JPH04293161A - Method and device for retrieving document - Google Patents

Abstract

PURPOSE:To obtain the concrete method and device for retrieving a document, which can decide a composite condition of a neighborhood condition, a context condition, a logical condition, etc., in a full text search. CONSTITUTION:By a device constituted of a character-string collating circuit 200 and a composite condition deciding circuit 300, a character-string collation of a document of a document data base and a designated retrieval word in a retrieval conditional expression is executed, and when the collation is executed, an identifier of the retrieval word collated with document discriminating information and collating information in the document are outputted as collating information. At the time of deciding a composite condition, a decision of a neighborhood condition is executed by checking the formation of a close distance condition between the retrieval words designated by the retrieval conditional expression, based on the collating information, the collating information and collating information of a result of decision are outputted as the whole collating information, and subsequently, a logical condition is decided by checking the common occurrence of the designated retrieval words in the same phrase, sentence, etc., based on the whole collating information, the whole collating information is outputted, and next, the logical condition is executed by checking a logical condition between the designated retrieval words, and the whole collating information obtained therein is outputted as final retrieval result information.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention relates to full text search in information processing systems, particularly information retrieval systems.
The present invention relates to a method and apparatus for quickly realizing complex condition determination processing such as neighborhood conditions, context conditions, and logical conditions. It can be used for searching in text databases, word processors, document filing systems, etc.

0002

BACKGROUND OF THE INVENTION In recent years, large-scale database services that include not only secondary information (bibliographic information) such as literature information and patent information, but also primary information (original texts) have become increasingly important. Conventionally, information searches in databases have been performed using secondary information such as keywords and classification codes controlled based on a thesaurus. However, this method only narrows down the results from a few dozen to a few hundred results, and there is an efficiency problem in that the searcher must read the text directly to confirm the content at the final stage. Furthermore, because the classification system itself changes over time, the problem arises that keywords and classification codes must be constantly updated. Furthermore, since assigning keywords (referred to as indexing) takes time, new documents are registered in bulk by batch processing. Therefore, there is a problem that the information to be searched always has a certain period of time.

[0003] As one method for dealing with these problems, a full-text search system is being considered in which a searcher can directly refer to the main text of a document and search for content based on a free search term. Several document search devices have been proposed to realize such a full-text search system. The configuration of a typical document retrieval device among them is shown in FIG. 2, and its contents will be explained. (L.A.
Horror: “Hardware Systems for Text Information Retrieval”, ACM, SIG
IR, 6th Conference 1983
Year, L. A. Hollaar:”Hardware
systems for TextInf
“Ormation Retrieval”, ACM
SIGIR6th Conference 19
83) In the document search device 1, the search control means 101 controls the entire search device and communicates with the host computer. That is, it receives a search request 201 sent from the host computer, analyzes it, and sends it as search information 202 to the character string matching means 200 and the complex condition determining means 300. Further, the search control means 101 controls the storage device control means 104 to read the document data 204 stored in the character string storage means 105 to the character string collation means 200. The character string matching means 200 searches the document data 204 for a search word (hereinafter referred to as a search term) specified in the search request.
It is checked whether there is a matching string, and if there is a matching string, information 205 identifying the matching string is output to the compound condition determining means 300. Complex condition determination means 300
checks whether the character string identification information 205 satisfies logical conditions such as AND and OR specified in the search request. If the compound condition is satisfied, the identification information and document contents of the corresponding document are displayed in the search results 20.
6 and returns it to the host computer.

[0004] In this system, in order to perform narrowing down with high precision, in addition to logical conditions, the following conditions targeting English sentences are proposed as search conditions for the compound condition determining means 300. “A.n.B”
(1-1) “<A, B>n”
(1-2) “A AN
D B IN SENT” (1-3) (1-1)
The conditional expression “A .n. B” in the expression is “A” and “B”.
”, which appear in this order, and where these two search terms appear close to each other within n words. The conditional expression is “A
” and “B” in any order, i.e. “A” appears before “B” or “B” appears before “A”. also represents searching for documents in which these search terms appear close to each other within n words.A search that measures the degree of proximity between search terms, as in equations (1-1) and (1-2). We will call this condition the neighborhood condition. “A AND B IN SE” in equation (1-3)
The conditional expression ``NT'' indicates that documents in which the two search terms ``A'' and ``B'' appear in the same sentence are found, regardless of their order. As in the expression (1-3), A condition for determining the co-occurrence of two search terms in the same context (also called a field) such as a sentence or a paragraph is called a context condition.

[0005] As described above, in this cited document, search conditions such as neighborhood conditions and context conditions, which are constrained by distance connections and contextual connections between search terms, are proposed as complex conditions. Using these conditions allows you to search by taking into account the semantic connections between keywords, compared to simply using logical conditions. This will allow you to narrow down your search. However, this cited document does not describe a specific method for realizing the neighborhood condition or context condition. In addition, full-text search requires an enormous amount of processing time because document data is directly searched. Therefore, in order to search for search terms at high speed, string matching hardware called a term comparator has been proposed. A concrete implementation method for this is disclosed in, for example, Japanese Patent Laid-Open No. 105039/1983. With this term comparator, several MBy
It is possible to perform character string matching at high speeds ranging from te/s to several tens of MByte/s. However, these term comparators only have the same functionality as the string matching means 200, and they do not have the same functionality as the string matching means 200, and they only have the same function as the string matching means 200, and they do not have the same functionality as the character string matching means 200, and they do not have the same functionality as the string matching means 200.
Composite condition judgment functions such as logical conditions and logical conditions are not included. Furthermore, the compound condition determination means 300 uses a large number of search terms (
(hereinafter referred to as a matching term) must be processed at the same high speed as the matching processing speed of the character string matching means 200. This is because, no matter how fast the character string matching means 200 can perform processing, if the processing of the complex condition determining means 300 is slow, the search speed as a system will drop, making it impossible to shorten the search time. Therefore, the compound condition determining means 300 must be capable of high-speed determination processing of neighborhood conditions, context conditions, and logical conditions.

[0006]

[Problems to be Solved by the Invention] The problem of the present invention is to make specific judgments of complex conditions such as neighborhood condition judgment, context condition judgment, and logical condition judgment, which enable fine-grained narrowing down that is unique to full-text search. In addition to providing a method,
It is an object of the present invention to provide a composite condition determination method that can perform these combination processes at a speed equivalent to that of a hardware string matching means.

[0007] The complex condition search function which is specifically attempted to be realized by the present invention is as follows. First, as a neighborhood condition, in the case of Japanese, a character distance condition search with an upper or lower limit specified for the number of characters that exist between search terms is used, and in the case of English, an upper or lower limit is specified for the number of words between search terms. This allows for inter-word conditional searches. Examples of character distance conditions include the following. “Document [8C] Search” (2-
1) “Document [10c] Search” (2
-2) “Document [8c, 10c] Search” (2-3
) “Document <10c> Search” (2-
4) The conditional expression "Document [8C] Search" in equation (2-1) means that the two search terms "document" and "search" appear in this order, and there are 8 characters between these two search terms. Indicates that the document containing the characters within the range is to be found. Therefore, in the example sentences shown in Figure 3, ■
You will be searching for and ■. The conditional expression “Document [10c] Search” in expression (2-2) uses the two search terms “Document” and “Search” regardless of their order, that is, “Document” comes before “Search”. ``Search'' or ``Search'' appears before ``Document'' indicates that documents in which these search terms appear closely within 10 characters are to be found. Therefore, in the example sentences shown in Figure 3, ■, ■, and ■
will be searched for. The conditional expression “document [8c, 10c] search” in equation (2-3) searches for documents in which the two search terms appear close to each other within 10 characters and are 8 or more characters apart, regardless of their order. It means to find out. Therefore, in the example sentence shown in FIG. 3, ■ and ■ will be searched. The conditional expression “document <10c> search” in formula (2-4) indicates that the two search terms “document” and “search” appear 10 or more characters apart, regardless of their order, to be found. . Therefore, in the example sentence shown in FIG. 3, ■ and ■ will be searched.

Next, examples of word distance conditions include the following. “text [8W] search” (3-1) “t
ext [10w] search” (3-2) “text
[8w, 10w] search” (3-3) “text
<10w>search” (3-4) “ of formula (3-1)
The conditional expression “text[8W]search” is “t
The two search terms “ext” and “search” appear in this order, and there are 8 search terms between these two search terms.
Indicates to search for documents containing the following words. The conditional expression “text[10w]search” in equation (3-2) means that the two search terms “text” and “search” can be used regardless of their order, that is, “t
Whether ``ext'' appears before ``search'' or ``search'' appears before ``text,'' the search term searches for documents in which these search terms appear closely within 10 words. "text[8w,10w]search" in equation (3-3)
The conditional expression “h” is “search” and “text”
This means searching for documents in which two search terms appear in close proximity within 10 words and 8 or more words apart, regardless of their order. (3-4) The conditional expression “text<10w>search” in the expression
This means finding documents in which the two search terms appear 10 or more words apart, regardless of their order. The above issues are issues regarding neighborhood conditions.

Next, as a context condition search, there are the following in both Japanese and English. “Document [P] search”, “text [P] search” (4-1) “Document [p] search”, “text [p] search” (4-2) “Document [S] search”, “text [S] search” (4-3) “Document [s] search”, “text [s] search” (4-4) “Document [PH] search”, “text [PH] search” (4-5) "documents[
ph] search”, “text[ph]search” (4-6)

001
0] Below, we will explain using a Japanese example. The conditional expression “document [P] search” in equation (4-1) means that the two search terms “document” and “search” appear in the same paragraph in this order. represent The conditional expression “document [p] search” in equation (4-2) indicates that the two search terms “document” and “search” search for documents that appear in the same paragraph, regardless of the order. . The conditional expression “document [S] search” in equation (4-3) means that the two search terms “document” and “search” appear in the same sentence in this order. represent The conditional expression “document [s] search” in equation (4-4) indicates that the two search terms “document” and “search” appear in the same sentence regardless of their order. . The conditional expression “document [PH] search” in equation (4-5) is
This indicates that the two search terms "document" and "search" are used in this order to find documents that appear in the same phrase. In Japanese, phrases are sentences separated by ",", ",", and ".". In English, this means sentences separated by "," and ".". The conditional expression “document [ph] search” in equation (4-6) is
This means that the two search terms "document" and "search" will search for documents that appear in the same phrase, regardless of the order. The above points are issues regarding the context conditions.

