JP2718062B2 - Learning type character string search device and control method of the device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial application field) The present invention relates to a character string search device capable of efficiently and quickly processing the contents of text information such as patents and academic documents.

(Prior art) Text information such as patents and academic literature is
(Automation) equipment is increasing steadily as technology advances. They are stored in the file memory, which increases in size one after another to be reused. However, there is not much progress in creating a database of file information that can retrieve those contents at high speed and extract necessary information. That is, the operation of adding a keyword matching the content to each record of the file information and rearranging the keywords so that the keyword search can be performed at a high speed depends on humans and is not easily achieved.
In addition, even if the contents match, there is a problem of variation in terms in which the keyword used in the search does not match the keyword added to the record.

Therefore, various techniques of real-time character string search have been studied to search for text information in a file memory at high speed without adding or rearranging keywords.

In general, a character string search device has a function of searching for where a search character string given as a keyword is included in text information. In order to realize this function, a character string collating means for comparing and collating each character string in the text information with a plurality of search character strings simultaneously is essential.

Therefore, the inventors of the present application have received in parallel an associative memory unit that stores a plurality of character strings of an arbitrary length and a character comparison result of the associative memory unit for each character of the input character string, and received the input character string and the stored character string. A patent application has been filed for an invention of a constituent symbol string identification device that enables character string collation with a sequence logic unit that compares the character arrangement (order) between characters. Although the names are different, the above devices are the basic elements of the character string search device.
(Japanese Patent Application Laid-Open No. Sho 60-211539 (Japanese Patent Application No. 59-068495),
No. 61-253536 (Japanese Patent Application No. 60-96213) Such a character string search device registers a keyword (search character string) to be searched, and then inputs a search target text character string. Generates a match signal when any one of the strings matches, and outputs which part of the text string matched which search string. Therefore, full-text search (full-text search) of text character string information without a keyword is possible.

In other words, without complicated work such as adding a keyword to text string information stored in a file,
A database of file information can be created.

However, the information search by the full-text search method is much longer than the in-desk search method in which only the keyword portion of the text string information to which a keyword is added is used to find the file storage position of the text string information that matches the search condition. Requires search time. I mean,
Even if the string matching speed reaches 10 million characters per second, if the length of the text string information becomes 1 billion characters (1 gigabyte) as an example, the search time will be as long as the entire text is searched.
It takes up to 100 seconds. If it takes only 100 seconds for each search with just one set of keywords (search string),
Many search requests cannot be received one after another.

(Problem to be Solved by the Invention) A keyword (search character string) included in the search condition of the conventional character string search device is registered in the character string collation means, and text information is input thereto, and which part of the text information ( Record) and find which keyword it contains, and then determine the record that matches the search criteria. Whether or not the search condition is satisfied is determined by software (program). For this reason, when searching for a large amount of text information in accordance with the full-text search method, the search time is too long to be practical.

Because this search time is too long, it was conceivable to register keywords included in a large number of search requests in the character string matching means in a single search of text information. If the operation of comparing the collation result information including the search result with a plurality of search condition expressions does not proceed in parallel, the response time until the end of the search only further increases. That is, when search requests with different search conditions arrive, without using the power of software, there is a problem that they can be sequentially processed at high speed and a matched record address cannot be output in a short time.

An object of the present invention is to provide a character string search technique that solves the above-described problem in a full-text search method. More specifically, if any of the keywords included in the search conditions have been used in the past search, immediately determine whether they satisfy the new search condition expression without the help of software. A learning type string that outputs a record address that matches the search conditions and searches the entire text information only when a keyword that has never been used is given, so that the average search time can be reduced. To provide search technology.

