JPH0752447B2 - Search method and device - Google Patents

Description

The present invention relates generally to a search method such as a method or apparatus for searching a text or a symbol string, and further defining and searching a unit of an arsenic including voice and images. In particular, the present invention relates to a high-speed symbol string and a search method and apparatus suitable for search processing.

[Conventional technology]

A conventional text search method, which is a typical application field of the present invention, will be described below.

Along with the shift to office automation, a database of document information is rapidly being developed, and the scale of the database is also increasing. Therefore, speeding up the database processing of document information is an important issue. One of the important processes is a text search process for searching for a specific character string called a pattern from character string data called a text. Therefore, this text
It is highly desirable to perform the search process at high speed.

In the past, several types of text search methods and devices have been proposed. For example, "Hardware Systems For Text Information Retrieve"
(Hollaar, LA: Hardware Systems Sor Text Insormat
ion Retrieval, ACM SIGIR 6th conf., 1983), each pattern is stored in a register in the form of an array.
The text is entered character by character from the beginning and the cellular array method that detects patterns and the finite automaton method are used to input the text character by character from the beginning and refer to the state transition table. A finite automaton method for detecting is known. However, these conventional methods are methods for inputting text one character at a time from the beginning, so if the text length is n characters, all n characters must be input, which becomes a bottleneck. It was impossible in principle to speed up.

On the other hand, as a software method for text search, the KMP method (KMP method (Kunuth, DE, etc .: Fast
Pattern Matching in Strings, SIAM J. Comput., Vol.
6, pp323-350, 1977)) and the BM method (BM method (Boye
r, RS, etc .: A Fast String Searching Algorithm, CAC
M, vol.20, pp.762-772, 1977)), but
These methods fetch text and patterns one character at a time, compare them, and perform various processing according to the results, which is not fast in terms of software and is not suitable for hardware. .

[Problems to be solved by the invention]

As described above, the conventional technique has an algorithmic problem that all n characters in the text length must be input one by one in order to detect the pattern.

An object of the present invention is to provide a search method and apparatus capable of detecting panto at high speed by reducing the number of characters or symbol strings of input text as much as possible.

[Means for solving problems]

The above-mentioned purpose is based on the combination of the state of the symbol string being searched and the one symbol or the number symbol in the newly input symbol sequence, and the state of the next symbol string and the one or several symbols to be newly input next. This is achieved by preparing a table (hereinafter referred to as a skip table) in which the address of is easily obtained.

The present invention is generally applicable to retrieval, and broadly includes, as the above-mentioned symbol string, characters, symbols, images, graphic patterns, voice patterns, etc. which are handled as a unit for retrieval.

[Action]

By using the above skip table, when executing the above-mentioned symbol string search, refer to the data stored in the skip table to input only the minimum necessary characters from the symbol string. Since the symbol string and the search can be realized only by itself, the speed can be significantly increased as compared with the method of inputting all the symbol strings.

〔Example〕

Hereinafter, embodiments of the text search method, which is a main application field of the present invention, will be described in detail. In the present invention,
In the text search method, m is the pattern length and n is the text length. Also, P (j) (1 ≦ j ≦ m) is the j-th character in the pattern, and P (j: k) is the character string T (i) (1 ≦ i ≦ n) from the j-th to the k-th pattern. The text i
Let the th character, T (i: k), be the i-th to k-th character string of the text. Also, when performing a text search, the text is logically as shown in FIG.
Random access memory (RAM) in the form of table 300
Stored in.

First, a text search state (i, l) is defined. Here, i represents the address of one character of the text to be input next, and l represents the state in which the pattern matches l characters from the rear. This is illustrated in FIG.