Finally, as logical conditions, there are the following in both Japanese and English. “Document [AND] search”, “text [AND] search” (5-1)“
“Document [OR] search”, “text [OR] search” (5-2)
“Document [NOT] search”, “text [NOT] search” (5-3)

[
[0012] The following will be explained using an example in Japanese. The conditional expression "document [AND] search" in equation (5-1) indicates that a document in which the two search terms "document" and "search" appear simultaneously is searched. The conditional expression “document [OR] search” in expression (5-2) is
It means to search for documents in which the search term "document" or "search" appears. The conditional expression “Document [NOT] search” in expression (5-3) means that the search term “Document” appears and “Search”
This means to search for documents in which the search term does not appear. The above is a problem as a logical condition.

[0013] To summarize these problems, the problem of the present invention is to specifically determine complex conditions such as neighborhood condition determination, context condition determination, and logical condition determination, which enables fine-grained narrowing down unique to full-text searches. The object of the present invention is to provide a complex condition determination method that can perform these combinations at a speed equivalent to that of hardware string matching means.

[0014]

Means for Solving the Problems In order to solve these problems, the method of the present invention includes a character string matching step and a complex condition determining step. In the string matching step, when a search term specified in a document is matched, the document identifier, which is the identifier of the document, the matched search term, that is, the identifier of the matching term, and the matching term in the document. Outputs the first character position and last character position of the text as matching information, and if a context condition is specified and a string identifying the context is matched, the identifier of the context identification string that was matched with the identifier of the document and the document The first character position and the last character position of the matching context identification character string in the text are output as matching information.

The complex condition determination step includes a neighborhood condition determination step, a context condition determination step, a logical condition determination step, or a combination of these steps. In the proximity condition determination step, the proximity distance condition expressed as the number of characters between the search terms specified in the search condition expression is determined based on the matching information output in the character string matching step, and the forward location that matches the condition is determined. The first character position of the located search term and the last character position of the subsequent search term are used as verification information of the determination result, and are added to the verification information output in the character string verification step and output.

[0016] In the context condition determination step, if the search condition expression includes a neighborhood condition, the search specified in the search condition expression is performed based on the matching information output in the neighborhood condition determination step. The co-occurrence conditions of terms in the same phrase, sentence, or same paragraph are determined, and the position of the first character of the preceding context identification string that matches the condition and the last character of the subsequent context identification string are determined. The position is used as verification information, and this is added to the verification information output in the neighborhood condition determination step and output. If the search condition expression does not include a neighborhood condition, the same phrase, same sentence, or the same search term specified in the search condition expression is determined based on the collation information output in the string matching step. The co-occurrence condition within one paragraph is determined, and the first character position of the preceding context identification character string that matches the condition and the last character position of the following context identification character string are used as collation information, and this is used as collation information. Output in addition to the collation information output in the column collation step.

In the logical condition determination step, if the search condition expression includes a neighborhood condition, the neighborhood condition and the context condition are determined in the search condition expression based on the collation information output in the neighborhood condition determination step. or if a context condition is included, based on the collation information output in the context condition determination step, and if only a logical condition is included in the search condition expression, the character is Based on the collation information output in the column collation step, the logical conditions between the search terms specified in the search condition expression are determined, and the collation information for each document that meets the conditions is output in the previous step. It is added to the matching information and output as the final search result information.

Further, one of the apparatuses of the present invention is composed of a character string matching means and a complex condition determining means as follows. When a search term specified in a document is matched, the string matching means identifies the document identifier that is the identifier of the document, the matched search term, that is, the identifier of the matching term, and the identifier of the matching term in the document. If the first character position and the last character position are output as matching information, a context condition is specified, and a string identifying the context is matched, the identifier of the context identifying string that was matched with the identifier of the document and the content of the document. The first character position and the last character position of the matching context identification character string in are output as matching information.

[0019] The compound condition determining means includes neighborhood condition determining means,
It consists of a context condition determining means and a logical condition determining means. The proximity condition determination means determines the proximity distance condition expressed by the number of characters between the search terms specified in the search condition expression based on the collation information output by the character string comparison means, and determines the proximity distance condition expressed by the number of characters between the search terms specified in the search condition expression, and determines the proximity distance condition expressed by the number of characters between the search terms specified in the search condition expression, The first character position of the search term and the last character position of the subsequent search term are used as verification information of the determination result, and are added to the verification information output by the character string verification means and output. The context condition determining means determines co-occurrence conditions of the search terms specified in the search condition expression in the same phrase, the same sentence, or the same paragraph based on the matching information output by the neighborhood condition determining means. The first character position of the preceding context identification character string that matches the condition and the last character position of the subsequent context identification character string that match the condition are used as matching information, and this is used as matching information output by the neighborhood condition determining means. Add and output. The logical condition determination means determines the logical conditions between the search terms specified in the search condition expression based on the collation information output by the context condition determination means, and determines the collation information for each document that matches the condition. is output as the final search result information.

[0020]

[Operation] In string matching, the document identifier, the identifier of the matching term, and the first and last character positions of the matching term in the document are output as matching information, and when a context condition is specified, the context Since the identifier of the identification string and the first and last character positions of the matching context identification string in the document are output as matching information, neighborhood conditions can be determined using the above matching for the search term in the search condition expression. The determination is made by checking the match with the identifier of the term, and the character distance condition in the search condition expression is determined by comparing the first character position and the last character position of each matching term. The context condition is determined by checking the match between the search term in the search condition expression and the identifier of the collation term obtained by character string matching and neighborhood condition judgment. Regarding the conditions for the range of co-occurrence, by comparing the position of the identifier in the preceding and following context identification strings with the first and last character positions of each matching term, This is done by determining whether the matching terms co-occur between the positions of the identifiers. Regarding logical conditions, the search term in the search condition expression is judged by checking the match with the identifier of the matching term obtained by character string matching, neighborhood condition judgment, and context condition judgment, and the matching is performed when a match is found. This is done by determining whether the term identifier satisfies the logical condition in the search condition expression.

[0021] By configuring the character string matching means and the compound condition judgment means, it is possible to consistently realize the judgment of compound conditions such as the above-mentioned neighborhood conditions, context conditions, and logical conditions, which is unique to full-text search. It becomes possible to perform a detailed search. Furthermore, for example, by having three microcomputers each execute the neighborhood condition determination process, the context condition determination process, and the logical condition determination process, it becomes possible to operate the processes without synchronizing each process. In other words, these microcomputers can perform pipeline processing in which condition judgment processing is started in response to collation information stored in each input buffer, making it possible to realize high-speed complex condition judgment processing. can.

[0022]

EXAMPLES First, the principles of the method and apparatus of the present invention will be explained. In the character string matching means, when document data is first input, a document identifier stored at the beginning of the document is detected and output as a matching result. Next, when the search term specified in the document is matched, the identifier of the matched matching term and the first and last character positions of the matching term are output as matching information as the matching position of the matching term in the document. be done. That is, as shown in FIG. 13, the collation information for one document has a document identifier at the beginning, followed by the collation information for the search term. The above process is repeated for each document until all document data has been read.

A specific search term matching method will be explained using FIG. 4. For example, it is assumed that the search term "document" is set in the character string matching means and the document "...This is a search system using document understanding..." is input. The document identification information and matching term identification information obtained as the output of the character string matching means in this case are expressed as (6-1) and (6-2). (D1, 0, 0) (6-1)
(T1, Xs, Xe) (6-2) In document identification information (6-1), D1 represents a document identifier,
The second term following this is a constant 0 (zero). In the collation term identification information (6-2), T1 represents the identifier of the search term (hereinafter referred to as collation term identifier), Xs represents the first character position of the collation term searched in the document, and Xe similarly represents the last character position of the collation term searched in the document. Represents character position. In the example of FIG. 4, the document identification information is (D1, 0, 0), and the collation term identification information of "document" is (T1, 31, 32).

Next, the composite condition determining means performs the following determining process. First, the proximity condition determining means determines the proximity distance condition expressed by the number of characters between search terms based on the collation information output by the character string collation means. In other words, the character distance between the last character position of the search term located in the front specified in the search condition expression and the first character position of the search term located in the back is calculated, and this character distance is calculated as the distance specified in the neighborhood condition and A determination is made as to whether the order is satisfied. If the proximity distance condition is satisfied, the determination result is the first character position of the search term located in the front that matches the condition and the last character position of the search term located in the back that meet the condition as matching information, which is used as matching information by the character string matching means. It is added to the verification information output in . A specific example of neighborhood condition processing will be explained using FIG. 5. For example, the neighborhood condition "Document [4C
]Understanding” is set, and the string matching method is set to document “. ．． ．． . This is a search system that uses document understanding. ．． ．． ．． ．． ” is input. First, the search terms “document” and “understanding” are set in the string matching method. When a document is input, string matching processing is performed for these two search terms. The following document identification information and matching term identification information are obtained. (D1, 0, 0) (6-
3) (T1, 31, 32) (6-
4) (T2, 33, 34) (6-
5) (6-3) is document identification information, (6-4) is “document”
and (6-5) is the verification term identification information of "understanding". Next, processing is performed on the neighborhood condition "document [4C] understanding" based on this information.

In this example, from 32, which is the last character position of the search term "document" located at the front that matches the condition, and from 33, which is the first character position of the search term "understanding" located at the back that matches the condition. The character distance is 0 ((33-32)-
1) = 1 - 1 = 0), and since it is smaller than the specified 4 characters, the neighborhood condition "Document [
4C] is determined to be satisfied.Finally, as a judgment result, the identifier PID of this neighborhood condition is set to P1, and the first character position of the search term "document" located ahead that matches the condition is 31. is Xs, and 34, which is the last character position of the search term "understanding" located later, is Xe (6-6) (hereinafter referred to as neighborhood condition identification information. In the case of context condition, context condition identification (In the case of information or logical conditions, it is called logical condition identification information, and this information is collectively called complex condition identification information.) is added to the matching information output by the character string matching means (6- 3), (6-4), and (6-5) and output. (D1, 0, 0) (6-
3) (T1, 31, 32) (6-
4) (T2, 33, 34) (6-
5) (P1, 31, 34) (6-
6) In other words, the judgment result of the neighborhood condition is (PID, Xs,
It is output in the same format as the matching information for the search term (Xe) in addition to the matching term identification information.