(Means for Solving the Problems) Accordingly, the present invention provides a character string collating means for storing a plurality of search character strings and determining which of the plurality of search character strings the input character string matches, A first storage unit for storing which search character string is included and an access address determined by a class code of the character string determined by the character string matching unit; and an output of the first storage unit according to a search condition logic between the input character strings. Search condition collating means for performing the logical processing of the above, encoding means connected thereto, and the storage address of a text record containing the collated character string are stored, and the access address is determined by the output of the encoding means.
A learning type character string search device, characterized by comprising a storage means, and a switch means for selecting a text character string as a search character string as an input character string of the search character string matching means; and storing a plurality of search character strings, String matching means for determining which of the plurality of search strings the input string matches;
A first storage means for storing which text record contains which search character string, an access address determined by a class code of the character string determined by the character string collating means, and a first storage means in accordance with a search condition logic between input character strings. Search condition matching means for performing logical processing of the output of the storage means, and
Encoding means for specifying a text record that satisfies a search condition; second storage means for storing an actual storage address of the text record; an access address determined by an output of the encoding means; and input characters for the character string matching means A control method for a learning-type character string search device, comprising: a switch means for selecting a search character string or a text character string as a string, wherein the character string collation means and first and second storage means respectively store past search characters. Initialize mode for initial setting of the column, string collation result and the storage address of text co-record. When there is a search string that does not match the search string in the new search string, the search string is matched with the string collation means. Enter a text string in the search string registration mode to add and register the text string, and enter a text record that contains the search string A text string search mode for writing to the first storage means and the second storage means, and inputting a search condition logic and a search character string to the search condition matching means and the character string matching means; A search condition matching mode in which the output of the storage means whose access address is determined by the search condition matching means is processed by the search condition matching means, and a search result output mode in which the storage address of the text record matching the search condition is output from the second storage means using the result. When there is a match between all new search strings and past search strings in the search string matching mode, the mode is switched to the search condition matching mode and the search result output mode, and the matched text record storage address is output. If there is no match in the search character string matching mode, the mode shifts to the additional registration mode, where the mismatched character string is registered in the character string matching means. Thereafter, the mode is shifted to the text character string search mode, the search result is written in the first storage means and the second storage means, and then the storage address of the text record matched in the search result output mode is output through the search condition matching mode. This is a control method of a learning-type character string search device characterized in that it is controlled in the following manner.

(Operation) According to the present invention, a search-type character string search is performed, in which a search character string is additionally registered instead of being re-registered each time a search is performed, and a search result obtained by searching the entire text information is stored for each search character string. Since the apparatus does not need to redo the search due to the difference in the search condition formula, the search time can be reduced, and as a result, the effect of increasing the number of receptions of search service requests (transactions) per minute is expected.

In the conventional method in which the search conditions are reset every time a search is performed, for example, if the search time for the entire text information is one minute,
Only one search request is accepted per minute. In the learning-type character string search method of the present invention, since the results of full-text searches for past search requests are stored, in most cases, the search results for new search requests can be output immediately without searching the full text information. become. In most cases, the search time is reduced to less than 1msec. Only when there is a search request that cannot output the search results immediately, a search character string is additionally registered and a full text information search is performed, and the results are added to the stored search result data. In the long term, the required number of full-text searches for new search requests gradually decreases. If the number is reduced to 1/10 or less, the number of search request acceptances per minute increases by 10 times or more.

(Embodiment) FIG. 1 shows an embodiment of the present invention and is also a basic configuration diagram. The character string matching means 110 stores the search character string 121 input via the switch means 120, and then accepts the text character string 122 coming from the text file device 130, and converts the character string equal to the stored search character string 121. A match signal is output when a record that includes the match is received, and an identification code of the search character string is output.

An address is designated in the first storage means 140 by the identification code output from the character string collation means 110, and is accessed by a match signal. It is assumed that the multi-bit memory of each address stores a mark bit indicating which of the matched records contains the keyword corresponding to the address. That is, if the mark bit at the i-th bit position is "1" corresponding to the matched record at address i, it indicates that the record includes the keyword corresponding to address i, and if it is "0", it does not.

The search condition matching means 150 is a part for processing mark bit data read from the first storage means 140 whose access address is determined by the identification code output from the character string matching means 110 according to a logical condition sequentially provided from the input terminal 151. is there.

The encoder 160 encodes the “1” position in the output of the search condition matching means 150. If the number of occurrences of “1” is more than one and encoding is to be performed in ascending order of number, the encoder 160 must have a priority encoding function. This is because the encoder 160 has a Manchester chain circuit. Can be realized simply by adding

The second storage means 170 is a part for storing the storage address of the matched record in the file device 130. Naturally, the access address is determined by the output of the encoder 160, and when there are a plurality of matching records, the text information is output in order from the storage address of the record at the beginning of the text information.

