, depending on the number of documents in which the key item appears,
Can be classified into groups. The arranging means 115 stores the document information holding means 112 in accordance with this classification result .
Pair each document information on a storage medium such as a magnetic disk.
By respond to key fields are arranged for each group belonging, serial document information document the number of occurrences corresponding to the key item is comparable has stored document information holding means 112
The distribution range in the storage medium can be limited.

Here, a key item that is included in almost all documents or a very rare key item is rarely designated as a search key, and normally, it appears that the appearance frequency is medium. The key item is specified as the search key. Therefore, as described above, by controlling the distribution range of the document information according to the appearance frequency, it is possible to form the area in which the key items having the medium appearance frequency are concentrated and arranged. It is possible to limit the range in which the document information holding unit 112 searches for the target document information.

According to a second aspect of the present invention, in the full-text search system for searching a document including the search key from all the documents stored in the document database 101 in response to the input of the search key, each of the documents stored in the document database 101 is searched. Character string extraction means 1 for extracting a continuous character string by N-gram as a key item from a document and outputting information for identifying the extracted document as document information corresponding to each key item.
21, the document information holding unit 112 that holds the document information corresponding to each key item, and the characteristics of the combination of the character types that make up each of the key items obtained by the character string extraction unit 121 are determined, and the determined features are determined. Information for each key item
A discriminating unit 122 that outputs as a corresponding index , a classifying unit 114 that classifies each key item into a plurality of groups based on the index corresponding to each key item, and a document information holding unit based on the obtained classification result. Each sentence held in 112
Statement written information, for each group the corresponding key item belongs
Arrangement means 115 arranged in a storage medium provided in the written information holding means 112, and a search means for searching and outputting corresponding document information from the document information holding means 112 in response to an input of a search key. And 116.

According to the second aspect of the invention, the character string extracting means 121
For each key item extracted by, the feature extraction unit 122 extracts the feature of the character type configuration ,
Get the characteristics of character type composition as an index corresponding to each key item.
You can Based on this index, the classification means 114
Classifies the key items so that the key items
It can be divided into multiple groups according to the characteristics of the combination.
You can By providing this classification result to the processing of the arranging means 115, for example, the characteristics of the combination of character types can be determined.
Document information corresponding to similar key items is distributed in the storage medium storing the document information holding unit 112.
You can control the range you are in.

In this case, if the tendency of the combination of character types forming the search key input by the user is known,
By centrally arranging the key items having the corresponding characteristics in the document information holding means 112, the search means 116
The range in which the document information holding unit 112 searches for the target document information can be limited. FIG. 2 is a block diagram showing the principle of the information retrieval device according to the third aspect.

According to the invention of claim 3, in the kana-kanji conversion system in which the dictionary search means 102 searches and provides conversion candidates registered in advance in the dictionary 103 in response to the input of the reading of a word, a predetermined standard document is set. A reading extraction unit 131 that receives and extracts different readings included in the standard document as key items, and a frequency counting unit 132 that counts the number of times each key item appears in the standard document and outputs it as an index corresponding to the key item. , A classifying means 1 for classifying each key item into a plurality of groups based on an index corresponding to each key item
14 and the obtained classification result in the dictionary 103 .
A dictionary arranging unit 133 is provided for arranging the held conversion candidates in a storage medium storing the dictionary 103 for each group to which the corresponding key item belongs and providing the dictionary arranging unit 133 for the processing of the dictionary searching unit 102. .

According to the third aspect of the present invention, the frequency of occurrence of the reading of each word in the standard document obtained by the operations of the reading extraction unit 131 and the frequency counting unit 132 is used as an index to classify the classifier.
The operation of the column 114 causes each of the items corresponding to the key items.
Classify word readings into multiple groups according to frequency of occurrence
You can Based on the classification results obtained in this way
Then, the dictionary array means 133 stores the dictionary 103.
Storage media that belong to each group
Arrange the conversion candidates corresponding to the reading. This allows
The appearance frequency of reading in the standard document can be reflected in the conversion candidate array in the dictionary 103.