Next, the context condition determining means determines the co-occurrence condition of the search terms specified in the search condition expression in the same phrase, the same sentence, or the same paragraph based on the matching information output by the neighborhood condition determining means. Make a judgment regarding. In co-occurrence condition determination, two search terms are found in the order specified in the condition within the context range from the first character position of the context identification string specified in the condition to the last character position of the next context identification string. Determine whether it has appeared. If the co-occurrence condition is met, the determination result is the identifier of this context condition, the first character position of the preceding context identification string specified in the condition, and the last character position of the subsequent context identification string. This is added as verification information to the verification information output by the neighborhood condition determination means and output.

A specific example of context condition determination processing will be explained using FIG. 6. Here, “document” and “understanding” appear in this order, and the contextual condition “document [
S]Search" is set, and the string matching method is set to ". ．． ．． . This is a search system that uses document understanding. ．． ．． ．． ．． ” (document identifier = 1) is input. First, the search terms “document” and “search” are
Furthermore, since a context condition is specified, a character string "." for identifying the context is set in the character string matching means. In the string matching means, these three search terms are matched and
As shown in Figure 6 (D1, 0, 0) (6-7)
(S1, 30, 30) (6-8) (T
1,31,32) (6-9)(T3,
33,34) (6-10)(S1,4
8, 48) (6-11) is output. In the collation term identification information (6-8) and (6-11), S1 represents the identifier of the character string "." that identifies the context.

These collation information are sent to the neighborhood condition determination means, but in this example, since no neighborhood conditions are set, the input collation information is output as is from the neighborhood condition determination means as shown below. Ru. (D1, 0, 0) (6-7)
(S1, 30, 30) (6-8) (T
1,31,32) (6-9)(T3,
33,34) (6-10)(S1,4
8, 48) (6-11) (6-7) is document identification information, (6-9) is “document”, (6-10)
are (6-8) and (6) of the matching term identification information of “Search”.
-11) is the verification term identification information of “.”.

Next, a co-occurrence condition determination regarding the context condition "document [S] search" is performed based on this collation information. In this example, the context, or the range of sentences, is
From the 30th character, which is the first character position of the context identification character string "." specified in the condition, to the 48th character, which is the last character position of the next context identification character string ".", that is, (6-
8) to (6-11). In this example, within this context range, the search term specified in the condition “
"Document" and "Search" are included in this order, so
It is determined that “document [S] search” is established. Here, the order relationship between "document" and "search" is determined by comparing the last character position (32) of "document" and the first character position (40) of "search". In other words, since the last character position (32) of "Document" is smaller than the first character position (40) of "Search"(32<40), it is determined that "Document" is located before "Search". can.

Finally, as a result of the determination, the identifier CID of this context condition is set to C1, and 30 (start position information), which is the first character position of the context identification character string "." located at the front specified in the condition, and the rear The last character position 48 (end position information) of the context identification character string “.” located in
6-12) (hereinafter referred to as context condition identification information),
This is added to the collation information output by the neighborhood condition determination means and output as follows. (D1, 0, 0) (6-7)
(S1, 30, 30) (6-8) (T
1,31,32) (6-9)(T3,
33,34) (6-10)(S1,4
8,48) (6-11)(C1,30
,48) (6-12)

Finally, the logical condition determining means determines the logical conditions between the search terms specified in the search condition expression based on the collation information output by the context condition determining means, and determines whether the document specified in the condition is Output unit matching information as final search result information. Figure 7 shows an example of a specific logical condition determination process.
Explain using. For example, a logical condition "Document [AND] search" is set in which two search terms "document" and "search" appear in the same document, and the string matching means is "document"...Search using document understanding It is a system...
．． ．． ” is input. First, the search terms “document” and “search” are set in the string matching method, and when the document in the example in this figure is input, the string matching method returns the following: Verification term identification information is obtained. (D0, 0, 0) (6-13) (
T1, 31, 32) (6-14) (T3,
39, 40) (6-15) (6-13) is document identification information, (6-14) is “document”, (6-15
) is the collation term identification information for “search”. If the neighborhood condition and context condition are not set, this information is input to the logical condition determining means as is via the neighborhood condition determining means and the context condition determining means, and the logical condition "document [AND] A logical condition determination regarding "Search" is performed.

[0032] In this example, the logical condition determining means checks that the collation terms "document" and "search" exist simultaneously in one document, and determines that the logical condition "document [AND] search" is satisfied. Lower. Then, the document identification information (6-13) that satisfies this condition and the identifier LID of this logical condition as the result of this judgment are set as L1, and the first character position and the last character position of the corresponding document are used as matching information (6-16). This is added as (hereinafter referred to as logical condition identification information) to the output of the context condition determining means and output as follows. (D0, 0, 0) (6-13)(
T1, 31, 32) (6-14) (T3,
39,40) (6-15)(L1, 0
,99) (6-16)

As described above, the proximity condition determination means determines the proximity distance condition expressed by the number of characters between the search terms specified in the search condition expression, and the proximity condition determination means for determining the proximity distance condition expressed by the number of characters between the search terms specified in the search condition expression, , a complex condition judgment consisting of a context condition judgment means that judges the co-occurrence conditions in the same sentence or the same paragraph, and a logical condition judgment means that judges the logical conditions between search terms specified in the search condition expression. By using this method, it becomes possible to narrow down the search in detail, which is unique to full-text searches. Furthermore, since the input and output information formats of the neighborhood condition judgment processing, context condition judgment processing, and logical condition judgment processing that make up each complex condition judgment are exactly the same, distributing these and performing pipeline processing can speed up processing. Composite condition determination processing becomes possible.

Next, a first embodiment of the present invention will be explained using FIG. 1. In this embodiment, the character string matching circuit 2
00 and a complex condition determination circuit 300. The character string matching circuit 200 uses the search term of the search target sent from the search control means 101 (FIG. 2) and the character string storage means 105 (see FIG.
Compare the document data 204 read from Figure 2),
If there is a match, the match term information is shown as the match result 20.
5 to the complex condition determination circuit 300. The compound condition determination circuit 300 determines whether the compound condition is satisfied with respect to the matching term identification information output from the character string matching circuit 200 based on the compound condition sent from the search control means 101, and determines whether the compound condition is met. In this case, the judgment result 206 for each document is the corresponding matching term identification information and compound condition identification information.
Output as .

First, the character string matching circuit 200 will be explained in detail. The character string matching circuit 200 includes a term comparator 210, a document identifier detection circuit 220, a character count circuit 230, a position information addition circuit 800, and a search term length table 250. The term comparator 210 matches the specified search term with the sent document data 204, and if there is a match, it outputs a match term identifier 211 (positive integer value data) that is the identifier of the search term. , if there is no verified item, 0 (zero) is sent to the position information adding circuit 800. That is, as the value of the matching term identifier 211,
If it is 0, it is invalid data, and if it is a positive integer, it is valid data and represents the identifier of the collation term. As this term comparator 210, the one disclosed in Japanese Patent Laid-Open No. 60-105039 can be used. The document identifier detection circuit 220 has a register 224 as shown in FIG.
~228 and 229a, comparator 223, selector 2
It consists of 29. The register 224 is initially set with a top-of-text code (TOT) that is added to the beginning of document data for each document, and the register 229a is initially set to 0 (zero). Document data 204 sent in 8-bit units is sent to registers 225 to 228.
The signals are sequentially input to a four-stage shift register 228b consisting of. The comparator 223 uses the TO stored in the register 224 for this final stage output 228a.
A comparison is made to see if it is equal to T. The registers 225 to 228 send each output as a 32-bit output 222 to the selector 229, and when the final stage output 228a is TOT, the comparator 223 outputs the select signal 223.
a is sent, and the 32-bit output 222 is sent as the document identifier 22
1 and sends it to the position information addition circuit 800. Further, while the selector signal 223a is not sent, the selector 229 selects 0 stored in the register 229a. That is, the TOT added to the beginning of the document data for each document is detected from the document data, and the stored 32-bit document identifier 221 is sent to the position information addition circuit 800, and the TOT is sent to the position information addition circuit 800. While the document identifier 221 is not detected, a document identifier 221 of 0 is sent.

[0036] The character count circuit 230 counts the number of 8-bit character codes from the beginning of each document for the document data 204 sent, and converts it into the number of characters in which each character consists of 2 bytes. , and sends this to the position information adding circuit 800. The TOT detection signal 223a output from the document identifier detection circuit 220 is used to reset the character code count value for each document. As shown in FIG. 9, the search term length table 250 stores the length of the corresponding search term in the slot whose address is the matching term identifier 211, receives the matching term identifier from the location information adding circuit 800, and adds the location information. Additional circuit 80
The search term length 873 corresponding to the matching term identifier is returned to 0. In this figure, information is set such as 2, which is the term length of the search term "understanding" whose verification term identifier is 1, and 4, which is the term length of the search term "system" whose verification term identifier is 2. Therefore, for example, if 1 is received as the matching term identifier 211 corresponding to "search", 2 will be sent back as the search term length.

The position information adding circuit 800 includes registers 810 to 816 and OR gates 880 to 8 as shown in FIG.
81, selectors 820 to 822, a subtracter 830, and an adder 831. As an initial setting of this circuit, 0 (zero) is set in register 813, register 814, and register 815, and 0 (zero) is set in selector 820, selector 821, and selector 822, respectively.
is outputting. Further, 1 is set in the register 816, and 1 is output to the adder 831. Selectors 820 to 822 select the Z port when both select signals 890 and 891 are 0. In other words, each selector selects register 813, register 814, and register 815, respectively, and selectors 820 to 8
22, 0 is output as verification information 205. In addition, when the select signal 890 is 1 and the select signal 891 is 0, the X port is
If 0 is 0 and the select signal 891 is 1, the Y port is selected. The register 810 includes the document identifier detection circuit 22.
Each time a document identifier 221 is sent from 0, the document identifier 22
1 is stored and output to selector 820 and OR gate 880. Here, the document identifier 221 is 0.
In the case of , it indicates that something other than a document identifier was detected. When the document identifier 221 is stored in the register 810, the OR gate 880 calculates the logical sum between each bit of the document identifier 221, and sends a select signal 890, which is the result of the operation, to the selectors 820-822.