By adopting such a configuration, a new search character string is registered in the character string matching means 110, and thereafter, text information is input, and the character string matching result can be temporarily stored in the first storage means 140. It goes without saying that the switch means 12
By switching 0, the new search character string can also be compared with the old search character string in the character string matching unit 110. By storing the search results of the text information for the old search character string in the first and second storage means 140 and 170, if all of the new search character strings match the old search character string, the search of the entire text information is performed. Instead, the storage address of the match record can be output only by inputting the condition logic for search to the search condition matching means 150. The new search string is matched to the string only when the new search string does not match the old search string.
It is possible to additionally register in 110 and search the entire text information. Therefore, the more you do a search service,
The number of old search character strings registered in the character string matching unit 110 increases,
Most new search strings will now match the old search string. For this reason, it becomes possible to output a match record without performing a full-text search of text information, thereby improving performance. This shows the learning effect.

For this reason, the present invention is called a learning type character string search device.

FIG. 2 shows an embodiment of the first storage means, the search condition matching means and the second storage means, and also shows an example of the contents stored in the first storage means and the second storage means. Those having the same numbers as those in FIG. 1 have the same names as those in FIG.

The identification code of the search character string output from the character string matching means 110 is input to the X address recorder 141 of the first storage means 140 through the input terminal 119, and the Y address decoder 142 is connected to the address counter 143. Immediately before the start of the full text search for text information, the content of the address counter 143 is set to 0 (zero), and the flip-flop 144
The content is counted up by the falling signal from. The flip-flop 144 receives a match signal pulse from the character string collation means 110 through the input terminal 118, and changes the content to a high level. The same applies to two or more match pulses in one record. The input terminal 145 is supplied with a record clock signal of text information. Each time a record changes, the record clock signal (RCLK) lowers the contents of flip crop 144 to a low level. At that time, the content of the address counter 143 is increased by +1. Hereinafter, first, how the results of the text information full-text search are stored by the first and second storage units will be described.

The first storage means 140 is set to the writing mode during the text information full-text search, and the match signal and the identification code of the search character string are input from the character string collation means 110 through the input terminals 118 and 119. The first storage means uses the identification code.
140 access addresses are determined. Assuming that a match signal for the search character string B occurs for the first time in the 39th record, the second signal and the first column (the leftmost column) of the first storage means 140 are selected by the match signal, and are selected there. The mark bit “1” is written. In addition, the second storage means 170
In the first column, the address data of the matched record is written via the address decoder 171 and the R / W circuit 172. In other words, if the 39th record is the first matched record, 39 is the first match in the second storage means 170.
It is written to the address. When several search character strings match in one record, one of the first storage means 140
The mark bit “1” is written to several addresses in one column.

When the record of text information switches (input terminal 14
When a record clock signal is applied from 5), the contents of the address counter 143 are increased by one, so that the column on the right is selected by the Y address decoder 142. However, if no match signal is received from input terminal 118, no mark bit is written in that column. Even if the next record clock signal is applied to the flip-flop 144 and no match signal is generated thereafter, the flip-flop 144
Does not change, the content of the address counter 143 does not increase. Therefore, the Y address decoder 14
2 keeps selecting the same row.

After the 39th record where the match signal occurred, 47th, 48th, 91st, 140th, 275th, 302nd, ...
In this case, 39, 47, 48, 91, 140, 275, 302,... Are written in the second storage means 170 in order from address 1 via the address decoder 171 and the R / W circuit 172 as shown in the figure. The 39th record contains search strings B and C, the 47th record contains only search string A, the 48th record contains A and D, and similarly, the 91st record contains A, B If the 140th record contains C and D, the 275th record contains A and B, and the 302nd record contains A, B and D, the first storage means 140 Respectively,
0110 ‥ in the first row, 1000 ‥ in the second row, 1001 ‥ in the third row
Is 1110 ‥ in the fourth row, 0011 ‥ in the fifth row, and 1100 ‥ in the sixth row.
110 is written in the seventh column as 1101 ‥. As described above, the search result is divided into the mark bit and the address data of the match record and stored in the first and second storage means.

Next, the search condition matching means 150 will be described with reference to FIG. Conditional logic (eg, OR logic, NOR logic, AND logic, NAND logic, etc.) indicated by the search condition expression is input terminal 152.
, The logical operation unit 155 performs a logical operation between the content of the read signal register 146 of the first storage means 140 and the content of the processing result register 156, and leaves the result in the register 156. The processing function of the logical operation unit 155 changes according to the conditional logic given from the input terminal 152. It is assumed that the input terminal 151 is supplied with a set pulse to the register 156 on the result of the arithmetic processing at a timing with a time delay from the match signal supplied to the input terminal 118.

When a match signal pulse comes from the input terminal 118, a read signal of the first storage means 140 is set in the register 146, and the logical operation unit 155 starts logical operation on the contents of the registers 146 and 156. Priority encoder
Reference numeral 160 designates a selected address to the address decoder 171 of the second storage means 170 in accordance with the search result transfer pulse provided from the input terminal 161 when the last symbol of the search condition expression is input. When there are a plurality of matched record addresses, the selected addresses are output in ascending order.

For example, if the search condition expression is (A * + C) * D, it is converted to + A) *) + C) * D) =, and conditional logic such as * or + is input to the input terminal 152. . First, the character strings A, B, C, and D are input to the character string matching means 110, and the subsequent right parenthesis []] is input to the input terminal 151 as a set pulse to the register 156 on the result of the logical operation processing. The initial value of the register 156 is naturally all 0 (zero).

The search condition logic given to the input terminal 152 is +, *,
If + − and * −, the logical operation unit 155 functions as an OR gate, an AND gate, a NOR gate, and a NAND gate, respectively.

When the first search character string A is entered, the contents 0111011 # of the first row of the first storage means 140 are read out by the match signal pulse and set in the register 146. Simultaneously with the setting, the logical operation by the OR gate of the contents of the registers 146 and 156 is executed. When a set pulse is supplied from the input terminal 151, the result of the OR operation of the contents of 156 of the register 146 is set in the register 156.

Thereafter, when *-(NAND) is applied to the input terminal 152 and the second search character string B is input, the content 1001011 # of the second line of the first storage means 140 is set in the register 146. When the setting is completed, a logical AND operation of the knot (complement) of the contents of the register 146 and the contents of the register 156 is executed. At the next right parenthesis input, the operation result is stored in register 156.
Is set to Therefore, when + A) *) has been input, the contents of the register 156 become 0110000 ‥.

Next, when a conditional logic of + is specified from the input terminal 152 and the search character string C is input to the character string matching means 110, the first
The contents 1001100 # on the third line of the storage means 140 are stored in the register 146.
And the logical sum of the content and the content 0110000 of the register 156 is executed by the logical operation unit 155. When the right parenthesis is input, the operation result of the logical operation unit 155 is stored in the register 1
Set to 56. The contents of register 156 changes to 1111100 °.

Finally, the conditional logic of * is specified from the input terminal 152,
When the search character string D is input to the character string matching means 110, the contents 0010101 # are stored in the register 14 in the fourth line of the first storage means 140.
6 and its contents and the contents of register 156 1111100
The logical product with ‥ is executed by the logical operation unit 155. When the subsequent right parenthesis is input, the operation result of the operation unit 155 is set in the register 156. Its content is 0010100 ‥.

The result is that, when the last conditional logic is input, a transfer pulse is given to the input terminal 161 and converted to match record address data (47 and 91) via the priority encoder 160 and the second storage means 170. 4
Output in order of 7,91.

FIG. 2 shows that the result of full-text search of text information can be divided into first storage means 140 and second storage means 170 and stored so as to be associated with each search character string, and new search conditions are given. At the time, it was shown that the address data of the matched record could be output using the above search results. If record address data that satisfies the new search condition cannot be found, it is necessary to additionally register the mismatched search character string in the character string matching unit 110 and perform a full-text search of text information. Further, it is necessary to store the search result for the new search character string in the first storage means 140 and the second storage means 170. However, if the entire text information is searched after the new search character string is additionally registered in the character string collation unit 110, a search result for the entire search character string is obtained. This means that the storage of the search results in the first and second storage means is not added, but the latest one can be written. That is, even if the text information is changed or added, the text information is corrected at this time. It is practical that the search result is not simply added but rewritten with a new one every time a full-text search is performed, and the search time is not increased.

FIG. 3 is an explanatory diagram of one embodiment of the character string collating means.
As an example that seems to be the smartest, a case where a sequential logic circuit 320 and an encoder 330 are added to the output stage of the associative memory 310 is described. The associative memory 310 stores the search character string. That is, each character code of the search character string is stored in the address order from the top. In the associative memory in which the character code Xi of the _i- th address is stored at the i-th address, the time t of the input text character string is stored.
When the character code X (t) is equal to X _i , the comparator 316 on the _i -th row outputs y _i = “1”. Y _i = "0" is output when not equal to X _i.

Order in the logic circuit 320 array of flip-flop FF _i flag bits are connected by AND gates 325, i-th FF _i-1 content (flag bits) to the AND gate 325 and S _i-1 Output of the i-th row comparator 315 of the associative memory 310
y _i has been entered. Then, the output of the i-th AND gate 325 is set to FF _i , and the content S _i being 1 means that the text character string matches the i-th character of the search character string. FF ₁ is set to “1” when the first character of the search character string is included in the text character string and input. The length of the search string can determine whether a match string by the S _n When an n checks whether becomes "1".

As in the FIG. 3 can cope well when the search string is a variable length, delimiter flip-flop DF _i is introduced. If the start position of the search character string is set arbitrarily, the registration position of the search character string can be specified every fourth line because the number of inputs to the encoder 330 is too large. For this purpose, a delimiter signal for initial setting of a flag bit can be set in DF ₀ , DF ₄ , DF ₈ , DF ₁₂ and DF ₁₆ .

In the associative memory 310, for example, ABCD, CAT, DONKEY
The registration example of the three search character strings is shown. When the delimiter signal is set to be equal to, shorter than 4 characters, and longer than 4 characters, generally, "1" is set in the delimiter flip-flop of the start line and the end line of the search character string registration. "Is set. For clarity in the figure, the DF with "1" set is shaded. That is, DF ₀ and DF ₄ are set to “1” for ABCD, DF ₄ and DF ₇ are set to “1” for CAT, and “1” is set for DF ₈ and DF ₁₄ for DONKEY. Is set.

Delimiter flip-flops DF ₀ , DF ₄ ,
“1” of DF ₈ is a flag bit flip-flop FF ₀ , FF ₄ , F
Logically it is added to the output signal of the F _8. Thereby,
Initial setting of the flag bit is performed. Delimiter flip-flop DF ₄ , DF on the line indicating the end of the search string
₇ and DF ₁₄ are used to transmit whether or not the flag bits have reached FF ₄ , FF ₇ and FF ₁₄ to the encoder 330 through the OR gates OG ₁ , OG ₂ and OG ₄ . The fact that the flag bits reach FF ₄ , FF ₇ , and FF ₁₄ indicates that the text string matches the search string from beginning to end. Therefore, OR
Outputting “1” at any of the gates OF ₁ , OG ₂ , OG ₃ , and OG ₄ means that there is a character string match. here,
When DONKEY is registered, the matching result of 4 characters of DONK
OG ₃ is not output because DF ₁₂ is not set to “1”. That is, to from DF ₉ so as not to issue a match signal to DF ₁₃ "0" when they do not match until the end of DONKEY.

As described above, the character string matching circuit can be realized by adding the sequential logic circuit 320 having the delimiter flip-flop and the encoder 330 to the associative memory 310. The match signal is output from the output terminal 331 in the form of the OR of the outputs of the OR gates of OG ₁ , OG ₂ ,…, OG ₄ ,… and a class code indicating which search string the input text string matched is output. Terminal
Sent from 332. When each component of the text string is input to the associative memory 310, it is simultaneously compared with all the character codes of the search string stored therein, and a response result is output from the encoder 330 within the cycle of the input clock. You.

1 to 3, the associative memory 310, the first storage means 140, and the second storage means 17 in the character string collation means 110 are described.
0 assumes that the character code, mark bit of the search result, and the address data of the matched record can be written at high speed at any time, and that they can be read at high speed. Resulting in. Therefore, the contents of the associative memory 310 and the contents of the first and second storage means are stored in a part of the text file device 130, and when the power is turned on, the text file device 130
It is desirable that the stored contents be loaded into these memories. For this purpose, the learning-type character string search apparatus shown in FIG. 1 includes character string matching means 110, first storage means 140, and second storage means.
An initialization mode for initial setting of past search conditions and search result data must be prepared in 170.

In the initialization mode, first, the search character strings registered in the past are sequentially registered in the character string collating means 110 after reset, and then the mark bit data for each search character string is stored in the first storage means 140 in the first line. Are written in order from. The address counter 143 of the first storage means 140 in FIG. 2 is set so as to return to zero when counting up to the number of mark bits in each row, and sequentially writes data bits corresponding to past search results.

That is, the write data switch 14 in FIG.
7 sets the mark bit “1” from the input terminal 148 to the input terminal 149
This is a portion that receives data bits from the first storage unit 140 as a write signal. In the initialization mode, the switch falls toward the input terminal 149, and when searching for full text information, the switch falls toward the input terminal 148.

Writing of data relating to the record address matched in the past search to the second storage means 170 is performed from the input terminal 173 through the R / W (read / Write) circuit 172, and the write address is transmitted from the address counter 175 to the OR gate 176. Given through. The contents of the address counter 175 are incremented (increased by +1) by a write clock signal supplied from the input terminal 174.

As described above, the associative memory 310 of the character string matching unit 110,
The setting of the initial state (past search conditions and search results) in the first storage means 140 and the second storage means 170 is achieved.

In addition to the initialization mode, the learning-type character string search device shown in FIG. 1 performs character string collation on a new search character string when the new search character string does not match a past search character string (a character string registered in the associative memory 310). Search character string registration mode for additionally registering in the associative memory 310 of the means 110, and thereafter, the character string matching means
A text character string is input to the text file device 130 from the text file device 130, a text record including the search character string is marked, and the search result is written to the first storage means 140 and the second storage means 170 (or the storage content of the search result is (Rewrite) Text string search mode, search condition logic and search character string are input to search condition matching means 150 and character string matching means 110, respectively, and first storage means 140 whose access address is determined by the output of character string matching means 110 And a search result output mode in which the storage address of a text record matching the search condition is output from the second storage means 170 using the result. The operation in each mode is as described above.

With these operation modes, the learning-type character string search apparatus can greatly reduce the number of times of full-text search of text information using past search results. For that purpose, first, after the power is turned on, data is initialized in the memory in the initialization mode. Next, new search conditions are given in the search result output mode. The search character string at that time is not A, B, C or D, but is specifically given as Magnetic, Memory, Material or Thin-film, respectively, and the search condition expression is simply +
A) *) + C) * D) =, not + Magnetic) * ▲
▼) + Material) * Thin-film) = The condition logic is given to the search condition matching means 150, and the character string is given to the character string matching means 110, and the output of the first storage means 140 whose access address is determined by the output is processed by the logical operation unit 155. Then, it is set in the register 156 when the cut ")" symbol is input. When the = (equal) symbol is input, the contents of the register 156 are encoded, and the storage address (record address) of the match record is output from the second storage means 170 via the R / W circuit 172.

If the character string matching means 110 generates a mismatch signal when a new search condition is input in the search result output mode, the search character string is temporarily stored in the memory of the host computer and the contents of the register 146 are all After "1", the input of the next search character string is accepted, and if there is no mismatch, the search condition matching is continued as it is. Each time a mismatch signal is generated, the new search character string is temporarily saved in the memory of the host computer. Although it is possible to output the search result from the second storage means 170 at the end of the search condition collation, the mismatched search character string is ignored. If at least one new search character string matches a past search character string, a text record address including the character string is obtained, and a flash report on the search target can be provided.

Next, the search character string registration mode is set, and the mismatched search character string temporarily saved in the memory of the host computer is additionally registered in the character string matching unit 110. Search results in full-text search by mode are stored in first storage means 140 and second storage means 170
Can be written to. Thereafter, when returning to the search result output mode via the search condition matching mode, a record address matching the new search condition is obtained. By adopting such a control method, as long as the new search string matches the past search string, even if the search conditions are different, it is possible to output a record address that matches the new search conditions without performing a full-text search of text information. Become.

(Effect of the Invention) As described above, even if the search character string of the new search condition is equal to the past search character string, the search condition expression is different, so that the past search result cannot be used, and a new full-text search of text information is performed. According to the present invention, the problem that the reception speed of the search request does not increase due to the need for the search request can be easily solved.

In the learning-type character string search device, as the number of search request receptions increases, the more search character strings
Most of the search strings included in the new search conditions are string matching
It will match the search string already registered in 110. Therefore, assuming that 90% of the search character strings match (hit rate 90%), the full-text search only needs to be performed once out of 10 search requests.
It only needs to be done once for 0 search requests. If it took 1 minute to search the text information full text, the search request could only be received on the first floor per minute in the past, but according to the present invention, the search request was received 10 or 100 times per minute. It can be said that it will be possible. The hit rate depends on the contents of the search request condition, and cannot be said to be unconditionally high. However, a considerable effect is expected in the search of a personal text database.

[Brief description of the drawings]

FIG. 1 is a diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing an embodiment of the first storage means, search condition matching means and second storage means of the present invention, and FIG. FIG. 4 is an explanatory diagram of one embodiment of a character string collating unit. 1. In FIG. 1, 110... Character string matching means, 120... Character string switch, 140, 170... Storage means, 150.
19 Search character string identification code input terminal, 141, 142 Address decoder, 143 Address counter, 144
Flip flop, 145: Match record clock input terminal, 146: Read data register, 147 ...
Write data switch, 148 ... Mark bit input terminal, 149 ... Search result data input terminal, 151 ... Set pulse input terminal, 152 ... Search condition logic input terminal, 156 ...
Search condition collation result register, 160
Encoder, 161: Search result transfer pulse input terminal, 171: Address decoder, 172: R / W circuit, 1
73 Match record address data input / output terminal, 174
…… Address data R / W clock input terminal, 175 …… Address counter, 176 …… OR gate.

Claims (2)

(57) [Claims] A character string matching means for storing a plurality of search character strings and determining which of the plurality of search character strings the input character string matches, and storing which text record contains which search character string. A first storage means for determining an access address according to a class code of a character string determined by the character string matching means, and a search condition matching means for performing a logical process on an output of the first storage means according to a search condition logic between input character strings. Encoding means connected thereto, a second storage means for storing a storage address of a text record including a matched character string, and an access address determined by an output of the encoding means;
A learning-type character string search device, comprising: switch means for selecting a search character string or a text character string as an input character string of the search character string collation means. 2. A character string collating means for storing a plurality of search character strings and determining which of the plurality of search character strings the input character string matches, and storing which text record includes which search character string. A first storage unit whose access address is determined by a class code of the character string determined by the character string comparison unit; and a search condition matching unit that performs a logical process on an output of the first storage unit in accordance with a search condition logic between input character strings. Encoding means for specifying a text record that satisfies the search condition, storing the actual storage address of the text record, and determining the access address by the output of the encoding means; And a switch for selecting a search character string or a text character string as an input character string of the character string collating means. An initialization mode for initial setting the past search character string, the character string comparison result and the storage address of the text record in the character string matching means and the first and second storage means, respectively; When there is a search string that does not match the search string registration mode that adds the search string to the string matching means when there is a search string, and a text that contains a search string by entering a text string into the string matching means A text string search mode for marking a record and writing it to the first storage means and the second storage means;
A search condition matching mode in which the search condition logic and the search character string are input to the search condition matching means and the character string matching means, and the output of the storage means whose access address is determined by the output of the character string matching means is processed by the search condition matching means. A search result output mode for outputting a storage address of a text record matching the search condition from the second storage unit using the result,
When there is a match between all new search strings and past search strings in search string matching mode, control is switched to search condition matching mode and search result output mode to output the matched text record storage address If there is no match in the search character string matching mode, the process proceeds to the additional registration mode, in which the mismatched character string is registered in the character string matching unit, and thereafter, the process proceeds to the text string search mode to store the search result in the first storage unit. A control method for a learning-type character string search apparatus, characterized in that control is performed such that the storage address of a text record matched in a search result output mode is output via a search condition matching mode after writing in a second storage means.