Here, if a document having a sufficiently large size as a standard dictionary is input to the reading extraction means 131, the frequency counting means 132 can obtain a counting result corresponding to the frequency of reading of each word in an arbitrary document. . Therefore, by the arrangement processing by the dictionary arrangement unit 133, the conversion candidates corresponding to the readings having a high appearance frequency can be arranged in the dictionary 102 in a concentrated manner.
The average range for searching the conversion candidates in the dictionary 103 can be reduced.

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings. FIG. 3 is a configuration diagram of a document filing system to which the information search device according to the first aspect of the invention is applied. In FIG. 3, a key detection unit 411 and a registration processing unit 412 correspond to the word detection unit 111 described in claim 1, and when a document to be searched is input to the document database 401 in the same manner as in the conventional case. Then, the index table 402 for full-text search may be created and used for the processing of the arrangement changing unit 210.

In the arrangement changing unit 210 shown in FIG. 3,
The frequency counting unit 211 is the counting unit 112 described in claim 1.
The number of document numbers included in the data section corresponding to each key item is counted by referring to the data area 404 corresponding to the document information holding unit 112, and the counting result is passed through the frequency table 212. It is configured to be provided to the processing of the classification processing unit 213.

Here, the frequency table 212 may be configured to store the number of documents including the key item as the appearance frequency of the corresponding key item, corresponding to the key number indicating each key item. Further, in FIG. 3, the classification processing unit 213, 請
It corresponds to the classifying means 114 described in claim 1 ,
According to the result of comparison between the appearance frequency stored in the frequency table 212 and a predetermined threshold value, the key items are classified into two groups, and each key item of the frequency table 212 is associated with a group indicating a group to which the key item belongs. It is configured to add a flag.

In FIG. 3, the sort processing unit 214 extracts the key items belonging to each group based on the group flag, rearranges them in the order of appearance frequency for each group, and supplies them to the processing of the data arranging unit 215. Has become. This data arrangement unit 215 is a sort processing unit 2
According to the sorting result received from 14, the data area 404 of the data section corresponding to the key item belonging to each group
The storage location in is changed.

The index updating unit 216 updates the contents of the index unit 405 in accordance with the placement result of the data placement unit 215, and reflects the placement result described above in the pointer information corresponding to each key item. . In this way, the data arranging unit 215 operates according to the sorting result by the sorting processing unit 214 to realize the function of the arraying unit 115, and the data unit corresponding to the key item that appears at the same frequency is provided in the data area. They can be arranged centrally in 404.

For example, the above-mentioned predetermined threshold is set to the total number of documents n.
When the classification processing unit 213 classifies into a group consisting of key items whose appearance frequency exceeds a predetermined threshold and a group consisting of key items below the threshold, the classification processing unit 213 registers them in the document database 401. The key items contained in the majority of the documents can be separated.

Here, the key items having a high appearance frequency included in most of the registered documents are, for example, particles and verbs in the case of free terms, and directives and connectives, and these are search keys. It is thought that it is seldom specified as. However, as described above, since the data part corresponding to the key item is a list of document numbers including the respective key items (words in this case), the data part corresponding to the key item having the higher frequency of occurrence is The data length is long, and the wide distribution of such a huge data portion in the data area 403 is a cause of expanding the seek range when searching for the data portion to be searched.

Therefore, by classifying the key items according to the appearance frequency as described above and arranging them separately in the data area 404, the search processing section 403 corresponding to the search means 116 searches the data section. It is possible to greatly reduce the seek range. In addition,
For the key items belonging to the group whose appearance frequency is less than or equal to the threshold value, the data arrangement unit 215 arranges the data parts corresponding to the key items in the data area 404 in descending order of the appearance frequency, so that as shown in FIG. In addition, the key items that appear in the document relatively frequently and the key items that appear significantly less frequently can be gently separated and placed in the data area 404.

Here, a word such as a general noun that is highly likely to be designated as a search key does not appear as frequently as a word having a remarkably high frequency of appearance such as the above-mentioned noun, but it appears more frequently than a proper noun or the like. Is high. Therefore, as described above, it is necessary to separate the key items having a very low appearance frequency from the key items having a significantly high appearance frequency, and arrange the key items appearing in the document with a moderate frequency in the data area 404 in a concentrated manner. Thereby, the seek range at the time of searching the data part can be further limited, and the effect of shortening the average seek time can be enhanced.

As a result, the time required for the full-text search processing can be greatly reduced, and the operability of the document filing system can be improved. Note that the classification processing unit 213 may be configured to operate the sign bit of the numerical data representing the appearance frequency in the frequency table 212 in accordance with the result of comparison between the appearance frequency and a predetermined threshold value.

For example, when the appearance frequency is equal to or higher than a predetermined threshold value, the sign bit of the corresponding numerical value data in the frequency table 212 may be operated to make the numerical value indicating the appearance frequency a negative number. In this case, the sorting unit 214 collectively sorts all the key items based on the frequencies stored in the frequency table 212, so that the sorting is performed after the grouping as described above. Similarly, it is possible to obtain a sorting result in which key items having extremely high or low frequency of occurrence are separated.

By the way, since N-gram, which is one of the methods for extracting key items from a document, is a method for extracting a character string up to N characters starting from each character in the document in a crushed manner. By using this, key items can be extracted regardless of the nature of the document. However, when this method is used, key items that are obviously not useful as a search key, such as one katakana character or a combination of katakana and alphanumeric characters, are also extracted, and the key items and corresponding data are also extracted. Each part is registered in the index table 402.

Next, focusing on the character type of the key item, a method of arranging in the data area will be described. Figure 5
FIG. 9 shows a block diagram of a document filing system to which the information retrieval apparatus of claim 2 is applied. In FIG. 5, the key detection unit 411 operates as a function of the character string extraction means 121 described in claim 2, and N-gra is used as a key item extraction method.
The key processing unit 412 detects a key item from each document in the document database 401 by using m, and the registration processing unit 412
The registration process is performed on the index table 402 for each key item in the order obtained by the above.

Further, in the arrangement changing unit 210 shown in FIG. 5, the feature extracting unit 221 extracts and obtains the feature of the combination of the character types constituting each key item registered in the index unit 405. The feature information indicating the feature is provided to the processing of the matching processing unit 222. The matching processing unit 222 matches the feature information corresponding to each key item with the extraction condition held in the condition holding unit 223, and, according to the matching result, a matching flag indicating whether or not the extraction condition is matched. Is added to correspond to the corresponding key item of the frequency table 212, and is provided for the processing of the sort processing unit 214.

As described above, the collation processing unit 222 operates based on the characteristic information obtained by the characteristic extraction unit 221, so that the discrimination means 122 and the classification means described in claim 2 are executed.
Each key item can be classified into two groups based on the characteristics of the character types that form each key item obtained by applying the N-gram by implementing the function of the step 114 .

Further, the sorting result is based on this classification result.
The data arranging unit 215 is subjected to processing via the
The arranging means 1 described in claim 2 by the operation of 6
It is possible to realize the 15 functions and arrange the data portion corresponding to each of the key items in the data area 404 in accordance with the characteristics of the character type forming each key item.

Here, for example, the condition holding unit 22 described above.
3 stores information indicating a combination of character types that are likely to be specified as a search key, such as a character string in which two kanji characters, two katakana characters, one kana character followed by one kana character, and the like are stored as extraction conditions. You can leave it.

In this case, the key items are classified into a group consisting of character strings and a group consisting of other character strings which are considered to be highly likely to be designated as a search key by the operation of each section as described above. However, the data parts corresponding to each can be arranged separately in the data area 404. As a result, as shown in FIG. 4B, the key items that meet the extraction condition can be concentrated in the head portion of the data area 404, so that the search processing unit 403 responds to the designation of the search key. It is possible to significantly reduce the range for searching the data part by shortening the average seek time.

Further, by performing the appearance frequency counting process by the frequency counting unit 211 in parallel with the above-mentioned determination process, the arrangement change according to the appearance of the key item can be applied at the same time. For example, if the classification processing unit 213 is configured to operate the sign of numerical data indicating the appearance frequency in the frequency table 212 according to the comparison result between the appearance frequency and the threshold value, the sorting processing unit 214 causes the frequency table 212 to operate.
The key items of the frequency table 212 may be sorted in descending order of the appearance frequency for each group indicated by the conformance flag of No. 1, and provided to the processing of the data placement unit 215.