Therefore, when the document identifier 221 is stored in the register 810, the value of the document identifier 221 is 0.
(not zero), the operation result of OR gate 880 is 8.
90 becomes 1. In response to this, selectors 820 to 822
Since 1 is output to each selector, the X port is selected. The collation information 205 includes the register 81 as an identifier.
The document identifier 221 stored in 0 is the 0 (zero) stored in the register 814 as the start position information.
0 stored in register 813 as end position information
(zero) will be output as document identification information. Therefore, as shown in FIG. 11, the document identifier consists of a 32-bit document identifier and a 32-bit fixed value of 0 in which the start position information is 0 and the end position information is also 0.

The register 811 has a term comparator 2.
Each time a collation term identifier 211 is sent from 10, the collation term identifier 211 is stored, and this collation term identifier 211 is output to the selector 820, the OR gate 881, and the search term length table 250. Furthermore, from the search term length table 250, the matching term identifier 21
1, the search term length 873 is read out and the subtracter 83
Output to 0. Here, the matching term identifier 211 is 0.
If , it indicates that the search terms are not matched. The register 812 also includes a character count circuit 23.
Every time the character count 231 is sent from 0, the character count 231 is stored as the end position information 812a of the collation term, and is also sent to the subtracter 830 and selector 82.
2, and further subtracted by a subtractor 830 to output the end position information 81.
Subtract the search term length 873 from 2a, and add the adder 83.
The head position information 831a of the collation term incremented by 1 is output to the selector 821. When the collation term identifier 211 is stored in the register 811, the OR gate 881 performs the logical sum between each bit of the collation term identifier 211, and the select signal 891, which is the result of the operation, is sent to the selectors 820 to 881.
822.

Therefore, when the collation term identifier 211 is stored in the register 811, the value of the collation term identifier 211 is not 0, so the output of the OR gate 881 becomes 1. In response, the select signal 891 is sent as 1 to the selectors 820 to 822, so each selector selects the Y port, and the collation information 205 includes the collation term identifier 2 stored in the register 811 as an identifier.
11, the start position information 831a output from the adder 831 as the start position information, and the end position information 812a stored in the register 812 as the end position information are output as the collation term identification information. Therefore, as shown in FIG. 12, the collation term identification information is output as 32-bit collation position information consisting of a 32-bit collation term identifier, 16-bit head position information, and 16-bit tail position information.

As is clear from the above explanation, collation information can be expressed as (Tn, Xs, Xe). Here T
n represents a document identifier or a collation term identifier. Also,
Xs represents the head position information of the collation term, and Xe represents the tail position information of the collation term. Therefore, in the example of verification information (T2, 33, 34) in FIG. 4, T2 represents the verification term identifier, 33 represents the start position information of the verification term, and 34 represents the end position information of the verification term.

The term comparator 210 described above,
Document identifier detection circuit 220, character count circuit 230
, the search term length table 250, and the position information addition circuit 800, the character string matching circuit 200 outputs the information shown in FIG.
Verification information as shown in 3 will be output for each document. That is, for each document, the document identification information comes first, followed by the collation term identification information. In addition, the document identification information has the same structure as the collation term identification information,
The position information can be considered as verification term identification information of 0. Therefore, since it can be handled in the same way as collation term identification information, it is possible to perform batch processing without being conscious of the structure of document identification information. The above is a detailed explanation of the character string matching circuit 200.

Next, the operation of the character string matching circuit 200 will be explained using a specific example. Here, the search condition expression “Q=((Document [4C] Understanding) [S] System) [AN
D] (Document [S] search)" (7-1) will be explained as an example. In this example, the search terms are "T1: document" and "T2
: Understanding”, “T3: Search”, “T4: System” and “S1:. ” are sent from the search control means 101,
The search term is set in the term comparator 210, and the length of the search term is set in the search term length table. here,
T1, T2, T3, T4, and S1 represent matching term identifiers for the search terms "document", "understanding", "search", "system", and ".", respectively. “S1:.”
is for detecting "." as a context identification character string (hereinafter referred to as a context marker) of a sentence in response to the context condition "[S]", that is, the specification of a sentence as a context. .

[0044] As for documents, “...It is a retrieval system that uses document understanding.”
..." (7-2) is input. The document identifier is D1. When this document data is input to the character string matching circuit 200, the matching result 205 is the following matching shown in FIG. Information is output. (D1, 0, 0) (8-1) (S1,
30, 30) (8-2) (T1, 31, 32)
(8-3) (T2, 33, 34) (8
-4) (T3, 39, 40) (8-5) (T4
, 41, 44) (8-6) (S1, 48, 48
) (8-7) (8-1) represents document identification information. In the document identification information (8-1), D1 represents a document identifier, and the following two terms are constants 0. (8-2
) and (8-7) are for the context marker ".", (8-3) is for "document", (8-4) is for "understanding", and (8-5) is for "."
"Search" and (8-6) represent the collation term identification information of "System". Also, S1 is ". ”, T1
is for “document”, T2 is for “understanding”, T3 is for “search”,
and T4 represents the collation term identifier of "system". Verification information 205 for these (8-1) to (8-7)
will be sent as an input to the composite determination circuit 300.

Next, the condition judgment processing of the complex condition judgment circuit 300 will be explained. The complex condition determination circuit 300 is shown in FIG.
As shown in the figure, the MPUa of the three microcomputers
301, MPUb302, and MPUc303. The microcomputer MPUa301 uses a neighborhood condition judgment program 310, and the microcomputer MPUb302 uses a context condition judgment program 320.
However, a logic condition determination program 330 is executed in the microcomputer MPUc303. Furthermore, between each MPU there are buffers 350, 360, and 37 that use first-in, first-out (FIFO) memory.
0 is placed and used for data transfer between each MPU.

First, the determination processing of the neighborhood condition determination program 310 will be explained. Neighborhood condition determination program 31
0, the matching information 205 sent from the character string matching circuit 200 to the buffer 350 is read out, and the search information 202 is read out.
Determine whether the specified neighborhood condition is met. As an example of the neighborhood condition, in (7-1) there is “Document [4C
] Understanding”. The conditional expression “Document [4C] Understanding” requires that the two search terms “document” and “understanding” appear in this order, and that these two search terms are within 4 characters. It means to search for documents that appear in close proximity.Here, we consider the search expression as a type of arithmetic expression, and use “
"Document" is called the forward operand Ta, "search" is called the backward operand Tb, and "[4C]" is called the operation. Furthermore, the identifier representing this neighborhood condition is called "Pi". Note that Pi is the collation term. A code that can be distinguished from the identifier is assigned.By defining in this way, the neighborhood condition can be written as "Pi:Ta[nC]Tb".The following explanation will be made using this definition.

As mentioned above, there are other operations for neighborhood conditions such as "Pi:Ta[nC,mC]Tb""Pi:Ta<nC>Tb""Pi:Ta[nc]Tb""Pi: Examples include "Ta[nc,mc]Tb" and "Pi:Ta<nc>Tb". Figure 15 shows the procedure for processing this neighborhood condition.
This will be explained in detail using . First, iterative processing step 1
At 000, all collation information 205 in the buffer 350 is
In other words, 1 until the verification information of the last document is read out.
Repeat the processing steps from 001 to 1010.

In the verification information reading processing step 1001, one verification information 205 is read from the buffer 350, and
Output to buffer 360. In the verification information identification processing step 1002, the verification information reading processing step 100 is performed.
Is the collation information 205 read in step 2 document identification information?
Or check if it is collation term identification information. That is, if the lower 32 bits of the collation information are 0 (zero), it is determined that it is a document identifier. If it is determined that it is a document identifier, initialization processing step 1004 for each document is executed. Here, the forward operand buffer 311 used as a work area is cleared to 0 (zero). If the verification information 205 is verification term identification information, backward operand identification processing step 1003 is executed. In backward operand identification processing step 1003, the collation term identifier of the collation term identification information is checked to determine whether or not the collation term is specified as a rear operand in the neighborhood condition. In the case of a backward operand, the distance from the collation term identification information stored in the forward operand buffer 311, which will be described later, is determined, and it is determined whether a specified neighborhood condition is satisfied.

[0049] Hereinafter, the collation term identification information regarding the search term specified as the forward operand will be referred to as forward collation term identification information, and the collation term identification information regarding the search term specified as the rear operand will be referred to as backward collation term identification information. do. That is, in the forward operand buffer iterative processing step 1005, the proximity condition between the forward matching term identification information stored in the forward operand buffer 311 and the backward matching term identification information designated as the backward operand is determined. In this proximity condition determination, first, forward matching term identification information reading processing step 1
At step 006, forward matching term identification information is read from the forward operand buffer 311. Next, in proximity condition determination processing step 1007, the position information of the read forward matching term identification information and backward matching term identification information are compared to determine whether or not the specified proximity condition is satisfied. If the condition is satisfied, the determination result is output to the buffer 360 and the forward operand buffer 311 as collation information.

Forward operand identification processing step 1009
In this case, when the collation term identifier of the collation term identification information is specified as a forward operand, this collation term identification information including the one specified as a rear operand is output to the forward operand buffer 311. For example, this can be expressed as “Pi:Ta[nC]Tb” as a neighborhood condition.
When "Pj:Tb[nC]Tc" is specified, that is, when Tb is specified as both a backward operand and a forward operand, the backward operand identification process step 1003 and the forward operand identification process are performed for Tb. This is because both processes in process step 1009 are required. For this purpose, a backward operand identification processing step 1003 and a forward operand identification processing step 1009 are performed.
will be processed separately. By repeatedly performing each of the above processing steps on the collation information stored in the buffer 350, the neighborhood condition determination process can be realized.