Next, transition of the search state will be described with reference to FIGS. 4 and 5. When the search state is (i, l) and one character T (i) of the text whose address is i is equal to the m-1st character P (m-1) of the pattern, i = i-1, l =
1 + 1 (FIG. 4). If they are not equal, move the pattern before the (m-1) th pattern until a string that exactly matches the i-th to i + 1-th character string T (i: i + 1) of the text appears in the pattern. think of.
That is, the character string in the matching pattern is P (H-1:
H), it suffices to move the pattern by m−H and set the state 1 to 0, so that i = i + m + l−H and 1 = 0. .

The above-mentioned H is calculated by the following equation.

H (l, z) = max {x | (l + 1 ≦ x ≦ m−1 and P (x−1) = z and P (x−1 + 1: x) = P (m−1 + 1: m)) or (1 ≤x≤l-1 and P (1: x) = P (mx-1: m)) or x = 0} (0≤l≤m-1) This H (l, z) is determined only from the pattern. Is a function. As an example, considering the case where the pattern is “AABCAAABCAA”, H (1, C) is the position in the pattern where the character string “CA” is found first after the pattern (the end of the character string). Position) is "10". In the case of this example, H (l,
The table of z) is shown in FIG.

Next, a second example of a concrete flowchart of this method will be described.
Shown in the figure. First, the pattern length m is set to the text address i, and 0 is set to the state 1 (101). Next, it is determined whether or not i is less than or equal to the text length n. When i is greater than n, the text search is terminated, and if not, the search is continued (102).

Next, the text table is accessed using i as an argument, the i-th character T (i) of the text is input, and it is assigned to z. And the current state l and the entered character z
Of the table S (l,
z) is referenced and substituted for y (103).

Here, the table S (l, z) is Is. That is, it is the information on the difference between the address of the text to be input next and the address of the text input earlier. This is called a skip width, and this table S (l,
Let us call z) the skip table. In particular, one character z = T (i) of the current text is P
Note that it is "-1" when it is equal to (ml). As an example, FIG. 7 shows the skip table S (l, z) when the pattern is "AABCAAABCAA". Next, it is checked whether y = S (l, z) is "-1" (104). If it is "-1", it means that one character of the current text and one character of the pattern are equal, so when the current state 1 is not "m-1", the state 1 is counted up (106). , 107). The current state 1 is "m-
When it is 1 ", it is found that the text contains a pattern, so for example, the pattern discovery information OUT is set to" 1 ". Then, the state 1 is reset to" 0 ", and the skip width is further set. Substituting a constant I = 2m−F−1 for y (108), where F = max {x | (1 ≦ x ≦ m and = P (1: x) = P (m−x + 1: m)) or x = 0} Finally, since the skip width y = S (l, z) contains the difference between the address of the text to be input next and the address of the text previously input, Substitute i + y for the address i of the text to be input to and return to step 102 (109).

Next, an operation example of this system will be described with reference to FIG. As an example, the text is "AABCAAABABAABCBAABBAAABAA" (Fig. 8) and the pattern is "AABCAAABCAA". The case of this pattern and the skip table S (l, z) are shown in FIG. First, i = m = 11 and l = 0 are set. In the first cycle, the text T (11) = A is input and assigned to z. Further, the skip width S (l, z) = S (o, A)
Fetch = -1 and substitute into y. Then, since y = -1, the state 1 is counted up and l = 1, and the address i of the next text becomes i + y = 10. Two
In the cycle, the tenth text T (10) = B is input and substituted into z. Furthermore, skip width y = S (1, B)
= 11 is fetched. Then, since y ≠ -1, the state l
Becomes l = 0, and i = i + y = 21. The same operation is performed in the third cycle and the fourth cycle. In the fifth cycle, the text address i becomes 30, and n = 25 is exceeded, so the text search processing is completed. Thus, using this method, in this example, the text search process for 25-character text can be completed in 5 cycles. Therefore, it enables faster processing than the finite automaton method or the cellular array method, which requires inputting all 25 characters of text.

