JPH0587865B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an information retrieval system for searching data matching search conditions from a database consisting of a large amount of character information in Japanese or the like.

(Prior art) As a free word-based information retrieval system,
Japanese Patent Application No. 61-055683 proposes a ``Japanese information retrieval system''. This has only a character index for each character, consisting of the character type and the data number designation bit of the actual data that contains that character.
For all character types, perform a logical operation between the bit strings that indicate the data number of the index, and from the result, know the data number of the actual data that includes all the character types of the specified character string, and then is read out and scanned, and it is determined whether or not it matches a character string specified as a search condition, thereby searching for the target actual data.

The above-mentioned method does not have an index specifically for characters and character combinations. Therefore, for example, as a combination of characters, "information retrieval"
If you try to search for the string ``jo''
After searching for data containing the four characters ``information'', ``search'', and ``search'', it is necessary to search for the character string ``information search'' in the data.

(Problem to be Solved by the Invention) In such a conventional method, logical operations using a character index cannot satisfy the requirement that the data contain at least all the characters in a specified character string. However, since the combination of characters is not taken into account, it is necessary to scan the actual data, and therefore, when there is a large number of candidate actual data, there is a problem that it takes an extremely long search time.

(Means and effects for solving the problem) The present invention, as a means for solving the above-mentioned problems, provides an information retrieval system that adds a data number to registered data when registering information, and adds a data number to the registered data. Extract collocations made up of characters and character combinations that appear, convert the collocation number of the collocation into a plurality of coordinate values (key values) that are represented as unique points in a multidimensional space consisting of a plurality of coordinate axes, and combine the key values and In a collocation index having a data number designation bit string corresponding to the key value, the collocation is registered by turning on the corresponding bits of the plurality of key values and the data number, and when searching for information, the collocation is registered. Extracts collocations based on combinations of characters included in the character string as a condition, and converts the collocation number of the collocations into multiple coordinate values (key values) that are represented as unique points in a multidimensional space consisting of multiple coordinate axes. Then, calculate the logical product of bits at the same position in the bit string corresponding to the key value of the collocation to be searched in the collocation index, and from the data number indicated by the resulting on-bit, determine the collocation of the character string as the search condition. The present invention provides a search method using a collocation index in an information retrieval system, which is characterized by searching for data containing .

Further, as a calculation method, a collocation number is set in the collocation according to the kanji code table, and key values of multiple coordinate axes indicating unique points in a multidimensional space are obtained from the collocation number.
A MOD operation is performed to obtain a remainder when the compound word number is divided by a plurality of prime numbers, and a type mark is added to the result of the operation, thereby converting it into a single kanji character number and using it as the plurality of key values. The present invention attempts to solve the above problem by using a search method using a collocation index in an information retrieval system.

A feature of the present invention is that, instead of pointing to a compound word consisting of a combination of characters using a single index as in the past, the compound word number of the compound word is indicated as a unique point in a multidimensional space consisting of a plurality of coordinate axes. The goal is to convert it into coordinate values (key values) and map it to a multidimensional space.

Methods for converting the collocation numbers mentioned above into multiple coordinate values (key values) include the most common division method, numerical analysis method, median square method, folding method, and Lin's method. , any function may be used as long as it is distributed as uniformly as possible within an appropriate range.

The method of the present invention expresses a predetermined combination of characters by the logical product of several keys. Although the number of permutation combinations is large, in the tens of millions, the total number of keys can be reduced to a few thousand, and indexes on characters and combinations of characters can be used just as well as traditional character indexes. It made it possible to handle it.

(Example) Next, an example of the present invention will be described with reference to the drawings.

Figure 3 shows how kanji are defined in the JIS code. Kanji are divided into level 1 and level 2 depending on their frequency of use, and one kanji character is represented by two code numbers. Let's say that the first code number is ``bian'' and the second code number is ``旁'', then the first level kanji shows 32 kanji from (30) to (4F) in hexadecimal notation. The second-level Kanji characters similarly have code numbers indicating 36 flats from (50) to (73). Each of these 68 旁 is combined with the 94 旁 code numbers from (21) to (7E), and the total number of kanji is the product of these numbers.
This results in 6392 characters (including the last 43 undefined characters of the first level).

The method of the present invention can be applied to any combination of characters, but in order to avoid complicating the explanation, we will explain the case of a combination of two kanji characters with reference to Figure 3 above. .

There are various methods of assigning numbers to combinations of characters, but here we will explain the case of using a serial numbering method.

Furthermore, there are various ways to determine the key when expressing the above-mentioned collocation by the logical product of several numbers (keys), but here we will use the modulus (MOD) of several integers. I will explain about it.

First, according to the table in Figure 3, the first character ``A'' is number 1, the next character ``唖'' is number 2, and so on, and so on, and the last character ``龠'' is It will be number 6392. Also, based on this, if we number the combinations of two kanji characters in order, ``aya'' is number 1, ``aya'' is number 2, ... ``aya'' is number 6392,
"Mua" is number 6393...The last "Koya" is 40857664
It can be the number.

The general formula for this is, if the first character number is P and the second character number is S, it is expressed as the conjunctive number N = (P-1) * 6.392 + S... (1), and from this, the predetermined number is The modulus of several integers (MOD) allows us to find keys and define collocations.

A more detailed explanation of the above will now be given using the operations of this embodiment when storing data and when retrieving data.

FIG. 1 is for explaining the operation of collocation index registration at the time of data storage in this embodiment, taking data "information storage and retrieval" as an example.

Assume now that data "information storage and retrieval" is input, assigned a data number (123 in this example), and stored in the actual data section (processing 1).

At this time, all characters included in the data "Information storage and retrieval" are extracted (Process 2...Hiragana characters are omitted here).

Next, the bit corresponding to the data number of each extracted character index bit string is set to logic "1" (◎) and registered in the character index (processing 3).

However, in the figure, only a part of the character index is schematically shown, and only the area around the 123rd bit corresponding to data number 123 and the area around the 456th bit are schematically shown, and the character index is shown schematically as a bit string. A continuous bit string for the required data numbers is prepared.

Next, all adjacent character and character combinations included in the data "information storage and retrieval" are extracted as collocations. In other words, 7 of "Storage/accumulation/product/examination/search of information/information"
However, in FIG. 1, combinations that are not kanji characters are not shown (processing 4).

Next, key values are determined so that each of these seven sets is expressed by the logical product of several keys (numeric values), but here we will focus on "information" and explain it.

First, the letter numbers for "information" and "information" are shown in Figure 3.
According to the JIS code, ``jo'' is number 1396, and ``information'' is number 1396.
is number 2527, so the collocation number N of "information" is N=(1396-1)*6392+2527=8919367 from equation (1) above.

Next, the key is determined by a modulus (MOD) of several predetermined integers. The number of keys and the value of the integer can be various, but here the number of keys is 4, and the integer is, for example (751 743
739 733), MOD (751) = remainder of (8919367 ÷ 751) = 491 MOD (743) = remainder of (8919367 ÷ 743) = 395 MOD (739) = (8919367 ÷ 739 ) remainder = 376 MOD (733) = (8919367÷733) remainder) = 223, we can obtain the four numbers (491 395 376 223).

If we were to use these as 4 keys, we would also need a type mark (required to distinguish between the 4 keys), so we would need to divide the 3008 first-level kanji into 4 ranges of 752 for each type. to enter,
Perform the following calculations.

1st key MOD (751) + 752 * 0 = 491 = "turtle" 2nd key MOD (743) + 752 * 1 = 1147 = "axis" 3rd key MOD (739) + 752 * 2 = 1880 = "favor" 4th Key MOD (733) + 752 * 3 = 2479 = "Me" The key value (491 1147 1880 2479) obtained from this is further regarded as a character number, and the result shown in Figure 3 is
Translated according to JIS code: "Turtle/axis/kei/hei"
becomes.

This final conversion to kanji is a device to treat the index in the same way as a conventional character index, and is not an essential problem. (For example, the simplest way is to express the type in the 1000's as (1491 2395 3376 4223).) In any case, the collocation "information" can be expressed using four keys (for example, four keys). (a combination of kanji) (processing 4).

Therefore, the bits corresponding to the data numbers of the bit strings of the collocation index of the four keys (kanji) of the collocation "information" mentioned above are set to logic "1" and registered in the collocation index, and similarly "storage", " 4 keys are also calculated for "Search" and registered in the collocation index (processing 5).

Next, the search of the collocation index will be described below.

FIG. 2 explains the operation during a search in this embodiment, taking the character string "information search" as an example of a search condition.

Also, in the actual data section, data number 123 "Information storage and retrieval" is included, as shown in the actual data section of Process 1 in Figure 1.
In addition, it is assumed that much data has already been stored, such as data number 456 "Tokai Television Information Search System."

When the character string "information search" is input as a search condition (process 1), first "information", "information", "search", "search" are input.
The bit strings of the four character indexes are ANDed with the same bit positions. By this calculation,
Assuming that a bit string in which only the 123rd and 456th bits are “1” is obtained, from this result, the data containing the four characters “information/information/search/retrieval” are the 123rd and 456th bits. It turns out that there are only two such data, and since it is impossible for other data to contain the character string "information search", it can be excluded from the search candidates (processing 2).

Next, we extract the three combinations of "information,""investigation," and "search" contained in the string "information search," and calculate the four keys for each using the calculation method used during registration as described above. do. As a result, the key for "information" is "kame/jiku/kei/hei", and by the same calculation, the key for "hoken" is "kan/ago/frozen/noodles".
The key for "search" is required to be "kutsu/sho/toro/he" (processing 3).

Therefore, a logical AND operation is performed between the bit string obtained as a result of the logical operation of the character index and the same bit position of the bit string of the conjunctive index of these 12 keys. By this operation, only the 456th bit is “1”
If it is obtained as a corrupted bit string, the only data that may contain the character string "information search" is number 456 (processing 4).