The above processing procedure will be explained using a specific example. The search condition is the formula “Q=((document[
4C] Understanding) [S] System) [AND] (Document [S]
This will be explained using "Search)" as an example. Each condition determination program is set with a conditional expression that is analyzed by the search control means 101 and separated into each condition. The neighborhood condition part “Document [4C] Understanding” of conditional expression (7-1) is the neighborhood condition identifier P1 and the search term identifiers T1 and T that correspond to “Document” and “Understanding”.
2, it is given in the form "P1:T1[4C]T2". The document shown in (7-2) is a search system that uses document understanding.
..." is input, the following (8-1) to (8-7) are output from the character string matching circuit 200 as matching information to the buffer 350 as described above. These matching information are shown in the figure. 17.

(D1, 0, 0) (8-
1) (S1, 30, 30) (8-2) (T1,
31, 32) (8-3) (T2, 33, 34)
(8-4) (T3, 39, 40) (8
-5) (T4, 41, 44) (8-6) (S1
, 48, 48) (8-7) (8-1) represents document identification information. D1 is a document identifier. (8-
2) and (8-7) represent verification term identification information of ".". Also, (8-3), (8-4), (8-5),
and (8-6) are “document”, “understanding”, “
"Search" and "System" represent the matching term identification information.Furthermore, S1, T1, T2, T3, and T4 are "," respectively. ”, “document”, “understanding”, “search”, and “system”. The neighborhood condition determination process under the above conditions will be explained step by step using FIG. 15. In the initial state As shown in the initial state of FIGS. 16a and 16b, collation information (8-1) to (8-7) is stored in the buffer 350, and the forward operand buffer 311 and buffer 360 are cleared to 0. There is.

[0053] The neighborhood condition determination processing program reads these collation information 205 one by one from the buffer 350,
A determination process is performed regarding the neighborhood condition "P1:T1[4C]T2". First, as step 1, read processing step 1
001 is executed and collation information (8-1), that is, (D1
, 0, 0) is read into the work area of the program as shown in step 1 in FIG. Next, verification information identification processing step 1002 is executed, and verification information (D1, 0, 0
) is checked to see if it is document identification information. Verification information (D
1, 0, 0) is determined to be document identification information because the last two terms are both 0 (zero), that is, the lower 32 bits are 0 (zero). Therefore, initialization processing step 100
4 is executed, and the forward operand buffer 311, which is an internal work area, is cleared to zero. Thereafter, as step 2, the collation information reading process step 1001 is executed again, and the collation information (8-2), that is, (S1, 30, 3
0) is read and similarly output to the buffer 360 as is. Next, verification information identification processing step 1002
is executed, and it is checked whether the verification information (8-2) is document identification information or verification term identification information. Verification information (8-
2), that is, (S1, 30, 30), since the lower 32 bits are not 0 (zero), it is determined to be collation term identification information rather than document identification information. Then, in the next backward operand identification processing step 1003, it is checked whether this verification term identification information corresponds to the backward operand specified in the neighborhood condition.

Neighborhood condition “P1:T1[4C]T2”
The search term specified as the rear operand is T2, and this matching term S1 does not correspond to the rear operand, so the processes 1005 to 1008 are not performed, that is, the neighborhood condition determination process is not performed, and the next front operand identification Processing step 1009 will be executed. In this processing step, it is checked whether the collation term identification information (8-2) corresponds to the one specified as the forward operand of the neighborhood condition "P1:T1[4C]T2". Since this collation term is S1 and does not correspond to the forward operand, processing step 1010 is not executed, that is, the processing ends without being stored in the forward operand buffer 311.

As step 3, the reading process step 1001 is executed by the iterative process 1000, and the third collation information (8-3), that is, (T1, 31, 32) is read and stored in the buffer 360 in the same way as the second input. Output to. At the same time, collation information identification processing step 1002 is executed. Whether it is document identification information or collation term identification information can be checked. Since (8-3) is collation term identification information, backward operand identification processing step 1003 is executed, and since it is not a backward operand, forward operand identification processing step 10 is executed.
09 is executed. The matching term identification information (8-3), that is, (T1, 31, 32) is based on the neighborhood condition “P1:T1[
4C]T2” is specified as the forward operand, so
Forward operand storage processing step 1010 is executed,
The data is stored in the forward operand buffer as shown in step 3 of FIG.

[0056] Also, as step 4, iterative processing 100
0, reading processing step 1001 is executed, and step 4
th verification information (8-4), i.e. (T2, 33, 34
) is read, and matching term identification information (8-4) is output to the buffer 360 in the same way as the third input. Next, a verification information identification processing step 1002 is executed to check whether the information is document identification information or verification term identification information. Since (8-4) is collation term identification information, backward operand identification processing step 1003 is executed. (8-4) That is, (T2
, 33, 34) is the neighborhood condition “P1:T1[4C]T2”
Since it is specified as the rear operand of , the proximity condition determination processing from processing steps 1005 to 1008 is executed. First, the forward operand buffer reading processing step 1006 is executed, and the forward collation term identification information (8-3) stored in the forward operand buffer 311 is read. Next, proximity condition determination processing step 1007 is executed to check the character distance between the forward matching term identification information (8-3) and the backward matching term identification information (8-4). (8-3) That is, the end position X of (T1, 31, 32)
e is 32, (8-4) or (T2, 33, 3
Since the starting position of 4) is 33, the character distance between them is 0, and the specified condition of 4 characters or less is satisfied. Therefore, judgment result output processing step 1008 is executed, and as a judgment result, 31, which is the starting position of (8-3), is set as the starting position, 34, which is the ending position of (8-4), is set as the ending position, and the matching term identifier Neighborhood condition identification information (P1,
31, 34) (FIG. 17) are output to the forward operand buffer 311 and buffer 360 as in step 4 of FIGS. 16a and 16b. Further, in the iterative process 1000, a reading process step 1001 is executed, and the fifth collation information (8-5), that is, (T3, 39, 40) is read and sent to the buffer 360 (8-5) in the same way as the fourth input. 5
) is output. Next, verification information identification processing step 100
2 is executed to check whether the information is document identification information or verification term identification information. Since the matching term identification information (8-5) is matching identification information that is not set in the neighborhood condition, the process moves to the next input without performing either backward operand identification processing step 1003 or forward operand identification processing step 1009. . After that, the final verification term information (8-7
), the neighborhood condition determination process is repeated in the same way.

As a result of the above-described neighborhood condition determination processing, the collation term identification information (9-1) to (9-8) shown below (FIG. 17) is output to the buffer 360. (D1, 0, 0) (9-1) (S1,
30, 30) (9-2) (T1, 31, 32)
(9-3) (T2, 33, 34) (9
-4) (P1, 31, 34) (9-5) (T3
, 39, 40) (9-6) (T4, 41, 44
) (9-7) (S1, 48, 48) (
9-8) What should be noted here is the neighborhood condition judgment result (9
-5) is also sorted and stored in ascending order of the end position information as verification term identification information. Therefore, as will be described later, when determining a context condition, it is not necessary to check inclusion relationships for all combinations of contexts and collation terms, and processing can be reduced. These collation term identification information will be sent to the context condition determination program 320.

Next, the determination processing of the context condition determination program 320 will be explained. Context condition determination program 32
0, the buffer 3 is sent from the neighborhood condition determination program 310.
The verification information sent to 60 is read out, and the search information 20 is retrieved.
It is determined whether the context condition specified as 2 is met. An example of the context condition is "document [S] search" as shown in (7-1). The conditional expression "document [S] search" indicates that the two search terms "document" and "search" appear in the same sentence in this order to find a document. Here, an identifier representing this context condition (hereinafter referred to as a context condition identifier) is assumed to be Ci. C
A code that can be distinguished from the collation term identifier is assigned to i. By defining this, the context condition becomes “C
i:Ta[S]Tb". The following explanation will be made using this definition. In addition to this, context conditions include "Ci:Ta[P]Tb", "Ci :Ta[PH]Tb", "Ci:Ta[p]Tb", "Ci:Ta[s]Tb", "Ci:Ta[ph]Tb", etc.

The principle of context condition determination will be explained using the output example of the neighborhood condition determination process shown in FIG. Assuming that the document shown in (7-2) "...This is a retrieval system using document understanding..." is input, the neighborhood condition determination program 310 returns the main text as described above. The following (9-1) to (9-8) shown in the figure are passed through the buffer 360 as collation information to the context condition determination processing program 320.
sent to. (D1, 0, 0) (9-1) (S1,
30, 30) (9-2) (T1, 31, 32)
(9-3) (T2, 33, 34) (9
-4) (P1, 31, 34) (9-5) (T3
, 39, 40) (9-6) (T4, 41, 44
) (9-7) (S1, 48, 48) (
9-8) (9-1) represents document identification information. D1
represents a document identifier, and the following two terms are constants 0. (9-2) and (9-8) represent collation term identification information of the context marker ".". Similarly, (9-3), (9
-4), (9-6), and (9-7) are “
It represents the verification term identification information of "document", "understanding", "search" and "system".Here, S1, T1, T2, T
3, and T4 represent matching term identifiers of ".", "document", "understanding", "search", and "system", respectively. Further, (9-5) represents the neighborhood condition identification information when the neighborhood condition "understand document [4C]" is verified. P1 is a neighborhood condition identifier of the neighborhood condition "document [4C] understanding".

Based on the above collation information, first, the context condition "Ci:Ta" which specifies the order of collation term identifiers is
[S] Executes co-occurrence determination regarding "Tb".Here, the identifier Ta is called the 0th identifier, and the identifier Tb is called the 1st identifier.In the co-occurrence determination, all identifiers specified in the context condition are It is determined whether or not it has appeared in the context.To determine the success or failure of the co-occurrence determination, a co-occurrence counter is used here.One co-occurrence counter is provided for each specified context condition. .A method of controlling the co-occurrence counter under a context condition in which the order is specified will be explained below.Context condition "Ci:Ta[S]Tb"
The range of the context, ie, sentence, is from the matching information (9-2) of the appearing context marker "." to the matching information (9-8) of the next appearing context marker ".". Therefore, (9-2) to (9-8), that is, (
S1, 30, 30) to (S1, 48, 48) are the range of sentences. Therefore, (9-2) or (
Check in order from S1, 30, 30). context marker “
. ” appears, the co-occurrence counter is reset to 0 (zero), and then it is checked whether the matching information specified in the context condition appears. First, if the 0th identifier Ta specified in the context condition is Do not change the value of the co-occurrence counter until it appears.In other words, when the value of the co-occurrence counter is 0, focus on the 0th identifier Ta and monitor the appearance of this identifier.Here, the collation information of identifier Ta appears. If so, the co-occurrence counter is counted up. That is, the value of the co-occurrence counter is changed from 0 to 1.