In this case, the data arrangement unit 215
First, the data parts corresponding to the key items belonging to the group considered to be useful as the search key are arranged in the data area 403 in descending order of the frequency of appearance of the key item, and then the usefulness as the search key is low. Similarly, the data section corresponding to the key item belonging to the selected group has the data area 40
3 (see FIG. 4 (b)).

As a result, it is possible to further limit the search range of the data part in the search processing and shorten the average seek time. Further, the above arrangement method considering the statistical property of the key item may be applied to the dictionary for converting kana to kanji.

FIG. 6 shows the configuration of a kana-kanji conversion system to which the information retrieval device of claim 3 is applied. In FIG.
The word dictionary 231 corresponds to the dictionary 103 described in claim 3, and stores a data area 232 in which data parts storing conversion candidates corresponding to readings are arranged in dictionary order and a data part corresponding to each reading. Index part 2 showing the place
And 33.

The conversion processing unit 234 corresponds to the dictionary search means 102 described in claim 3, receives the reading input by the user through the input / output control unit 235, and then the word dictionary described above. The corresponding conversion candidate is searched from 231 and the obtained conversion candidate is returned. Further, in the counting processing unit 240 shown in FIG. 6, the word analysis unit 241 corresponds to the reading extraction unit 131, receives a standard document via the input / output control unit 235, and refers to the analysis dictionary 242. This standard document is analyzed to extract individual words, and the readings of the obtained words are sequentially sent to the frequency counting unit 243.

In response to this, the frequency counting section 243 operates as the frequency counting means 132 and increments the count value corresponding to the corresponding reading of the frequency table 244 to count the appearance frequency for each reading. Has become. Here, if a sufficiently long document is input as the standard document, the appearance frequency of the reading in a general document can be predicted based on the appearance frequency of each reading in the standard document.

Therefore, the frequency table 244 obtained as described above is subjected to the processing of the classification processing unit 213,
The sort processing unit 214 and the data arranging unit 215 operate according to the classification result by the classification processing unit 213, so that the data stored in the word dictionary 231 is stored according to the frequency of reading of each word in the document. A data part corresponding to the area 232 can be arranged.

In this way, the data part corresponding to the word that is expected to appear frequently in the document and the word that is expected to appear infrequently are separated, and the reading of frequently used words is supported. By centrally arranging the conversion candidates to be converted in the data area 232, the search range when searching for the kana-kanji conversion candidates can be limited and the average seek time can be shortened, so that the kana-kanji conversion processing can be speeded up. It is possible to plan.

[0050]

As described above, according to the first aspect of the invention, the data portion corresponding to each key item is arranged in accordance with the appearance frequency in the document, so that the full-text target search process is performed. In addition, since the search range for searching the corresponding data part can be limited, the average seek time can be shortened and the search processing can be speeded up.

According to the second aspect of the invention, by arranging the data part corresponding to each key item in accordance with the characteristics of the combination of the character types forming the extracted key item, the number of key items is N -When extracted by gram,
Since the search range for searching the corresponding data part can be limited during the full-text target search processing, it is possible to shorten the average seek time and speed up the search processing.

According to the invention of claim 3, the appearance frequency of each word in a general document is estimated based on the frequency of appearance of the word in the standard document, and the reading of each word is handled according to the estimation result. By arranging the data part consisting of the conversion candidates, the search range when searching for the conversion candidates can be limited, so that the average seek time can be shortened and the kana-kanji conversion processing can be speeded up. it can.

[Brief description of drawings]

FIG. 1 is a principle block diagram of an information search device according to claims 1 and 2. FIG.

FIG. 2 is a block diagram showing the principle of the information retrieval device of claim 3;

FIG. 3 is a configuration diagram of a document filing system to which the information search device of claim 1 is applied.

FIG. 4 is a diagram illustrating the arrangement of data parts in a data area.

FIG. 5 is a configuration diagram of a document filing system to which the information search device of claim 2 is applied.

FIG. 6 is a configuration diagram of a kana-kanji conversion system to which the information search device according to claim 3 is applied.

FIG. 7 is a diagram showing a configuration example of a conventional information search device.

[Explanation of symbols]

101, 401 Document database 102 Dictionary search means 103 Dictionary 111 Word detection means 112 Document information holding means 113 Counting means 114 Classification means 115 Arrangement means 116 Searching means 121 Character string extracting means 122 Discriminating means 131 Reading extraction means 132 Frequency counting means 133 Dictionary Arrangement unit 210 Arrangement changing unit 211 Frequency counting unit 212, 244 Frequency table 213 Classification processing unit 214 Sort processing unit 215 Data arrangement unit 216 Index updating unit 221 Feature extracting unit 222 Matching processing unit 223 Condition holding unit 231 Word dictionary 234 Conversion processing unit 235 Input / output control unit 240 Frequency counting unit 241 Word analysis unit 242 Analysis dictionary 243 Frequency counting unit 402 Index table 403 Search processing unit 404, 232 Data area 405, 233 Index unit 410 Index creation 411 key detection unit 412 registration processing unit

Continuation of the front page (56) References JP-A-6-68159 (JP, A) JP-A-8-194719 (JP, A) JP-A-4-205560 (JP, A) T. Kikuchi, T. Kikuchi, Attributes of constituent characters / Acceleration method of full text search by encoding including character position, IEICE Technical Report (DE90-24), December 14, 1990, Volume 90, No. 362, pages 1-7 Tadashi Kikuchi, Acceleration of Kana Character Search in Full-text Search Using String Matching, IPSJ Research Report (91-DBS-83), May 24, 1991, Vol. 91, No. 46, No. 46 Pages 1-10 Chuichi Kikuchi, A Method for High-speed Full-text Search for Japanese Documents, IPSJ Research Report (92-FI-25), May 12, 1992, Vol. 92, No. 32, 9-16 Page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/30 G06F 17/22

Claims (3)

(57) [Claims] 1. In a full-text search system for searching a document including the search key from all documents stored in a document database in response to the input of the search key, all words that may be a search key are used as keys. A word detecting unit that detects a document including the key items as items, and outputs information that specifies the detected document as document information corresponding to each key item; and a word detecting unit that corresponds to each key item. Based on the document information held corresponding to each key item, the number of documents including each key item is counted, and
Counting means for outputting as an index corresponding to the key item, classification means for classifying each key item into a plurality of groups based on the index corresponding to each key item, and based on the obtained classification result, to document information holding means
Each of the retained document information is stored in the document information retaining means for each group to which the corresponding key item belongs.
An information retrieving apparatus comprising: an arranging unit arranged in a storage medium; and a retrieving unit that retrieves and outputs corresponding document information from the document information holding unit in response to an input of a retrieval key. 2. A full-text search system for searching a document including the search key from all the documents stored in the document database in response to the input of the search key, wherein N- g
A character string extraction unit that extracts a continuous character string by ram as a key item and outputs information that specifies the extracted document as document information corresponding to each key item, and the document information corresponding to each key item. The characteristic of the combination of the document information holding unit that holds the key information and the character type that constitutes each of the key items obtained by the character string extraction unit is determined, and the determined feature is determined.
The information indicating the signature is output as an index corresponding to each key item.
A determination unit that, on the basis of the index corresponding to each key item, and classifying means for classifying said each key field into a plurality of groups, based on the classification results obtained, in the document information storage unit
Each of the retained document information is stored in the document information retaining means for each group to which the corresponding key item belongs.
An information retrieving apparatus comprising: an arranging unit arranged in a storage medium; and a retrieving unit that retrieves and outputs corresponding document information from the document information holding unit in response to an input of a retrieval key. 3. A kana-kanji conversion system in which a dictionary search means searches and provides conversion candidates registered in advance in a dictionary in response to a word reading input, and receives a predetermined standard document and includes it in the standard document. Reading extraction means for extracting different readings as key items, and frequency counting means for counting the frequency of occurrence of each key item in the standard document and outputting it as an index corresponding to the key item. , based on an index corresponding to each key item, the classification means each key item is classified into a plurality of groups, based on the obtained classification result, each conversion candidates held in the dictionary, corresponding An information retrieval device comprising: a dictionary arrangement unit arranged in a storage medium storing the dictionary for each group to which the key item belongs and used for processing of the dictionary retrieval unit. Place