Therefore, if data No. 456 is read from the actual data section and confirmed that it actually contains the character string "information search", this data is found to match the search condition (processing 5).

In this way, in the present invention, by searching the collocation index together with the character index, the number of actual data candidates can be narrowed down to a much smaller number than when searching only with the character index.

Next, we will explain why it is possible to handle an extremely large number of combinations using a small number of collocation indexes. As mentioned above, there are about 40 million collocations consisting of two kanji of the first and second level, but for example, 350 to the power of 3 already exceeds 40 million, so there are three coordinate axes with a scale of about 350. By the value (key) of , it is possible to locate every collocation at a unique coordinate.

However, conversely, if a certain coordinate is "1", it does not necessarily mean that a corresponding collocation exists. This is because multiple collocations may interfere with each other and accidentally indicate the coordinates. This acts as search noise and reduces the accuracy of search results. In such a case, the occurrence of such search noise can be prevented to a considerable extent by doubling the scale and increasing the number of coordinate axes by one to have about 300 billion coordinates.

Data number 123 in the above example was eliminated as a result of the logical operation because data number 123 was "information storage and retrieval", so the eight keys corresponding to "information" and "retrieval" were excluded. Naturally, the 123rd bit of the collocation index bit string is “1”, but all four 123rd bits of the collocation index bit string of the four keys corresponding to “Hoken” are “1”. This is because the possibility is extremely small.

In addition, in the above example, we explained the case where the search condition is a single character string "information search", but it is not limited to this, and may include multiple character strings such as "information" and "search", or when the search condition is "information search". It is also possible to handle complex search conditions such as ``'' or ``search,'' which include any of multiple character strings, or even mixed types of strings. This can be easily realized by changing the logical operation contents.

In addition, in the above example, we explained collocations that are combinations of kanji, but it is also easy to create collocation indexes using the same method for combinations that are not kanji, such as combinations of kanji and hiragana, or combinations of katakana. realizable.

Further, in the above example, a combination of two characters was described, but it is also possible to apply to a combination of three or more characters.

Furthermore, in the above example, the number of collocation indexes between kanji is 4.
The number is 2966, which is the sum of the prime numbers (751 743 739 733), and when converted to characters, it falls within the first level range. In other words, over 40 million combinations of kanji are expressed within the range of characters at the first level, but the precision of logical operations depends on the number of coordinate axes (keys) and the range of possible values (key values). ), it is possible to easily increase the accuracy if necessary.

Note that although the above explanation has been made regarding information retrieval in Japanese, the present invention is more suitable for information retrieval in Chinese, which has a kanji structure.

(Effects of the Invention) According to the present invention, compared to narrowing down actual data to be candidates using only a character index as in the conventional method, by using a character index and a collocation index, it is possible to Actual data can be narrowed down to a smaller amount. Therefore, the time required to scan actual data can be shortened, and the search speed can be improved.

In addition, since the accuracy of this logical operation is extremely high, there is a possibility that search noise may occasionally enter and add unnecessary data to the candidate data, but data that matches the search conditions will never be missed. By utilizing this property, it is practically possible to omit scanning of the actual data. In that case, search results can be presented using only logical operations, which has the effect of significantly shortening the search time.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining the operation at the time of registering a character index and a compound word index according to an embodiment of the present invention. FIG. 2 is a diagram for explaining the operation at the time of search in this embodiment. FIG. 3 is a diagram showing how kanji are defined in the JIS code as a premise for explaining the present invention.

Claims (1)

[Claims] 1. In the information retrieval system, when registering information, a data number is added to the registered data, a collocation consisting of a combination of characters appearing in the registered data is extracted, and the collocation number of the collocation is In a collocation index that is converted into a plurality of coordinate values (key values) indicated as unique points in a multidimensional space consisting of a plurality of coordinate axes, and has the key value and a data number designation bit string corresponding to the key value, The collocation is registered by turning on the corresponding bits of multiple key values and data numbers, and when searching for information, the collocation is extracted from the combination of characters included in the string as the search condition, Converting the collocation number of the collocation into a plurality of coordinate values (key values) indicated as unique points in a multidimensional space consisting of a plurality of coordinate axes, and converting the collocation number into a bit string corresponding to the key value of the collocation to be searched in the collocation index. A collocation index in an information retrieval system, characterized in that a logical product is calculated for bits in the same position, and data including a collocation of character strings as the search condition is searched from the data number indicated by the resulting on-bit. Search method used. 2. A calculation method that sets a compound word number according to a kanji code table for the compound word, and calculates key values of multiple coordinate axes indicating unique points in a multidimensional space from the compound word number.
A MOD operation is performed to obtain a remainder when the compound word number is divided by a plurality of prime numbers, and a type mark is added to the result of the operation, thereby converting it into a single kanji character number and using it as the plurality of key values. A search method using a collocation index in an information search system according to claim 1.