Next, since the value of the co-occurrence counter shows 1, the appearance of the first identifier Tb specified in the context condition is monitored, and when it appears, the co-occurrence counter is counted up. If collation information of the identifier Tb appears here, the co-occurrence counter is counted up. That is, the value of the co-occurrence counter is updated from 1 to 2. When the value of the co-occurrence counter reaches 2, it means that all the identifiers specified in this context condition have appeared, so it can be determined that the co-occurrence condition has been met. By controlling the co-occurrence counter in this way, co-occurrence determination is performed under context conditions in which the order is specified. The co-occurrence counter is determined when the next context marker appears, as described below. Furthermore, collation information (9-8) of the context marker ".", that is, (S1, 48, 48) appears next. The value of the co-occurrence counter at this point is 2. Co-occurrence counter value is 2
This means that the identifiers of the two search terms specified in the context condition appear and the co-occurrence condition is satisfied. At this time, context condition identification information of the satisfied context condition is output. The start position information of this context condition identification information is set to 30, which is the start position information of the context marker "." that appeared before, and the end position information is set as the end position information of the context marker "." that appeared later. Set 48. In addition, the identifier of the context condition identifier “
In other words, the context condition identification information (Ci
, 30, 48).

Next, a method of controlling the co-occurrence counter in the case of the context condition "Cj:Ta[s]Tb" regardless of the order of collation term identifiers will be described below. The context range is similarly (9-2) to (9-8), i.e. (S1, 30,
30) to (S1, 48, 48) is the range of sentences. Therefore, similarly (9-2) or (S1, 30
, 30). First, the context marker “
. ” appears, the co-occurrence counter is reset to 0 (zero) in the same way. Next, it is checked whether the matching information specified in the context condition appears. First, if matching information for the identifier “Tb” appears , the co-occurrence counter is counted up.In other words, the value of the co-occurrence counter is updated from 0 to 1.After this, if the collation information of the identifier "Tb" that has already appeared appears again in the same context, the co-occurrence counter is updated. The value of the co-occurrence counter is not changed.Then, when collation information of the identifier "Ta" appears, the co-occurrence counter is counted up.In other words, the value of the co-occurrence counter is updated from 1 to 2.Here, the value of the co-occurrence counter is 2, all the two identifiers specified in this context condition have appeared, so it can be determined that the co-occurrence condition has been met.After this, the identifier “T” that has already appeared
Even if the collation information for "a" appears again in the same context, the value of the co-occurrence counter is not changed.Furthermore, the next context marker "a" appears again in the same context. ” verification information (9-8), i.e. (S1, 48, 4
8) appears, the co-occurrence counter is already 2, that is, the co-occurrence condition is satisfied, so the context condition identification information of this context condition is output. In other words, the context condition identification information (Cj
, 30, 48). By controlling the co-occurrence counter in this way, it can also be used to determine co-occurrence of context conditions regardless of order. By controlling the co-occurrence counter as described above, it is possible to determine whether the co-occurrence determination is successful or not.

The procedure for processing context conditions will be explained in detail using FIG. 19. First, iterative processing step 1100
1101 to 1 until all the collation information in the buffer 360, that is, the collation information of the last document is read out.
Repeat the processing steps up to 112. In the verification information reading processing step 1101, one piece of verification information is read from the buffer 360 and output to the work area. In the collation information identification processing step 1102, it is checked whether the collation information read in the collation information reading processing step 1101 is collation term identification information. That is, if the lower 32 bits of the collation information are not 0 (zero), it is determined that it is a collation term identifier. In this case, a context marker identification processing step 1103 for detecting a context marker is executed. If the verification information is not the verification term identification information, verification information output processing step 111 outputs the verification information to the buffer 370.
Execute 2. Context marker identification processing step 1103
Then, the matching term identifier of the matching term identification information is checked to determine whether it is a context marker of the context specified in the context condition. In the case of a context marker, a post-processing step 110
4 is executed.

In post-processing step 1104, the end position information of the contexts for which co-occurrence determination has been made is stored in the work area. This context identification information is sent to the buffer 3 in a matching term identification information output processing step 1110, which will be described later.
70. After that, pre-processing step 1104a
Execute. Here, preparations are made to perform co-occurrence determination for the next context with this context marker as the endpoint. First, the co-occurrence counter of the context conditions related to the context with this context marker as the endpoint is reset to 0. Further, the head position information of this context marker is set as the head position information of the context whose end point is this context marker. In the post-processing step 1104, the context condition identification information is stored in the work area and the context condition determination is performed again because the case where the context conditions are nested is taken into consideration. Nesting of context conditions means nesting of context conditions, such as “Cj:(Ta[s]Tb)[p]Tc”.
The two search terms "Ta" and "Tb" appear in the same sentence regardless of their order, and this sentence and the search term "Tc" appear in the same paragraph regardless of their order. This refers to the case where a context condition is specified in a context condition, such as searching for a document that appears in a context condition.

[0065] In this judgment procedure, the context condition is
:(Ta[s]Tb)" and "Cj:Ci[p]Tc", and first, the context condition for the included one, that is, the context condition for the sentence included in the paragraph in this example"
Ci: (Ta[s]Tb)". It is assumed that the context condition of this sentence is satisfied. When this is satisfied, the context condition identification information of the sentence is stored in the temporary storage buffer. Next, the context marker of the sentence is determined. appears, the end position information is set in the context condition identification information of this sentence, and it is determined as the context condition identification information.The context condition identification information of this sentence is stored in the work area, and the containing context condition, that is, this example Then, the context condition regarding the paragraph “C
j:Ci[p]Tc". After that, it is assumed that the collation term identification information of the collation term identifier Tc appears. At this time, the context condition of the paragraph is satisfied, and the context condition of the paragraph is stored in the temporary storage buffer. The identification information is stored.The next time the context marker of a paragraph appears, the end position information is set in the context condition identification information of this paragraph, and it is determined as the context condition identification information.By performing processing in this way, , nested context condition “Cj: (Ta[s]Tb)[p]
This realizes the conditional judgment of “Tc”.

After the context marker identification processing step 1103, a work area repetition processing step 1106 is executed. In the work area repetition processing step 1106, the co-occurrence determination processing of steps 1107 to 1110 is repeatedly executed for all the collation term identification information stored in the work area in the collation information reading processing step 1101 and the post-processing step 1104. Context condition specification identification processing step 1
In step 107, it is checked whether the collation term identifier of the collation term identification information stored in the work area is specified as a context condition. If specified in the context condition, co-occurrence determination processing step 1108 is executed, and it is checked whether the co-occurrence counter value, which is incremented every time the collation term identifier specified in the context condition appears in the context, is 2. . The co-occurrence counter value of 2 indicates that the two matching term identifiers specified in the context condition have been found and the context condition has been satisfied. If the context condition is satisfied, a temporary storage process step 1109 is executed to store the context condition identifier of the relevant context condition in the temporary storage buffer 321. Thereafter, context marker identification processing step 1110
a is executed, and if it is not the collation term identification information of the context marker, the collation term identification information output processing step 1110 is executed. Here, the verification term identification information is stored in the buffer 37.
Output to 0. Each of the above processing steps is stored in a buffer 360.
The context condition determination process can be realized by repeatedly executing the verification information stored in the .

The above processing procedure will be explained using a specific example. The search condition is the formula “Q=((document[
4C] Understanding) [S] System) [AND] (Document [s]
Search)" is used as an example. Each condition determination program is set with a conditional expression that is analyzed by the search control means 101 and separated into each condition. Specifically, the context condition determination program 3
20 contains the context condition part of conditional expression (7-1) “(document [
4C] Understanding) [S] System” and “Document [s] Search” are set. Here, the neighborhood condition “Document [4C] Understanding”
By setting the identifier of P1 to P1 and the search term identifier of “system” to T3, the context condition “(document [4C] understanding) [S] system” is expressed in the form “C1:P1[S]T3”. C1 is an identifier of a context condition. Similarly,
The context condition “document [s] search” is “C2:T1[s]T4
”. C2 is an identifier of a context condition, T1 is a search term identifier of “document”, and T4 is a search term identifier of “search”.

[0068] Now, if the document shown in (7-2) "...This is a retrieval system using document understanding..." is input, as described above, from the neighborhood condition determination program 310, is the following (9-1) shown in FIG.
(9-8) is output to the buffer 360 as collation information. (D1, 0, 0) (9-1) (S1,
30, 30) (9-2) (T1, 31, 32)
(9-3) (T2, 33, 34) (9
-4) (P1, 31, 34) (9-5) (T3
, 39, 40) (9-6) (T4, 41, 44
) (9-7) (S1, 48, 48) (
9-8) In this example, the context marker matching information (9-2)
It is determined whether the context condition "(document [4C] understanding) [S] system" and the context condition "document [s] search" are satisfied in the context, that is, the sentence with (9-8) as the end point. That will happen.

The context condition determination process for this conditional expression will be explained using FIGS. 20a and 20b and FIGS. 21a and 21b. First, in the initial state shown in FIGS. 20a and 20b, collation information (9-1) to (9-9) is stored in the buffer 360.
-8) is stored, and the temporary storage buffer 321 and buffer 370 are in a state where they are cleared to 0. The context condition determination processing program 320 reads these collation information one by one from the buffer 360 and sets the context condition "C1:P1[S]T3" and the context condition "C2:T1".
Judgment processing is performed for “[s]T4”. First, step 1
, the reading process step 1101 shown in FIG.
0,0) is read into the program work area as shown in step 1 of FIGS. 20a and 20b. Next, the verification information identification processing step 1102 is executed, and the verification information (
It is checked whether 9-1) is the verification term identification information. In the verification information (D1, 0, 0), the last two terms are both 0 (
(zero), that is, the lower 32 bits are 0 (zero), so
It is determined to be document identification information. Therefore, matching information (D
1, 0, 0) is not matching term identification information, it is determined by the context marker identification processing step 1110a that it is not a context marker. Therefore, the next verification information output processing step 1110 is executed, and the information is output to the buffer 370 as it is.

Thereafter, as step 2, the collation information reading process step 1101 is executed again, and the collation information (9-
2) That is, (S1, 30, 30) is read. Next, a verification information identification processing step 1102 is executed, and it is checked whether the verification information (S1, 30, 30) is verification term identification information. Since the lower 32 bits of the collation information (S1, 30, 30) are not 0 (zero), it is determined to be collation term identification information rather than document identification information. Then, the next context marker identification processing step 1103 is executed, and it is checked whether this matching term identification information corresponds to the context marker of the context specified in the context condition. Verification information (
Since the collation term identifier S1 of S1, 30, 30) corresponds to "." specified as the context marker, the post-processing step 1104 is executed.

In post-processing step 1104, first, the end position information of the context with the main context marker S1 as the end point is set. Next, it is checked whether collation information is stored in the temporary storage buffer 321, and if it is stored, the entire contents of the temporary storage buffer 321 are read into the work area. The temporary storage buffer 321 stores context condition identification information corresponding to the specified context condition as collation information, as will be described later. Since no collation information is stored here, it will not be read into the work area. After this, a preprocessing step 1104a is executed. Here, the context condition of the sentence, ie “C1
:P1[S]T3" and "C2:T1[s]T4", set co-occurrence counter a and co-occurrence counter b to 0. Next, set the context start position information to the start position information of context marker S1. 30 is set. After that, the work area repetition processing step 1106 is executed. Here, since the main collation term identifier "S1" is not specified as a context condition, the context condition specification identification processing step 1107 is not executed. Identification processing step 1110a is executed, but since the present collation term identifier "S1" is the collation term identification information of the context marker, the collation term identification information output processing step 1110 is not executed.Therefore, the collation term identification information of the context marker is buffer 370
will not be output.

As step 3, iterative processing step 1
A reading process step 1101 is executed by 100,
The third verification information (9-3), that is, (T1, 31,
32) is read into the work area. Next, a collation information identification processing step 1102 is executed to check whether or not it is collation term identification information. Verification information (T1, 31, 32)
Since is collation term identification information, context marker identification processing step 1103 is executed. Verification term identification information (T1
, 31, 32) is not the matching term identifier S1 of the context marker, so post-processing step 1
104 is not executed. After this, work area repetition processing step 1106 is executed, and steps 1107 to 111 are performed for the collation term identification information stored in the work area.
0 co-occurrence determination processing is performed. In the context condition designation identification processing step 1107, it is checked whether or not the verification term is specified in the context condition by referring to the verification term identifier T1 of the verification term identification information (T1, 31, 32) stored in the work area. Since the present collation term identifier T1 is specified in the context condition "C2:T1[s]T4", the co-occurrence determination process 1108 is executed. Here, the co-occurrence counter a corresponding to the context condition "C2:T1[s]T4" is counted up, and the value of the co-occurrence counter a is updated from 0 to 1. However, since the value of the co-occurrence counter a is not 2, the co-occurrence determination will not hold. After that, the context marker identification processing step 1110a is executed, but since the matching term identification information is not of the context marker, the matching term identification information output processing step 1110 is executed. Here, the collation term identification information (T1, 31, 32) is output to the buffer 370.

Furthermore, as step 4, reading processing step 1101 is executed under the iterative processing step 1100, and the fourth collation information (9-4), that is, (T2
, 33, 34) are read into the work area. Then, a verification information identification processing step 1102 is executed, and it is checked whether or not it is verification term identification information. Verification information (T2, 3
3, 34) are matching term identification information, so the context marker identification processing step 1103 is executed, but the matching term identifier of the matching term identification information (T2, 33, 34) is S1.
Therefore, post-processing step 1104 is not executed. Thereafter, a context condition specification identification processing step 1107 is executed under the work area repetition processing step 1106, and it is checked whether the collation term identification information in the work area is specified as a context condition. Verification term identification information stored in the work area (T2, 33, 34)
Since the collation term identifier T2 is not specified in the context condition, co-occurrence determination processing step 1108 is not executed. Also, since the matching term identification information is not a context marker,
Verification term identification information output processing step 1110 is executed, and verification term identification information (T2, 33, 34) is output to buffer 370.

Furthermore, as step 5, a reading process step 1101 is executed under the iterative process step 1100, and the fifth collation information (9-5), that is, (P1
, 31, 34) are read into the work area. Further, in a matching information identification processing step 1102, it is checked whether or not it is matching term identification information, and since the matching information (P1, 31, 34) is matching term identification information, a context marker identification processing step 1103 is executed. Here, since the collation term identifier of the collation term identification information (P1, 31, 34) is not S1, the work area repetition processing step 1106
is executed, and the co-occurrence determination processes 1107 to 1110 are executed for the collation term identification information stored in the work area. In the context condition designation identification processing step 1107, it is checked whether or not it is specified as a context condition by referring to the verification term identifier P1 of the verification term identification information (P1, 31, 34) stored in the work area. In this case, the value of the co-occurrence counter b corresponding to the context condition “C1:P1[S]T3” is 0, and the collation term identifier P
1 is specified as the 0th collation term identifier of the context condition "C1:P1[S]T3". Therefore, co-occurrence determination processing 1108 is executed. Here, the co-occurrence counter b is counted up, and the value of the co-occurrence counter b changes from 0 to 1. However, since the value of the co-occurrence counter b is not 2, the co-occurrence determination will not hold. After this,
Since this verification term identification information is not that of a context marker, verification term identification information output processing step 1110 is executed, and actual verification term identification information (P1, 31, 34) is output to the buffer 370.

Furthermore, as step 6, a reading processing step 1101 is executed under the iterative processing step 1100, and the sixth collation information (9-6), that is, (T3
, 39, 40) are read into the work area, and then the collation information identification processing step 1102 is executed. Here, it is checked whether the collation information (T3, 39, 40) is collation term identification information or not, and since the collation information (T3, 39, 40) is collation term identification information, context marker identification processing step 1103 is executed. , matching term identification information (T3,
39, 40) are not context markers, the post-processing step 1104 will not be performed. After this, the work area repeat processing step 1106 is executed, and the verification term identification information stored in the work area is
Co-occurrence determination processes 107 to 1110 will be executed. In context condition specification identification processing step 1107, matching term identification information (T3,
39, 40), it is checked whether or not this matching term identifier is specified as a context condition. In this case, the context condition “C1:P1[S]T3
The value of the co-occurrence counter b corresponding to " is 1, and the matching term identifier T3 is the context condition "C1:P1[S]T3"
Since it is specified as the first identifier of
108 is executed. Here, the co-occurrence counter b is counted up, and the value of the co-occurrence counter b is updated from 1 to 2. Since the value of the co-occurrence counter b has become 2, it can be determined that the co-occurrence determination has been established. At this time,
Temporary storage processing step 1109 is executed, and the context condition “
C1:P1[S]T3'' context condition identification information (C1,3
0, Xe1) is stored in the temporary storage buffer 321. Here, since the position information of the context marker whose end point is after the context is not known yet, we will temporarily set the context end position information to
Set it as e1. After this, when the collation term identification information of the context marker appears, this context end position information is set in post-processing step 1104. After that, since this verification term identification information is not that of a context marker, verification term identification information output processing step 1110 is executed, and the verification term identification information (T3, 39, 40) that has just been established is transferred to the buffer 3.
70.

As step 7, iterative processing step 1
100, reading processing step 1101 is executed, and the seventh collation information (9-7), that is, (T4, 41
, 46) are read into the work area. Thereafter, a verification information identification processing step 1102 is executed, and it is checked whether this verification information is verification term identification information. Verification information (T
4, 41, 46) are matching term identification information, the context marker identification processing step 1103 is executed, and since the matching term identification information (T4, 41, 46) is not a context marker, the post-processing step 1104 is not executed. After this,
Work area repeat processing step 1106 is executed,
The co-occurrence determination processes 1107 to 1110 are executed for the collation term identification information stored in the work area. In the context condition specification identification processing step 1107, the matching term identification information (
With reference to the matching term identifier T4 of T4, 41, 46), it is checked whether or not this matching term is specified as a context condition. In this case, since the collation term identifier T4 is specified in the context condition "C2:T1[s]T4", the co-occurrence determination process 1108 is executed. Here, the context condition “C2
:T1 [s] Verification term identifier T in T4”
4 appears, the co-occurrence counter a corresponding to this context condition
is counted up, and the value of co-occurrence counter a changes from 1 to 2.
Therefore, the co-occurrence determination is established. therefore,
Temporary storage processing step 1109 is executed, and the context condition identification information (C2, 30, Xe2) of the context condition “C2:T1[s]T4” that has just been established is stored in the temporary storage buffer 321. Since the context end position information has not been determined,
Set e2 temporarily. Thereafter, since this verification term identification information is not that of a context marker, verification term identification information output processing step 1110 is executed, and verification term identification information (T4, 41, 46) is output to the buffer 370.

Finally, in step 8, the collation information reading processing step 1101 is executed again, and the main collation information (9
-8), that is, (S1, 48, 48) is read. Next, a verification information identification processing step 1102 is executed to check whether the verification information (S1, 48, 48) is verification term identification information. Verification information (S1, 48, 48
) is verification term identification information, the next context marker identification processing step 1103 is executed to check whether this verification term identification information corresponds to a context marker related to the context specified in the context condition. Since the collation term identifier is S1, it corresponds to the context marker “.” of the specified sentence, and the post-processing step 110
4 is executed. In post-processing step 1104, first, 48, which is the end position information of this context marker, is set as the context end position information. Next, the context condition identification information (C
1, 30, Xe1) and (C2, 30, Xe2) are stored in the temporary storage buffer 321, and the processing results are shown in step 8 of FIGS. 21a and 21b. (C1, 30, 48) and (
C2, 30, 48). This processing result is stored in the work area. Further pre-processing step 1104
a is executed, the co-occurrence counter of the context condition of the sentence is cleared to 0, and the start position information 48 of this context marker is set as the context start position information.