Next, FIG. 1 shows an example of a circuit that realizes the above method. 1
Is the RAM that stores the text table 300T (i),
2 stores the skip table 200S (l, z)
RAM, 6 is a text address register for accessing 1, 11 is a skip table address register for accessing 2, and 10 and 12 are 1 respectively
And memory registers corresponding to 2 and logically,
6 is text address i, 11 is skip table
The address (l, z), 10 corresponds to the text z = T (i), and 12 corresponds to the skip width y = S (l, z). Furthermore, 4
2 is a logic circuit that realizes 110 in the flowchart of the present system in FIG. 2, 5 is an adder circuit that realizes 109 part, and 3 is a determination circuit that realizes 102 part.
This operation example is similar to the operation example of the present system described above.

In the above embodiment, the case where the table is operated in units of one character has been described, but it is possible for those skilled in the art to easily operate the table in units of a plurality of characters as needed. Further, according to the present invention, a plurality of patterns can be detected by appropriately changing the content type and the number of the skip table 200.

Furthermore, the present invention can be implemented by any of the software means and the hardware means as described above, and can exert its effect.

Further, in the above-mentioned embodiment, the description has been focused on the case of the text symbol string, but as described above, the present invention can be applied to all the voice patterns, images / figures, etc. in which the search unit can be specified similarly to the above-mentioned symbols. Is.

〔The invention's effect〕

According to the present invention, when the pattern length is m symbols and the symbol string length is n symbols, in the minimum case, the process can be completed by only inputting a symbol string having a length of n / m. Compared with the conventional method that requires inputting all, it is possible to realize significant improvement in processing performance. In addition, hardware is also
It can be realized with a significantly smaller amount of material than the conventional method such as the array method.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of a text search device according to an embodiment of the present invention, and FIG. 2 is a text search device for realizing the present invention.
Flow chart of the search method, FIGS. 3, 4, and 5 are explanatory views of the method of the present invention, FIGS. 6 and 7 are diagrams showing examples of tables used in the present invention, and FIG. 8 is an example of text. FIG. 9 is an explanatory diagram of an operation example of the present invention. 1,2 ...... Random access memory, 3 ...... Decision circuit, 4 …… Logic circuit, 5 …… Adding circuit, 6,11 …… Address register, 10,12 …… Memory register, 200 …… Skip table, 300 ... text table.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masashi Tsuchida 1099 Ozenji, Aso-ku, Kawasaki-shi, Kanagawa Hitachi Ltd. System Development Laboratory (72) Inventor Nobuhiro Taniguchi 1st Horiyamashita, Hadano, Kanagawa (72) Inventor Yoshiaki Yamashita 1 Horiyamashita, Hadano City, Kanagawa Prefecture Hitate Manufacturing Co., Ltd. Kanagawa Factory

Claims (4)

[Claims] 1. A method for searching a symbol pattern for a predetermined pattern using a processing device, the method for detecting at least one of the predetermined patterns in the symbol string, From the state of the symbol string search of and the information of at least one symbol entered,
Create a table to store the address of the symbol string to be input next and the information to calculate the next state, and access the table from the current state and the information of the input symbol string according to the input of the symbol string. Then, the next state and the address of the symbol string to be input next are extracted from the data stored in the table, the symbol string of the address is then input, and the above process is repeated. . 2. In the table, information corresponding to a useless search is excluded based on information obtained as a result of an already-executed search and information on a newly input symbol string including at least one symbol. The search method according to item 1, wherein 3. In order to detect that at least one predetermined pattern is included in the text, the next input from the current text search state and the information of one character in the input text is performed. A table that stores the address of one character of the power text and information for calculating the next state,
In response to the input of one character of the text, means for accessing the table from the current state and the information of the one character of the input text, and the next state and the next input from the data stored in the table A search device comprising: a means for calculating a text address; and a means for repeating inputting one character of text corresponding to the address. 4. In the table, information corresponding to a useless search is excluded based on information obtained as a result of an already-executed search and information on a newly input symbol string consisting of at least one symbol. The retrieval device according to claim 3, further comprising: