JP2669601B2 - Information retrieval method and system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system and method for searching a large number of documents stored in a disk, for example, in a text file format, at high speed and with a desired ambiguity.

[0002]

2. Description of the Related Art There has been a demand for a high-speed search for a large number of documents written in a natural language and stored on a disk, such as newspaper articles, patent gazettes, and scientific and technical literature, and various search methods have been proposed. Have been. Such search methods are roughly classified as follows.

(A) Keyword search method In this method, indexing of individual documents and keywords representing the contents of the documents is performed in advance. At this time, as a method of determining a keyword, there are automatic keyword extraction by morphological analysis, manual keyword addition, and a method combining them. However, this method can only search for character strings indexed as keywords, and the accuracy of automatic keyword extraction by morphological analysis depends on the accuracy of word / grammar dictionaries. It has the drawback of requiring human work.

(B) Full-text search method without index This is a method of scanning all documents to be searched each time a character string to be searched is specified without using an index. Some systems use special hardware to speed up the search. However, a system using special hardware is costly and, in a client-server environment, there may be restrictions on the types of machines that can be used.

(C) Full-text search method using an index The present invention belongs to a full-text search method using an index. The following techniques are known for the purpose of speeding up full-text search by using an index.

In Japanese Patent Laid-Open No. 4-205560, a character string to be searched is divided into search units, which are units for searching, and an ascending code is assigned to each search unit. On the other hand, an attribute code indicating its logical division is given, a character position order code indicating the position in the search unit of the character string to be searched is given for each character, the search unit identification code and the character position It is disclosed that character position information including an order code and an attribute code is created, and the character position information is stored in an area for each character type to create a search file.

In Japanese Patent Laid-Open No. 4-215181, a character string to be searched is constructed in order to reduce the number of times the character string is matched for search processing and to enable high-speed matching in a general-purpose information processing device. It is disclosed to create a search file in which character set position information indicating the position of each character set in a character string is grouped for each character set type.

By the way, it is often necessary to search not only for a character string that completely matches the search character string but also for a document that includes a character string that partially matches the search character string. For example, when the user's memory of the search string is ambiguous, or the search string can appear with various modifications, and it is difficult to list all such modifications. There is.

A typical method of specifying a partial character string in the prior art is to use a regular expression. According to this, 0 or more repetitions of any character, 1 of any character
It is possible to specify more than once, a line end position, a line start position, an arbitrary character within a specific character code range, and the like.

Further, Japanese Patent Application Laid-Open No. 63-99830 discloses a system having a function of partially matching search character string data with search target character string data. Disclosed is to provide a table for accumulating data indicating whether the search character string data appears in any of the searched character string data.

Further, Japanese Patent Laid-Open No. 62-221027 discloses that when a character string cut out from the beginning of a partial object character string is not retrieved from the dictionary, the length of the next cut out character string is increased by one. By performing a forward search,
It is disclosed that the number of invalid searches is reduced to enable relatively fast and efficient word segmentation.

Further, Japanese Patent Laid-Open No. 4-326164 and Japanese Patent Laid-Open No. 5-174067 disclose self-correlation information for each property to be searched in a database search system, and the self-correlation information of the search key is stored. Disclosed is the provision of a search means for obtaining the degree of matching between the correlation information and the autocorrelation information of the search target 10 for each property, and outputting the property numbers in descending order of the degree of matching.

However, with these conventional character string search techniques, it is difficult to specify the degree of ambiguity of the character string to be searched, and the search result is undesired or unnatural for the user. It will contain many strings.

[0014]

SUMMARY OF THE INVENTION An object of the present invention is to provide a character string search technique capable of arbitrarily specifying the degree of ambiguity of a character string to be searched.

Another object of the present invention is to provide an index structure for realizing a character string search technique capable of arbitrarily specifying the degree of ambiguity of a character string to be searched.

Still another object of the present invention is to provide a character string search technique capable of searching for something close to human senses by fuzzy search.

[0017]

According to the present invention, a unique number (or symbol) is assigned to each document in order to perform a full-text search on a database composed of a plurality of documents.
The search file stores N consecutive characters in each document, a document number in which the N characters appear, and positional information in the document. Preferably, the index file is composed of two files, a character pattern file and a position information file. The character pattern file stores where the character pattern / separation pattern and the corresponding document number / in-document position number are located in the position information file. The location information file contains
The document number and the position number in the document are stored.

According to the present invention, there is proposed a method for quickly searching for a document including a character string whose character sequence is similar to a designated character string using such an index file.
In this method, a character string to be searched and a search precision (greater than 0 and less than or equal to 1) are specified, and a "similar character string" in which the "similarity degree" of the character string to be searched is equal to or greater than the specified search precision. It is possible to specify the position in the document of the document including "and similar character string".

Specifically, this is done by selecting a "character string" similar to the character string to be searched from the document, and checking how many characters are consecutively matched and how many extra characters are inserted in the middle. This is a process of digitizing the "similarity" from two perspectives.

At this time, when the "similarity degree" reaches the maximum value of 1, the character strings are completely matched, and if the character strings are completely matched, the "similarity degree" is always the same. Becomes 1. "Similarity" when a similar character string has extra characters that are not in the character string you want to search for, or when only part of the character string you want to search appears in a similar character string
Is less than one, but in accordance with the present invention, such a "similarity" value is reasonable and fairly consistent with human normal sensation.

Since the index file enables a high-speed search for any continuous N characters in a document, the N characters are sequentially compared with a character string to be searched using the index file. This makes it possible to detect at a high speed how many consecutive characters match and how many extra characters are interposed in the middle.

[0022]

Embodiments of the present invention will be described below with reference to the drawings.

A. Hardware Configuration Referring to FIG. 1, there is shown an overview of a system configuration for implementing the present invention. This configuration corresponds to the bus 101
In addition, the central processing unit (CP
U) 102, loads the program, and CPU 10
Main memory (RAM) 104, which provides a work area for
A keyboard 106 for inputting commands, character strings, and the like, and a hard disk 108 storing an operating system for controlling the central processing unit, a database file, a search engine, an index file, and the like.
With a normal configuration, a display device 110 for displaying a search result of a database and a mouse 112 for pointing an arbitrary position on the screen of the display device 110 and transmitting the position information to a central processing unit are connected. is there.

As an operating system, W
Windows (trademark of Microsoft), OS / 2 (I
BM trademark), X-WIN on AIX (IBM trademark)
GUI as standard, such as DOW system (trademark of MIT)
Although it is desirable to support a multi-window environment, the present invention relates to PC-DOS (trademark of IBM), MS-
It can be realized in a character-based environment such as DOS (registered trademark of Microsoft), and is not limited to a specific operating system environment.

FIG. 1 shows a system in a stand-alone environment. Generally, a database file requires a large-capacity disk device.
The present invention is realized as a client / server system, a database file and a search engine are arranged on a server machine, and the client machine is connected to the server machine via LAN by Ethernet, token ring, etc. -On the machine side, only the display control unit for viewing the search result may be arranged.

B. System Configuration Next, the system configuration of the present invention will be described with reference to the block diagram of FIG. It should be noted that the elements shown as individual blocks in FIG. 2 are individually or collectively stored in the hard disk 108 in FIG. 1 as a data file or a program file. .

The database 202 mainly stores a plurality of documents such as a newspaper article database and a patent publication database.
However, it should be noted that the scope of the present invention is not limited to searching a database composed of a plurality of documents, but can be applied to searching within a single document. At this time, the content of each individual document is stored in a searchable manner, for example, in a text file format. Further, a unique document number is given to each document. The preferred document number is an ascending sequential number starting from 1, but in the case of a patent publication database, an application number or a publication number can be used as a unique document number. Also,
Use "AB" instead of a sequential number to identify individual documents
Symbols such as "C", "&XYZ", etc. may be used, however, in general, representing such an identification symbol requires more bytes than numbers, so in practice a sequential number It is preferable to identify the document with.

Generally, it takes a long processing time to directly search a huge amount of contents such as newspaper articles or patent publications stored in the database 202, and therefore, it is necessary to search all newspaper articles stored in the database 202. The index file 204 is created in advance by the index creation / update module 206 for the content. In an embodiment to be described later of the present invention, the index file 204 thus created is composed of two files, a character pattern file and a position information file. The character pattern file stores where the character pattern / separation pattern and the corresponding document number / in-document position number are located in the position information file. The position information file stores the document number and the position number in the document.

The database 202 may manage individual documents as individual files, or may arrange all documents in a continuous single file in order. In short, the essential point is that A unique number is assigned to each document, and the contents of each document can be accessed with the unique number. In the former case, the database 202 manages a table that associates the unique number of each document with the actual file name that stores the document. In the latter case, the database 202 stores the unique number of each document. Will manage a table that correlates with the offset and document size in a single database file.

The search engine 208 searches the index file 204 by using the search character string provided by the search character input module 210 as an input, and the document number (there may be a plurality) of the document including the input search character string and its It has a function of returning a position (also a plurality of positions) where an input search character string appears in a document. The search character input module 210 is preferably composed of a dialog box in a multi-window environment, and is of a form in which a desired character to be searched is input by the keyboard 106 in the input box.

Further, according to a feature of the present invention, the search character input module 210 inputs the degree of similarity of the fuzzy search as a numerical value of 0 to 1 (a numerical value of 0 to 100 may be used on the basis of percentage). Make it possible. Therefore, the search character input module 210 displays a slider or a scroll bar having a handle indicating an arbitrary position between 0 and 1. The handle of the slider indicates, for example, 1 by default, and further, the handle is moved to the mouse 11
By dragging and moving in 2, it can be operated to point to another value.

The result display module 212 accesses the database 202 based on the document number which is the search result provided from the search engine 208 and the value of the position where the search character appears in the document, and the result is displayed in that position of the document. Corresponding rows are preferably displayed in a separate search result display window. If the search results do not fit on one screen of the window, a scroll bar will appear to allow the user to click through it to sequentially view the search results.

C. Structure of Index File and Creation Method In the present invention, a file is created in which all consecutive N characters, their positions in the document, and the division information in the document are indexed as documents. Here, the in-document division information is typically a sentence break such as “.”, “,” Or “first
It is a document break in a broad sense such as "chapter" or "summary".

C1. String Normalization The first process required to create an index file is string normalization, which is as follows.
That is, when the document to be searched is a Japanese text file, half-width characters and full-width characters may be mixed. Therefore, for example, a process of replacing a half-width character with a corresponding full-width character is performed.

C2. Extraction of character pattern information The next step to create an index file is N consecutive characters starting from that character for all characters in the normalized character string (hereinafter referred to as character pattern).
Is extracted and stored in the index file together with the document number and the position number within the document. However, N ≧ 1 and in the case of Japanese, N = 2 is appropriate.

The position number in the document is obtained by sequentially assigning a unique number in the document to all characters to be searched in the document. Then, the in-document position number of the first character of the character pattern is set as the in-document position number of the character pattern. If the number of characters at the end of the document is less than N including the following characters, a predetermined padding character such as X'00 'is padded to make a total of N characters.

In this embodiment, each individual document is divided into blocks in a manner that is meaningful at the time of retrieval, and the division information is stored in an index file. The division information is stored in the same format as the above-mentioned character pattern. That is, instead of the character pattern extracted from the normalized character string, a specially defined delimiter pattern is stored together with the document number of the document and the in-document position information of the character at the boundary of the block.

By defining several types of division patterns, it becomes possible to have several types of different division methods. However, the delimiter pattern must be defined so as not to overlap with the character pattern extracted from the normalized character string. In this embodiment, the 1-byte code is also converted into the 2-byte code by the normalization process. Therefore, when 2 bytes are regarded as 1 word, if the value of 1 word is 255 or less, It does not correspond to a normal character code. Therefore, an arbitrary word value of 0 to 255 can be individually assigned to a plurality of types of delimiter patterns.

The advantages of storing the division information in such a format similar to the character pattern are as follows. -Easy to create and update indexes. No special processing is required for the division information. -Does not significantly increase index capacity. For example, the increase in capacity is much smaller than in a format in which a block number to which a position number in a document belongs is added for each position number in a document.

C3. Specific example of the position number in the document For example, a document including the sentence "Today is fine weather. The microphone is currently being tested."
2 (FIG. 2). The position number in the document is given to each character of this sentence as follows.

[Table 1] Position number of the character in the document 1 2 3 4 5 6 7 8 9 10111213141516171819202122 Normalized character string Today is sunny. We are currently testing the microphone. Separation method 1 | | Separation method 2 |

It is assumed that the document number of the document is No. 1 and the number N of characters of the character pattern is 2.
Then, the document number and the position number in the document associated with each character pattern (length 2) are as follows.

[Table 2] Character pattern Document number Document position number -------------------------------------- ---- Today 1 1 is fine 1 2 fine 1 3 fine 1 4 heaven 1 5 1 7. 1 8 Separation pattern 1 1 8 Separation pattern 2 1 8 Only 1 9 1 1 10 Now 1 1 1, 1 1 2 and 1 1 3 Separation pattern 2 1 13 My 1 1 4 1 18 strikes 1 19 strikes 1 20 strikes. 1 21. 1 22 Separation pattern 1 1 2 2 Separation pattern 2 1 2 2

C4. Role of document division information Next, the utility value of document division information (delimiters) in a search will be described.

Retrieval Targeting Only Specific Blocks For example, when a document is composed of a title, an abstract, and a body, it is common to search for only a specific part such as only a title or only an abstract. It can be said that it is a request. About the end of the title, the end of the summary,
By storing the delimiter pattern and its position information,
Such a search can be realized.

Retrieval of Documents Strongly Related to a Plurality of Strings It can be said that it is a general desire to perform a search conscious of the strength of the relation between a plurality of character strings in the context. For example, rather than just strings being in the same document,
It is highly likely that the relationship is stronger in the same paragraph, and it is predicted that the relationship is stronger in the same sentence. By storing the delimiter pattern and its position information at the end of a paragraph or sentence, it becomes possible to search for documents in which multiple character strings are in the same block,
You will be able to search with awareness of the strength of the relationship.

C5. Structure of Index File The character pattern / separation pattern and its document number / position number in the document must be stored in a form that can be efficiently retrieved during retrieval. Therefore, in this embodiment, the index file is a character pattern file (a file that mainly stores character patterns and delimiter patterns), and a position information file (a file that mainly stores document numbers and in-document position numbers).
It consists of two files. The character pattern file stores where the character pattern / separation pattern and the corresponding document number / in-document position number are located in the position information file. The position information file stores the document number and the position number in the document.

FIG. 3 shows an example of such a character pattern file and the corresponding position information file.

In FIG. 3, character pattern file 3
The entry 02 is a character pattern of N consecutive characters (here, N = 2) in all documents in the database 202. The entries in the character pattern file 302 are preferably sorted in ascending order by the code value of the first character of the normalized character pattern to allow a binary search. "Separation pattern 1",
“Separation pattern 2”, “Nari”, “Haharu”, and the like are individual entries in the character pattern file 302.
Note that, for example, "separation pattern 1" is ",", ","
Or it is a comprehensive delimiter of sentences and phrases such as ".", And a special 2-byte value is assigned.

The position information file 304 in FIG. 3 stores at least one document number corresponding to each entry of the character pattern file 302 and at least one in-document position number associated with each document number. doing.

Although not shown, the character pattern file 302 is associated with the entry of the character pattern file 302 and the entry of the position information file 304.
Of each location information file 304
Of the entry of the position information file 304, and the size of the corresponding entry of the position information file 304. That is, in FIG. 3, for example, the character pattern file 302 seeks the position information file 304 from the beginning from the offset information stored therein in relation to the “delimiter pattern 2”,
Read from the position seeked by the number of bytes specified in the size information, and thereby, in relation to “separation pattern 2”, the position number value in the document of 8, 13, 22,. It is possible to collectively read the in-document position number value related to the number 2,... (If any) the in-document position number value related to the document number n.

Generally, the value of the position number in the document associated with the document number i is, for example, (document number i: 4 bytes) (the number of position numbers in the document k: 4 bytes) (first position number in the document: 4). (Byte) ... (k-th document position number: 4 bytes). In this example, the field for storing the in-document position number takes 4 bytes to store the absolute position of the document, but actually stores the offset from the previous in-document position number. Thus it is better to save this to 1-3 bytes. The same applies to the field for storing the number of the document number and the position number in the document.

C6. Index File Creation Process Next, the index file creation process will be described with reference to FIG. This processing is performed when the first database 202 is constructed or when a document is added to or deleted from the database 202 as shown in FIG.
This is a process performed by the index creation / update module 206 of FIG.

In FIG. 4, first, at step 402, a process of securing a memory area is performed. This is a process of acquiring a work area of a predetermined size on the RAM 104 by calling a function of the operating system, for example.

In step 404, the database 202
From the document is preferably loaded into the memory area obtained in step 402 above.

In step 406, the above-described normalization processing is performed on the document read in step 404.

In step 408, a character pattern / delimiter pattern is created by scanning the normalized document, and the character pattern / delimiter pattern, the document number of the document, and the position of the character pattern / delimiter pattern in the document. The number is stored in the memory area obtained in step 402 above.

In the process of step 408, as the character pattern, the document number, and the position number in the document are stored in the memory area obtained in advance in step 402, the obtained memory area becomes insufficient in free space. It is possible. Therefore, in step 410, a process is performed to check whether the obtained memory area is full. If so, in step 412, the character pattern / separation pattern stored in the memory area, the document number of the document, The position information in the document of the character pattern / delimiter pattern is sorted based on, for example, the character code value of the character pattern / delimiter pattern / the document number / the position number in the document, and is written to the disk 108 (FIG. 1) as an intermediate file. As a result, the memory area in which the data written in the intermediate file was stored is released for use in the following process. Then, this post-processing is step 4
Proceed to 14.

If it is determined in step 410 that the memory area still has room, the process immediately proceeds to step 414.
Proceed to.

In step 414, it is judged whether or not the document not read in step 404 remains in the database 202. If so, the process returns to step 404.

When it is determined in step 414 that the reading process of all the documents in the database 202 is completed,
The character pattern / delimiter pattern remaining in the memory area acquired in step 402, which is not written out, the document number of the document, and the document position number of the character pattern / delimiter pattern are the characters of the character pattern / delimiter pattern. The files are sorted based on the code value, the document number, and the position number in the document, and the disk 108 (FIG. 1) as an intermediate file
Is written out.

By writing the intermediate files in steps 412 and 416, there are a plurality of intermediate files on the disk 108, and each of the intermediate files has been sorted in advance. The character pattern file 3 shown in FIG. 3 is extracted from the plurality of intermediate files by the technique of merge sort.
02 and the location information file 304, and disk 1
The processing of storing in 08 is performed. It should be noted that, since the character patterns can appear repeatedly in the original plurality of intermediate files many times, here, the entries of the same overlapping character pattern are combined into one, and the document number and document A process of associating the internal position number is performed. afterwards,
The intermediate files are no longer needed and are deleted.

D. Search Process Using Index File Next, an example of a process of performing a character string search using the index file created as described above will be described with reference to the flowchart of FIG. In step 502,
First, for example, a process of displaying a dialog box having an input box and prompting the user to input a search character string in the input box is performed.

When the user inputs a search character string in the input box and clicks the OK button, the search character string is normalized as necessary, and then the first N character pattern of the search character string is used. Thus, the search process using the index file is performed. The length of the N character pattern here is the same as the length N of the character string pattern of the index file. Therefore, the N character pattern, which is a partial character string of the search character string, is used as a key for the index. Files can be searched for two minutes at high speed. One example of a suitable value for N for Japanese documents is 2.

In step 506, the first N of the search character string
If it is determined that the character pattern was not found, a message box indicating that the search character string was not found is preferably displayed in step 508, and the process ends.

At step 506, the first N of the search character string
If it is determined that the character pattern is found, then the index file returns one or more document numbers and at least one in-document position number for that document number, so that this information is available for further processing at step 510. The data is temporarily stored in a main storage or a predetermined buffer area on a disk.

In step 512, all the search character strings are set to N
It is determined whether or not the character string has been searched as a partial character string, and if so, the process proceeds to step 520.
Proceed to. If not, at step 518, a search process is performed using the index file with the next N character patterns of the search string. Note that the length of the search character string is generally not limited to a multiple of N, and therefore, when the process of searching for N character patterns and taking a partial character string to near the end of the search character string proceeds, In some cases, the key character string of the index file is shorter than N characters. In this case, the last N characters of the search character string are taken as the partial character string. Then, there is a character that overlaps with the N characters taken immediately before. If the search string is less than N characters,
The result of the binary search has multiple candidates, and the subsequent processing is
A sequential search will find multiple candidates.

In step 516, it is determined whether or not the N character pattern of the search character string is found in the index file, as in step 506. However, step 5
16 is essentially different from step 506 in step 516 in which the position number in the document obtained by incrementing the position number in any document by N in any document number related to the N character pattern immediately before the search character string. If it does not hit the character pattern that it has, it is not considered to have been found.

If it is determined in step 516 that the N character pattern of the search string was not found, then preferably a message box indicating that the search string was not found is displayed in step 508 and the process is terminated. finish.

If it is determined in step 516 that the N character pattern of the search character string has been found, the previous one of the document number returned as a result of the search of the index file and at least one position number in the document in the document number Only those whose position numbers are sequentially linked within the same document as the N character pattern information are temporarily stored in the main memory or a predetermined buffer area on the disk for subsequent processing in step 518.

If it is determined in step 512 that all the search character strings have been searched, the flow advances to step 520 to determine the document number in which the search character string exists from the document number and the document position number stored in the buffer. The position is determined. In step 522, the contents of the document in the database 202 are accessed from the document number and the position number in the document, and the corresponding line of the document in which the document search character string exists is
Preferably it is displayed in a separate window.

In order to check that the search character string appears in a specific block (eg, the third block) in the document, the delimiter position in the document appearing up to the position where the search character string appears in the document. , The block (the block number) in which the search character string is located in the document, and comparison with a designated block number.

E. Ambiguous search process The process shown in FIG. 5 was a process of performing an exact search using an index file, but according to the present invention, a specified character string and character sequence are used using the index file. A so-called ambiguous search process including a character string similar to can be executed at high speed for each document in the database. Especially, in this method, the character string you want to search,
Specify the search precision (greater than 0 and less than or equal to 1), and documents containing "similar character strings" and "similar characters" whose "similarity" with the character string you want to search is greater than or equal to the specified search precision Column "
To identify the position within the document.

E1. Human sense of feeling that character strings are similar There are the following Japanese character strings that humans who understand Japanese feel that the arrangement of characters is similar and the meaning is similar.

[Table 3] (1) Different notation of katakana small and large letters "Software""Software" Presence of long sound "-""Compiler""Compiler" Presence of Nakaguro "・""IBM""I-B"・
"M" and others "Building""Building" (2) Particles inserted between Kanji compound words and Kanji compound words "In-house prosecution""In-homeprosecution""Political world reorganization""Political world reorganization" (3) Kanji compound words The compound word of the compound and the compound word with a missing part “National Museum of Ethnology”, “National Museum”, “Ethnic Museum” (4) Partially applied by abbreviations, etc. “Software Development” “Software Development” (5) Wrong input "California""California"

What is common to these is that most characters match continuously but there are missing or extra characters.

Next, considering some words from the viewpoint of which one is similar, what is similar to “software maker” is “software maker”, “software development maker”, “software development”. In order of "maker",
Compared to the "Political Funds Control Bill," I feel that the order is similar to "Political Funds Control Act", "Political Funds Control", and "Political Funds".

Even though the characters match, it seems unreasonable to say that "machine maker whose main business is the manufacture of soft ice cream manufacturing machines" is a character string similar to "soft maker".

From these facts, the human sense of whether or not character strings are similar can be summarized as follows.

(A) The more consecutive characters are matched, the more similar the characters are, (B) The more mismatched characters are trapped in the middle, the more dissimilar the characters are, and (C) The too many mismatched characters are caught in the middle. It can be said that it is not felt as one character string.

At this time, it is necessary to consider a special case where a part of the input character string appears repeatedly at a close position in the document. For example, a case where the input character string is "Included as Dean of Science" and the document indicates "Inducted as Dean of Science" is described in the document. One of the characters "Dub" appearing redundantly is an extra character, but "No"
It is more reasonable to think that it is a character closer to the matching character than a completely unrelated character such as.

E2. Consistency with the structure of the index file The structure of the index file shown in FIG. 3 is an N-character character pattern indexed with the document number / position number in the document, and the minimum unit of search processing is one character pattern. To obtain the document number and the position number in the document. When searching for a character string of less than N characters, it is necessary to repeat the minimum unit search process for all the character patterns starting with the character string to be searched, and the number may be quite large. It can be said that a search for an input character string of less than N characters has a heavy load, whereas a search for an input character string of N characters or more may be performed by performing a minimum unit search of the number of characters of the input character string at most.

Therefore, it can be said that, in order to maintain high speed, it is appropriate to cut off the partial match of less than N characters and determine a similar character string based on the part of continuous match of N characters or more.

E3. Rules for determining similar character strings and degree of similarity Among character strings that continuously match the input character string with M characters or more, those that are close to each other in the same order as in the input character string are collected and made similar. The outline of the rule is to determine the degree of similarity based on the number of characters that match and the number of characters that do not match.

First, words used in the description are defined.

Matching character string: A part where the character string to be searched and the document text continuously match for M characters or more. Choose the one with the largest length starting with the same letter.

(Example) Character string to be searched: Political Fund Regulation Bill Document Text: ... For the purpose of fund regulation, the power of law ...

Let M = 2. Then, "Fund control" is a matching character string. At this time, the “fund” and the “fund rule” are not called the matching character strings because they are the longest selection. In addition, since "mod" is less than two characters, it does not become a matching character string.

Valid match character string: A match character string that constitutes a similar character string.

Maximum non-matching character string length L: The number of non-matching characters included in a similar character string is limited to consecutive L characters. L is a constant of 1 or more.

A method of selecting a “similar character string” and a method of digitizing the “similarity degree” will be described.

(1) Determination of First Valid Matching Character String The first matching character string is determined as the first valid matching character string in the order in the document. here,

The start position of the i-th effective matching character string in the document is s (D, i). The ending position of the i-th effective matching character string in the document is e (D, i).
i) The start position of the i-th valid matching character string in the character string to be searched is s (C, i) The end position of the i-th valid matching character string in the character string to be searched is e (C, i) Will be written as.

(2) Determination of the Next Valid Matching Character String When the i-th valid matching character string has been determined, the (i + 1) th valid matching character string is determined as follows.

The first matching character string that satisfies the following two conditions a) and b) is defined as the (i + 1) th valid matching character string.

[0093]

A) e (D, i) + 1 ≦ s (D, i + 1) ≦ e (D, i) + L + 1

This means that an extra character between the i-th valid matching character string and the (i + 1) -th valid matching character string is allowed up to L characters. (See Example 3 below)

B) s (C, i + 1)> e (C, i)-(M-1)

Repeat until such a valid matching character string cannot be selected.

(3) Determination of "similar character string" and its "similarity" (similarity) If no more valid matching character string can be selected, the first character to the last character of the first valid matching character string The "similar character string" is used up to the last character of the effective matching character string of "," and the "similarity degree" is calculated by the following expression.

[Equation 3] Similarity = minimum (Number of characters belonging to valid matching character string in character string to be searched / number of characters in character string to be searched, number of characters belonging to valid matching character string in "similar character string" / Number of characters in "similar character string")

E4. . How to count the number of characters that belong to a valid matching character string in "similar character strings" If the two characters correspond to the same character in the character string you want to search, count the first as 1 and 2 The second counts as 0.5. In other cases, 1 character is counted as 1. (See Example 4 below)

E5. Order of determination of "similar character string" The first "similar character string" is determined by starting the comparison from the beginning of the document. When the i-th "similar character string" is determined, the (i + 1) th "similar character string" is after the first character of the i-th "similar character string" and the i-th "similar character string" Determine by starting the comparison with the first character that does not belong to a valid match string that constitutes a "similar string".

By setting the constants L and M to appropriate values, it is possible to calculate the "similarity degree" which is substantially in agreement with general human judgment as to whether or not the arrangement of characters is similar.

When the “similarity” reaches the maximum value of 1, the character strings are completely matched, and if the character strings are completely matched, the “similarity” is always 1 become. E6. Ambiguous search processing flowchart The above processing is shown in a flowchart of FIG. In FIG. 6, first, in step 602, a search character string is prompted. Also, in step 604, the user is prompted to enter a degree of similarity of 0-1. Normal,
Input of character strings and values in steps 602 and 604 is performed using a single dialog box using an input box and a scroll bar.

In step 606, the number i of the valid matching character string is set to 1. In step 608, a search for a valid matching character string is performed. Assuming now that there is a condition that the length of the valid matching character string is M or more, it is advantageous if an index file of M character patterns is created in the processing of FIG. This is because, if such an index file exists in advance, a search for an arbitrary M character pattern is executed at high speed by a binary search of the index file.
Subsequently, the start position for taking the M character pattern in the search character string is shifted by one, and the search for the M character pattern is performed in the index file. And the position numbers in the document are sequential, it means that a valid matching character string of length M + 1 has been obtained. Thus, every time the condition that the document number is the same as the search of the M character pattern one time before and the position numbers in the document are sequential is satisfied, the length of the valid matching character string is also 1 Is incremented by one. However, if nothing is found in the search for the M character pattern using the index file, the returned document numbers do not match, or the position numbers in the document are not sequential, the end position of the valid matching character string is found. Will be.

In some cases, no valid matching character string can be found. In such a case, step 61
If the result is 0, the process proceeds to step 626, where the message "not found" is displayed and the process ends.

If it is determined in step 610 that a valid matching character string has been found, the process proceeds to step 612.
S (D, i) to e (D, i) in the document, s (C, i) in the search string
From to e (C, i) are marked as valid strings.

At step 614,

A) e (D, i) + 1 ≦ s (D, i + 1) ≦ e (D, i) + L + 1 and b) s (C, i)> e (C, i) )-(M-1)

The i + 1th valid matching character string satisfying the condition is searched again by using the index file, and if found, the process returns to step 612 and, regarding the i + 1th valid matching character string, In the document, s (D, i + 1) to e (D,
i + 1), s (C, i + 1) to e (C, i + 1) in the search string,
Marked as a valid string (step 618)
The increment of i in indicates that we are looking at the next valid match string).

On the other hand, if the valid matching character string is no longer found in step 616, the similarity is calculated in step 620. This is, for example,

[0107]

[Equation 5] Similarity = minimum (number of characters belonging to the valid matching character string in the character string to be searched / number of characters in the character string to be searched, number of characters belonging to the valid matching character string in the "similar character string" / Number of characters in "similar character string")

Is given by At this time, the “similar character string” is a character string in the document from the start position of the first valid matching character string to the last position of the last valid matching character string.

At step 622, a result is selected from the similarity calculated at step 620 and the similarity input at step 604, and only those whose results are equal to or higher than the similarity input at step 604 are selected. , Step 624
To display the result.

In step 624, the document content stored in the database is accessed based on the document number returned as a result of the search of the index file in steps 608 and 614 and the position number in the document, and the corresponding location is determined. The process of displaying the containing line is performed.

The "similar character string" for one search character string may be found in a plurality of documents at the same time, but may be found in a plurality of places even in a single document. Therefore, steps 606 to 622 are applied to each of the plurality of “similar character strings”, and in step 624, the similarity condition is satisfied among the plurality of “similar character strings”. Note that only the items are sorted and displayed.

E7. Example of determining “similar character string” and similarity M = 2, L = 3.

(Example 1) (* IBM is a trademark of IBM Corporation)

[0114]

Since the longest character string that matches first is "eye"

The first valid matching character string is “eye” s (C, 1) = 1 e (C, 1) = 2 s (D, 1) = 1 e (D, 1) = 2

Since e (C, 1)-(M-1) = 1, the character string starting from the second character of the character string to be searched is replaced with the character string 3 of the document.
Search for the second valid matching character string by comparing the character string starting from the fourth, fifth or sixth character (e (D, 1) +1 = 3, e (D, 1) + L + 1
= 6).

Second valid matching character string “B” s (C, 2) = 3 e (C, 2) = 4 s (D, 2) = 4 e (D, 2) = 5

Since e (C, 1)-(M-1) = 3, the character string starting from the fourth character of the character string to be searched is set to 5, 6, 6,
Search for the third valid matching character string by comparing the character string starting from the 7th or 8th character (e (D, 2) +1 = 5, e (D, 2) + L + 1 = 8
So). Third valid match string "M" s (C, 3) = 5 e (C, 3) = 6 s (D, 3) = 7 e (D, 3) = 8

Since the end of the character string to be searched has been reached, the third effective match character string is the last.

[0121]

[Table 4]

The number is the number of a valid matching character string.

Therefore, "similar character string" is s (D, 1)
"IBM" from to e (D, 3). "Similarity" = minimum (6/6, 6/8) = 6/8 = 0.75

(Example 2)

[Table 5]

[0125]

[Table 6]

Similar character string = "Software development maker" Similarity = minimum (7/10, 7/9) = 0.7

(Example 3)

[Table 7]

[Table 8] 1 2 3 4 5 6 7 8 9 101112131415 Document D: I came to prosecution while staying at home.

Since the longest character string that matches first is "at home"

## EQU00006 ## The first valid matching character string is at home s (C, 1) = 1 e (C, 1) = 2 s (D, 1) = 1 e (D, 1) = 2

A character string starting from the third, fourth, fifth or sixth character of the document and (e (D, 1) + 1 = 3, e (D, 1) + L + 1 = 6)
Replace the character string starting from the second character of the character string you want to search with (
e (C, 1)-(M-1) = 1) to find the second valid match.

Since the second valid matching character string is not found, the end of the character string to be searched has been reached, so that only the first valid matching character string is found.

[Table 9] was Tsu Ki tried fu between or in prosecutions of standing home prosecuted 1 standing home. 1

Thus, the first “similar character string” is s
"At home" from (D, 1) to e (D, 1). "Similarity" = minimum (2/4, 2/2) = 2/4 = 0.5

The first non-valid matching character after “to” is “
If you search for the second "similar string" after "." Of ",

[Table 10] were Tsu Ki tried fu between or in prosecutions of standing home prosecuted 1 standing home. 1

However, in the document, “at home” and “prosecution” are separated by 4 characters, and in this example, L = 3, so the above “promotion” is regarded as a valid matching character string. Absent.

(Example 4)

[Table 11]

[0135] The longest matching first character string is "bank".

[0136]

The first valid matching character string is “bank” s (C, 1) = 1 e (C, 1) = 2 s (D, 1) = 2 e (D, 1) = 3

A character string starting from the second character onward of the character string to be searched (because e (C, 1)-(M-1) = 1) is 4, 5, 6 of the document.
Or the character string starting from the 7th character and (e (D, 1) +1 = 4, e
(D, 1) + L + 1 = 7 (because this is the case) and find the second valid matching character string.

The second valid matching character string is “line member” s (C, 2) = 2 e (C, 2) = 3 s (D, 2) = 4 e (D, 2) = 5

Since the end of the character string to be searched has been reached, there are two valid matching character strings.

[Table 12] 1, 1, 0.5, 1-> 3.5

"Similar character string" is from s (D, 1) to e (D,
2) Up to "bank clerk". "Similarity" = minimum (3/3,
3.5 / 4) = 3.5 / 4 = 0.875

E8. Examples of vagueness close to human sense

[Table 13] Software Maker Software Maker 0.909 Software Development Maker 0.833 Software Development Maker 0.769

This example shows that the "similarity" decreases as extra characters are entered.

[Table 14] Knitting Towea Studio 0.800 software maker 0.700 software 0.600

In this example, as the number of matching characters decreases, ""
"Similarity" decreases.

[Table 15] Science elections Faculty of Science elections 1.000 Faculty of Science, director election 0.929 Science director election 0.857

E9. Relationship between Index Structure and Search for "Similar String" By appropriately determining the value of M, the fuzzy search for "similar string" can be realized at a considerably high speed by the index structure of the present invention. .

How to determine constants N and M

[Table 16] N: Number of characters of character pattern stored in index M: Minimum length of valid matching character string in fuzzy search L: Maximum length of non-valid matching character string in "similar character string" in fuzzy search

When N is increased, the number of types of character patterns increases and the amount of data per character pattern decreases, so the search becomes faster, but the capacity of the index file increases. For an average Japanese document, N = 2 gives a sufficient search speed.

If M is set so that M ≧ N,
Sufficient search speed can be obtained in fuzzy search. Considering that the smaller M is, the more detailed and ambiguous search can be performed, it may be better to determine M = N.

E10. Second Embodiment of Determining Similarity

In the fuzzy search process of the second embodiment, in particular,
A consideration is given to the fact that "the more mismatched characters that are trapped in the middle, the more dissimilar it feels" and "There are too many unmatched characters that are trapped in the middle to make us feel that it is not a single character string". If a document matches a character string that matches the input character string, a non-matching character string, and a matching character string in that order, the similarity is obtained by capturing the latter matching character string as a similar character string. It is unnatural for the to drop. For example, if the input character strings are “charged at home,” document 1 is “charged at home,” and document 2 is “charged at home,” document 1 is “charged at home,” and document 2 is “charged at home. Is a similar character string, and the degree of similarity is higher for "at home". This rule is contrary to human perception. On the contrary, it is unnatural to receive a low rating. Either the degree of similarity of “in-charge at home” is higher than “at-home”, or the similar character strings in document 1 are determined to be “at-home” and “indictment” But it is natural.

Next, the processing of the second embodiment will be described. Referring to the flowchart of FIG. 6, in this embodiment, steps 602 to 612 are the same, and the process of step 614 indicating the condition for searching for the (i + 1) th valid matching character string is changed as follows. To be done.

[0151]

S (C, i + 1)> e (C, i)-(M-1) ... (Equation 1) s (D, i + 1)> e (D, i) ... (Equation 2) and s (D, i + 1)-e (D, i)-1 + max (e (C, i)-s (C, i + 1) + 1, 0) ≤ L (Formula 3)

S (C, i), e (C, i), s (D, i), s (D, i)
The definitions such as are as described above.

Formula 1 shows that M-1 characters (up to M-1 characters), such as "part" of the "Department of Science", are allowed, and other characters appear in the same order as the character order in the input character string. Means that the character string to be valid is valid.

Expression 2 means that valid matching character strings do not overlap each other in the document.

In the expression 3, the non-matching characters sandwiched between them and the duplicate occurrence character such as "section" of "Department of Science" are combined into L
It means that even letters are allowed.

In this embodiment, as in the previous embodiment, the ratio of the effective matching character string to each of the similar character strings in the search character string and the document is calculated, and the smaller one of the two is calculated. Rather than selecting as a degree, calculate by scoring similar character strings and dividing by a perfect score (score in the case of perfect match) to obtain a ratio. The score of a similar character string is calculated by adding a score to each character according to the following rules and adding them. Therefore, the processing in step 620 of FIG. 6 is as follows.

Character belonging to the first valid matching character string: 1 point belongs to the i-th (i> 1) valid matching character string, and position in the search character string ≧ e (C, i−1) +1 (Equation 4) ・ ・ ・ 1 point Position in search character string ≤ e (C, i-1) (Equation 5) ・ ・ ・ -1 / (2 * L) points Characters that do not belong to valid matching character string ・ ・・ -1 / L point

Also in this embodiment, when the i-th similar character string is determined, the i + 1-th similar character string is i
The comparison is determined starting from the first character after the first character of the second similar character string and not belonging to the valid matching character string forming the i-th similar character string.

The minus point of the characters that do not belong to the valid matching character string is that "the more the number of non-matching characters that are trapped in the middle, the more dissimilar it feels", and the "there are too many non-matching characters that are trapped in the middle of one character string. It is set in consideration of the fact that "I can not feel it." The maximum of the sum of the negative points of one non-matching string is 1 / L * L = 1,
The minimum positive point by incorporating the following matching character string is N ≧ 1 (2 is recommended especially for Japanese)
Therefore, the negative point does not exceed the positive point. In addition, Expression 5 indicates a duplicate occurrence character such as “part” of the above-mentioned “Department of Science”, and Expression 4 indicates a simple matching character that is not a duplicate occurrence character. By adding a minus point smaller than that of a simple non-matching character to the character represented by Expression 5, it is possible to deal with the case where the character appears in duplicate.

E11. Example of Determining Similarity of Character String and Similarity in Second Embodiment An example will be shown again with N = 2 and L = 3.

[Table 17]

Since the first matching character string is "eye", the first valid matching character string is "eye".

S (C, 1) = 1 e (C, 1) = 2 s (D, 1) = 1 e (D, 1) = 2

The second valid matching character string is "Bee" according to Expression 1, Expression 2 and Expression 3.

S (C, 2) = 3 e (C, 2) = 4 s (D, 2) = 4 e (D, 2) = 5

The third valid matching character string "M" according to Expression 1, Expression 2 and Expression 3

S (C, 3) = 5 e (C, 3) = 6 s (D, 3) = 7 e (D, 3) = 8

Since the end of the input character string is reached, there are three valid matching character strings.

[Table 18]

Similar strings are from s (D, 1) to e (D, 3)
Up to "IBM". Degree of similarity = ((1
* 6 + (-1/3) * 2) / 6) = 0.88

[0166]

[Table 19] (Example 6) 1 2 3 4 5 6 7 8 9 10 Input character string C: Software maker 1 2 3 4 5 6 7 8 9 Part of document D:… Software development maker

[0167] Similar string = "Software developer" Similarity = ((1 * 7 + (-1/3) * 2) / 10) =
0.63

[0168]

[Table 20] (Example 7) 1 2 3 4 Input character string C: Prosecution at home 1 2 3 4 5 6 7 8 91011121314 Part of document D: ... Prosecution was completed while staying at home. ...

Since the first matching character string is “at home”, the first valid matching character string is “at home”. Since the next matching character string “indictment” does not satisfy Equation 3, the first valid matching character string is only the first. .

[Table 21] C: was Tsu Ki tried fu between or in prosecutions of standing home: standing home prosecuted 1 D. 1

A similar character string is “at home”. Similarity = 2/4 = 0.5

The first non-valid matching character after "present" is "no". The second similar string is searched after "no".

[0172] C: was Tsu Ki tried fu between or in prosecutions of standing home: standing home prosecuted 1 D. 1

Therefore, the second similar character string is "prosecution".

[0174]

[Table 22]

There are two valid matching character strings: "Science Department" and "Appointed Director".

[0176]

[Table 23]

A character string that resembles is "appointed as dean of science department".
The second “part” satisfies Expression 5. So the degree of similarity = ((1 * 7 + (-1/6) * 1) / 7) = 0.97
Becomes

E12. Summary of the results of the second example

[Table 24] Input character string Document similarity ======================================== the length of ============== software development of the manufacturer software manufacturers 0.95 software maker 0.85 political funds Control bill political funds Control Law 0.87 political funds 0.50 Faculty of Science 0.97 Faculty of Science, became the appointed Science Director in length Inauguration 0.95

[0179]

As described above, according to the present invention, an effect is obtained that a fuzzy search natural to human sense can be realized at high speed in a document file or a database using a unique index structure.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a hardware configuration.

FIG. 2 is a block diagram of a processing element.

FIG. 3 is a diagram showing a structure of an index file.

FIG. 4 is a flowchart showing an index file creation process.

FIG. 5 is a flowchart of a character string search process using an index file.

FIG. 6 is a flowchart of a fuzzy search process using an index file.

Continuation of front page (56) References JP-A-4-215181 (JP, A) JP-A-4-293161 (JP, A) JP-A-2-279973 (JP, A) Information Processing Society of Japan Research Report (92- FI-25) P.I. 1-8, P.I. 9-16

Claims (41)

(57) [Claims] 1. A method for finding the similarity between a search character string and a character string in the document in a document that is searchably stored by computer processing, the method comprising: (a) inputting the search character string. (B) From the beginning of the search string above, (M
Finds a matching start position and end position in the above document for a partial character string of 2 or more scheduled integers (hereinafter,
A subcharacter string having a length of M characters or more determined by its start position and end position is referred to as a valid matching character string) stage, and (c) no valid matching character string was found in the above stage (b). In response, for each partial character string having a length of M characters or more (M is a planned integer of 2 or more) obtained by shifting the start position of the partial character string of the search character string, the valid matching character string And (d) in response to the above valid match string being found,
The start position of the partial character string in the search character string and the start position of the search in the document are shifted by the length of the immediately preceding effective match character string, and the immediately preceding effective match character string of the start position is shifted. The distance from the end position of is within L characters (L
Is a scheduled integer greater than or equal to 1), and (e) as long as the above valid match string is found, above step (d)
And (f) at least from the start position of the first valid match string in the above document to the end position of the last valid match string in the above document, Based on the existence information, the similarity between the search character string and the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document is determined. An information retrieval method, comprising a step of calculating. 2. The information search method according to claim 1, wherein the M is 2 and the L is 3 or more. 3. The calculation of the similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, to the end of the last valid matching character string in the document. The information search method according to claim 1, wherein the method is performed using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 4. The calculation of the similarity is performed for each character of the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document, If it belongs to a valid match string, the score is added, otherwise it is deducted and the resulting score value is
The information search method according to claim 1, wherein the information search method is performed using a value divided by a score value of perfect match. 5. An information search method for finding a position where a search character string appears in a document stored so as to be searchable by computer processing, comprising: (a) inputting a search character string; The step of inputting the degree of similarity, and (c) From the beginning of the search character string above, (M
Finds a matching start position and end position in the above document for a partial character string of 2 or more scheduled integers (hereinafter,
(A partial character string having a length of M characters or more determined by its start position and end position is referred to as a valid match character string)); and (d) no valid match character string was found in step (c) above. In response to this, for a partial character string of M characters or more (M is an integer of 2 or more) in which the start position of the partial character string of the search character string is shifted by one, the valid matching character string And (e) in response to the above valid match string being found,
The start position of the partial character string in the search string and the start position of the search in the document are each shifted by the immediately preceding valid match character string, and the end position of the start position of the immediately preceding valid match character string is shifted. Distance from within L characters (L is 1
The step of finding a valid match string that is the above integer), and (f) the above step (e) as long as the above valid match string is found.
And (g) at least the valid match string in the string between the start position of the first valid match string in the above document and the end position of the last valid match string in the above document. Based on the existence information, the similarity between the search character string and the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document is determined. And (h) displaying the content of the document including the valid matching character string in response to the calculated similarity being equal to or greater than the similarity input in the step (b). An information retrieval method having steps. 6. The information retrieval method according to claim 5, wherein the M is 2 and the L is 3 or more. 7. The calculation of the similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, and the end of the last valid matching character string in the document. The information search method according to claim 5, wherein the method is performed by using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 8. The similarity calculation is performed for each character of a character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. If it belongs to a valid match string, the score is added, otherwise it is deducted and the resulting score value is
The information search method according to claim 5, wherein the information search method is performed by using a value divided by a score value of perfect match. 9. The information search method according to claim 8, wherein the added point is 1 for one character, and the deducted point is 1 / L for one character. 10. A method for finding a similarity between a search character string and a character string in a document in a database including a plurality of documents retrievably stored by computer processing, comprising: (a) a search character string; (B) From the beginning of the search string above, enter (M
Finds a matching start position and end position in the same document in the database for a partial character string of 2 or more planned integers (hereinafter, M characters or more determined by the start position and end position) (C) calling a substring of the search string in response to the fact that no valid match string was found in step (b). Finding the valid matching character string for a partial character string of M characters or more (M is an integer of 2 or more) shifted by one position; and (d) finding the valid matching character string. In response to what was done
The start position of the partial character string in the search string and the start position of the search in the same document are each shifted by the length of the immediately preceding valid matching character string, and the immediately preceding valid matching character of the starting position is shifted. A step of finding a valid matching character string whose distance from the end position of the string is within L characters (L is an integer of 1 or more); and (e) the above step (d) as long as the valid matching character string is found.
And (f) at least the valid match character string in the character string between the start position of the first valid match character string in the above document and the end position of the last valid match character string in the same document. Similarity between the search character string and the character string from the start position of the first valid matching character string in the same document to the end position of the last valid matching character string in the same document, based on the existence information of An information retrieval method having a step of calculating a degree. 11. The above M is 2, and the above L is 3,
The information search method according to claim 10. 12. The calculation of the degree of similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document to the end of the last valid matching character string in the document. The information search method according to claim 10, which is performed using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 13. The calculation of the similarity is performed for each character of the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. If it belongs to a valid matching character string, the score is added, otherwise it is deducted, and the score value of the result is divided by the score value of the perfect match.
The information search method according to claim 10. 14. An information search method for finding a place where a search character string appears in a database including a plurality of documents stored in a searchable manner by computer processing, comprising: (a) inputting a search character string; , (B) inputting the degree of similarity, and (c) from the beginning of the above search character string, (M
Finds a matching start position and end position in the same document in the database for a partial character string of 2 or more planned integers (hereinafter, M characters or more determined by the start position and end position) And (d) in response to the fact that no valid match was found in step (c), the start of the substring of the search string Finding a valid matching character string for a partial character string of M characters or more (M is an integer of 2 or more) shifted by one position; and (e) finding the valid matching character string. In response to
The start position of the partial character string in the search string and the start position of the search in the same document are each shifted by the immediately preceding valid match character string, and the end of the immediately preceding valid match character string at the start position is shifted. Distance from position within L characters (L
Is a scheduled integer greater than or equal to 1), and (f) as long as the above valid match string is found, above step (e)
And (g) at least the valid matching character in the character string from the start position of the first valid matching character string in the same document to the end position of the last valid matching character string in the same document. Based on the column existence information, the search string above,
Calculating a similarity between the start position of the first valid matching character string in the same document and the end position of the last valid matching character string in the same document; and (h) above. An information retrieval method, comprising: displaying content including the valid matching character string of the document in response to the calculated similarity being equal to or higher than the similarity input in step (b). 15. The information retrieval method according to claim 14, further comprising the step of previously assigning a unique number or symbol to the plurality of documents. 16. The information retrieval method according to claim 15, wherein the unique number or symbol is a sequential number. 17. The above M is 2 and the above L is 3.
The information search method according to claim 14. 18. The calculation of the degree of similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, to the end of the last valid matching character string in the document. The information search method according to claim 14, which is performed by using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 19. The calculation of the similarity is performed for each character of the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. If it belongs to a valid matching character string, the score is added, otherwise it is deducted, and the score value of the result is divided by the score value of the perfect match.
The information search method according to claim 14. 20. The information retrieval method according to claim 19, wherein the added point is 1 for one character, and the deducted point is 1 / L for one character. 21. An information retrieval system for finding a position where a search character string appears in a document stored so as to be searchable by computer processing, comprising: (a) means for inputting a search character string; For a partial character string of length M or more (M is an integer of 2 or more) of the search character string, find a matching start position and end position in the document (hereinafter, the start position and end position). (A partial character string of M characters or more determined by the position is referred to as a valid matching character string) means; and (c) applying the means (b) from the beginning of the search character string to find a valid matching character string. In response to this, the start position of the partial character string in the search character string and the start position of the search in the document are shifted by the immediately preceding valid matching character string, respectively, The end position of the valid matching string of Finds a valid matching character string whose distance is within L characters (L is an integer that is expected to be 1 or more), and responds to the fact that the valid matching character string is not found. Means for finding an effective matching character string for a partial character string of M characters or more (M is an integer of 2 or more) in which the start position of the effective matching character string is shifted by one; As far as it can be found, the above means (c)
And (e) at least the valid match character string in the character string from the start position of the first valid match character string in the above document to the end position of the last valid match character string in the above document. Based on the presence information of the above, the similarity between the search character string and the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document An information retrieval system having means for calculating. 22. An information retrieval system for finding a position where a search character string appears in a document stored so as to be searchable by computer processing, comprising: (a) means for inputting a search character string; Means for inputting the degree of similarity, and (c) for the partial character string of length M or more (M is an integer expected to be 2 or more) of the search character string, the start position and end that match in the document. Means for finding a position (hereinafter, a partial character string of M characters or more determined by its start position and end position is referred to as a valid matching character string); and (d) said means (c) from the beginning of said search character string. In response to finding a valid matching character string, the start position of the partial character string in the search character string and the starting position of the search in the document are immediately preceding the valid matching character string. Just before the start position In response to not finding a valid matching character string, a valid matching character string whose distance from the end position of the valid matching character string is within L characters (L is an integer of 1 or more) is found, Means for finding the effective matching character string for a partial character string of M characters or more (M is an integer of 2 or more) in which the start position of the partial character string of the search character string is shifted by one; e) As long as the above valid matching character string is found, the above means (d)
And (f) at least the valid matching character string in the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. Based on the presence information of the above, the similarity between the search character string and the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document And (g) in response to the calculated similarity being equal to or greater than the similarity input in step (b), displaying the contents of the document including the valid matching character string. An information retrieval system including means for performing. 23. The above M is 2, and the above L is 3,
An information retrieval system according to claim 22. 24. The calculation of the similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, and the end of the last valid matching character string in the document. The information search system according to claim 22, wherein the information retrieval system is performed by using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 25. The calculation of the similarity is performed for each character of the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. If it belongs to a valid matching character string, the score is added, otherwise it is deducted, and the score value of the result is divided by the score value of the perfect match.
An information retrieval system according to claim 22. 26. The information retrieval system according to claim 25, wherein the added point is 1 for one character, and the deduction point is 1 / L for one character. 27. An information search system for finding a position where a search character string appears in a database including a plurality of documents stored in a searchable manner by computer processing, comprising: (a) means for inputting the search character string; , (B) means for inputting the degree of similarity, and (c) substrings of the search character string having a length of M characters or more (M is an integer of 2 or more) are matched in the above document. Means for finding a start position and an end position (hereinafter, a partial character string of M characters or more determined by the start position and the end position is referred to as a valid matching character string); and (d) The means (c) is applied, and in response to a valid matching character string being found, the start position of the partial character string in the search character string and the start position of the search in the same document are respectively set. Shift by last valid match Te, of its starting position, within the distance L characters from the end position of the effective match the character string before the straight (L is,
Find a valid match string that is a planned integer greater than or equal to 1),
In response to the fact that no valid matching character string was found, the start position of the partial character string of the search character string was shifted by one and the length was M characters or more (M is an integer of 2 or more) For the partial character string, the means for finding the valid matching character string, and (e) As long as the valid matching character string is found, the means (d)
And (f) at least in the character string from the start position of the first valid matching character string in the same document to the end position of the last valid matching character string in the same document, Based on the existence information of the valid matching character string, between the search character string and the start position of the first valid matching character string in the same document to the end position of the last valid matching character string in the same document. Means for calculating the degree of similarity with the character string; and (g) in response to the calculated degree of similarity being greater than or equal to the degree of similarity input in step (b), the validity of the document An information retrieval system comprising means for displaying contents including a matching character string. 28. The above M is 2, and the above L is 3.
An information retrieval system according to claim 27. 29. The calculation of the similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, and the end of the last valid matching character string in the document. The information search system according to claim 27, which is performed using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 30. The calculation of the similarity is performed for each character of the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. If it belongs to a valid matching character string, the score is added, otherwise it is deducted, and the score value of the result is divided by the score value of the perfect match.
An information retrieval system according to claim 27. 31. The information retrieval system according to claim 30, wherein the added point is 1 for one character, and the deduction point is 1 / L for one character. 32. An information search method for finding a place where a search character string appears in a database including a plurality of documents stored in a searchable manner by computer processing, comprising: (a) inputting a search character string; , (B) inputting the degree of similarity, and (c) from the beginning of the above search character string, (M
Finds a matching start position and end position in the same document in the database for a partial character string of 2 or more planned integers (hereinafter, M characters or more determined by the start position and end position) (D) the sub-character string of is called a valid matching character string), and (d) the start position of the i-th valid matching character string
i) The end position in the document of the i-th valid matching character string is e (D,
i) The start position of the i-th valid matching character string in the character string to be searched is s (C, i). The ending position of the i-th valid matching character string in the character string to be searched is e (C, i). , E (D, i) + 1 ≤ s (D, i + 1) ≤ e (D, i) + L + 1 and s (C, i + 1)> e (C, i)-(M- 1) finding the (i + i) th valid matching character string satisfying the condition (where L is an integer greater than or equal to 1); and (e) performing the above steps as long as the valid matching character string is found. d)
And (f) at least the valid matching character in the character string from the start position of the first valid matching character string in the same document to the end position of the last valid matching character string in the same document. Based on the column existence information, the search string above,
Calculating a similarity between a character string between the start position of the first valid matching character string in the same document and the end position of the last valid matching character string in the same document; and (g) above. An information retrieval method, comprising: displaying content including the valid matching character string of the document in response to the calculated similarity being equal to or higher than the similarity input in step (b). 33. The information search method according to claim 32, wherein the M is 2 and the L is 3 or more. 34. The calculation of the similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, and the end of the last valid matching character string in the document. 33. The information search method according to claim 32, which is performed by using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 35. The search character is calculated in order to calculate a ratio occupied by the valid matching character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. 33. The information search method according to claim 32, wherein characters belonging to a plurality of valid matching character strings in a string are weighted lower than other characters. 36. An information search method for finding a place where a search character string appears in a database including a plurality of documents stored in a searchable manner by computer processing, comprising: (a) inputting a search character string; , (B) inputting the degree of similarity, and (c) from the beginning of the above search character string, (M
Is a planned integer of 2 or more), and finds a matching start position and end position in the same document of the above database (hereinafter, a length of M characters or more determined by the start position and end position). (D) the sub-character string of is called a valid matching character string), and (d) the start position of the i-th valid matching character string
i) The end position in the document of the i-th valid matching character string is e (D,
i) The start position of the i-th valid matching character string in the character string to be searched is s (C, i) The end position of the i-th valid matching character string in the character string to be searched is e (C, i) , S (C, i + 1)> e (C, i)-(M-1), s (D, i + 1)> e (D, i) and s (D, i + 1)-e The condition (D, i)-1 + max (e (C, i)-s (C, i + 1) + 1, 0) ≤ L (where L is a planned integer greater than or equal to 1) Finding the i + ith valid match string that satisfies, and (e) as long as the above valid match string is found, above step (d)
And (f) at least the valid matching character in the character string from the start position of the first valid matching character string in the same document to the end position of the last valid matching character string in the same document. Based on the column existence information, the search string above,
Calculating a similarity between a character string between the start position of the first valid matching character string in the same document and the end position of the last valid matching character string in the same document; and (g) above. An information retrieval method, comprising: displaying content including the valid matching character string of the document in response to the calculated similarity being equal to or higher than the similarity input in step (b). 37. The information search method according to claim 36, wherein the M is 2 and the L is 3 or more. 38. The calculation of the similarity is based on the ratio of the valid matching character string in the search character string and the start position of the first valid matching character string in the document, and the end of the last valid matching character string in the document. The information search method according to claim 36, which is performed by using a smaller value of the ratios of the valid matching character strings in the character strings up to the position. 39. The calculation of the similarity is performed for each character of the character string from the start position of the first valid matching character string in the document to the end position of the last valid matching character string in the document. If it belongs to a valid matching character string, the score is added, otherwise it is deducted, and the score value of the result is divided by the score value of the perfect match.
An information retrieval method according to claim 36. 40. The information search method according to claim 39, wherein the added point is 1 for one character, and the deducted point is 1 / L for one character. 41. It belongs to the i-th valid matching character string,
41. The information search method according to claim 40, wherein the character in the corresponding search character string deducts 1 / (2L) from the character belonging to the i-1th valid matching string.