Thereafter, work area repetition processing step 1106 is executed. First, context condition identification information (C1,
30, 48), the context condition specification identification processing step 1107 is executed, but since the collation term identifier C1 is not specified in the context condition, the co-occurrence determination processing 1108 is not executed, and the collation term identification information output processing step 1110
is executed, and the context condition identification information (C1, 30, 48) is output to the buffer 370. In other words, the location information of this context condition identification information is the context marker identifier (9-2).
Start position information 3 of the sentence with (9-8) as the end point
0 and the end position information 48 are used as position information. After this, the work area repetition processing step 1106 is executed again, and the context condition specification identification processing step 1107 is executed for the context condition identification information (C2, 30, 48), but the collation term identifier C2 is also specified as a context condition. Therefore, the context condition identification information (C2, 30, 48) is output to the buffer 370 by the matching term identification information output processing step 1110.

By executing the above context condition determination processing, collation information (10-1) to (10-8) shown in FIG. 22 is output to the buffer 370. (D1, 0, 0) (10-1) (T1
,31,32) (10-2)(T2,33,3
4) (10-3) (P1, 31, 34)
(10-4) (T3,39,40) (10-
5) (T4, 41, 44) (10-6) (C1
,30,48) (10-7)(C2,30,4
8) (10-8) Here, (10-7) is the context condition identification information of the context condition “(Document [4C] Understanding) [S] System”, and similarly (10-8) is the context condition identification information of “Document [4C] Understanding) [S] System”. s]
"Search" context condition identification information.This collation information is subsequently sent to the logical condition determination program 330.

Finally, the determination processing of the logical condition determination program 330 will be explained. Logical condition judgment program 3
At step 30, the collation information sent from the context condition determination program 320 to the buffer 370 is read out, and it is determined whether it matches the logical condition specified as the search information 202 by the search control means 101. An example of a logical condition is, for example, a conditional expression "document [AND] search". The conditional expression "document [AND] search" indicates that a document in which the two search terms "document" and "search" appear simultaneously is to be found. In addition, an identifier (
(hereinafter referred to as a logical condition identifier) is assumed to be Li. Furthermore, Li
Assign a code that can be distinguished from the collation term identifier. By defining in this way, the logical condition becomes “Li
:Ta[AND]Tb". The following explanation uses this definition. In addition to this, logical conditions include "Li:Ta[OR
]Tb” and “Li:Ta[NOT]Tb”.Here, the logical condition “Li:Ta[OR]Tb” means to search for documents in which the search term “Ta” or “Tb” appears. , logical condition “Li:Ta[NOT]
"Tb" indicates that documents in which the search term "Ta" appears and in which the search term "Tb" does not appear are searched.

These logical conditions are converted by the search control means 101 into the general form of a product shown below, and specified as search information 202 to the logical condition determination program 330. Li: (A11+A12+...+A1j)*(A21
+A22+...+A2k)*...*(An1+A22+...+Anm) (10-1) In formula (10-1), "+" represents a logical sum, and "
*” represents logical product. Also, Anm is called an element, and (
An1+A22+...+Anm) is called a term. Element A
There is also an element ¬Anm which is negated (represented by "¬") as nm. In addition, the term is also negated.
¬(An1+A22+...+Anm) exists. Here, ¬Anm is called a negative logic element. In contrast, terms that are not negated are called elements of positive logic. Also, ¬
(An1+A22+...+Anm) is called a negative logic term, and a term that is not negated is called a positive logic term.

Since the formula (10-1) is in the form of a logical product of terms, the formula (10-1) holds true (becomes true).
In order to do so, each term must all hold true. Therefore, a counter is provided to count the terms that hold true. If the value of this counter (hereinafter referred to as term counter) is equal to the number of terms, then equation (10-1) is established. Negative logic terms that do not include negative logic elements and positive logic terms that include negative logic elements are established from the beginning. Therefore, the initial value of the term counter is set to the sum of the number of negative logic terms that do not include negative logic elements and the number of positive logic terms that include negative logic elements. By controlling this term counter as follows, it is possible to determine whether equation (10-1) is successful or not. In other words, the number of negative logic terms is set as an initial value in the term counter, and when a term changes from not being true to being true, 1 is added to the term counter, and conversely, when a term changes from being true to being true, 1 is added to the term counter. subtracts 1 from the term counter. If the term does not change from not being true to not being true or from being true to true, the value of the term counter is not updated. By controlling the term counter in this way and checking whether the value of the term counter is equal to the number of terms for each document, it is possible to determine whether equation (10-1) holds true. Conversely, if the entire equation is negated, it can be determined whether equation (10-1) holds true by checking whether the value of the term counter is less than the number of terms.

Next, since all terms are the logical sum of elements, in order for a term to hold true, it is sufficient that any element holds true. A method for determining success or failure of one term will be described below. Here, a counter is used to check the success or failure of the term, and in the element specified in the term,
Count the number of elements that are satisfied. If the value of this counter (hereinafter referred to as an element counter) is 1 or more, that is, if any element is satisfied, the term is considered to be satisfied. Since negative logic elements can be considered to be established from the beginning, the initial value of the element counter is set to the number of negative logic elements. Therefore, one element counter is provided for each term, that is, the number of element counters is equal to the number of terms.

The success or failure of a term can be determined by controlling this element counter as follows. In other words,
The number of negative logic elements is set as an initial value in the element counter, and when an element changes from not being true to being true, 1 is added to the element counter, and conversely, when the element changes from being true to not being true, the number of elements is added to the element counter. Subtract 1 from the counter. Since the element here corresponds to the collation term identifier, the element does not change from not being true to not being true or from being true to true. Therefore, only positive logic elements change from not being true to being true, and only negative logic elements change from valid to not true. Therefore, when matching term identification information corresponding to a positive logic element appears, 1 is added to the element counter; conversely, when matching term identification information corresponding to a negative logic element appears, 1 is added to the element counter.
Subtract.

The element counter is controlled in this way, and it is determined whether a term is established by checking whether the value of the element counter is 1 or more each time collation term identification information is input. Conversely, when a term is negated, it is determined whether the term holds true by checking whether the value of the element counter is 0. After this,
Based on the result of determining whether a term is true, if the term changes from not to true, add 1 to the term counter; conversely, if the term changes from valid to not true, subtract 1 from the term counter. . Further, if the term including negation does not change from untrue to untrue or from valid to true, the value of the term counter is not updated. By controlling the term counter in this way and checking whether the value of the term counter is equal to the number of terms for each document, it is possible to determine whether equation (10-1) is successful or not. For example, in the case of the above “Li:Ta[AND]Tb”, that is, “Li:Ta*Tb”, it is converted as “Li:¬((¬Ta)+(¬Tb))” and the search information 202 is Logical condition judgment program 33
Passed to 0.

FIG. 23 shows the procedure for processing this logical condition.
This will be explained in detail using . For example, the above “Li:Ta[
AND]Tb”, that is, “Li:Ta*Tb”, it is converted as “Li:¬((¬Ta)+(¬Tb))” (10-2) and the logical condition is determined as the search information 202. It is passed to the program 330. This logic condition has one negative logic term. Furthermore, this term has two negative logic elements. Since there is one term, here, the element counter The initial value of the term counter is set to the sum of the number of negative logic terms that do not include negative logic elements and the number of positive logic terms that include negative logic elements. However, since the term in this example is a negative logic term that includes negative logic elements, it is not applicable and is therefore set to 0. Also, the element counter is set to 0.
Set 2, which is the number of negative logic elements. Here, since the term is multiplied by ¬, it is possible to count down the element counter and determine that the term has been established when the value of the element counter reaches 0. Also, the logical condition (10
-2) Since the equation is not multiplied by ¬, it can be determined that the logical condition (10-2) is satisfied when the value of the term counter is 1.

First, in the iterative processing step 1200, the processing steps 1201 to 1210 are repeated until all the collation information in the buffer 370, that is, the collation information of the last document is read out. Reading process step 12
At step 01, collation information is read from the buffer 370 and stored in the work area of the program. In the collation information identification processing step 1202, the above reading processing step 1201
Check whether the collation information read in is document identification information or collation term identification information. That is, if the last two terms of the collation information are both 0 (zero), that is, the lower 32 bits are 0 (zero), it is determined that it is document identification information. In this case, logical condition establishment determination processing step 1203 is executed. When the logical condition is satisfied, that is, in the logical condition (10-2) formula, when the value of the term counter is 1, it is shown that the logical condition is satisfied.

If the logical condition is satisfied, document identification information determination processing step 1203a is executed, and if the main document identification information is not the first document identification information, result output processing step 1204 is executed. In document identification information determination processing step 1203a, no determination processing is performed in the case of the first document identification information, because no logical condition determination processing has been performed yet. In the result output processing step 1204, the contents of the output buffer 331 storing collation information to be described later and the logical condition determination result (hereinafter referred to as logical condition identification information) are output as the composite condition determination result 206. In this way, when document identification information is detected,
The logical condition determination process is performed because it is necessary to determine whether the specified logical condition is satisfied or not for each document.

[0089] As for the contents of the logical condition judgment result information output here, the position information is set with the document start position information and the document end position information, and the identifier is collation information in which the logical condition identifier Li is stored. (Li, document start position information, document end position information). This document start position information is always 0 (zero) because the position information is always cleared to 0 at the start of the document. Therefore, it becomes logical condition identification information (Li, 0, document end position information). Further, after the logical condition establishment determination processing step 1203, an initialization processing step 1205 is executed to clear the output buffer 331 to 0 and initialize the element counter and term counter. In this example, the element counter is set to 2 and the term counter is set to 0.

If it is determined in the matching information identification processing step 1202 that the matching information is matching term identification information, the matching term identification processing step 1206 is executed, and the matching term identifier in the matching term identification information is determined to be a logical condition. Check whether it is specified in the element. If specified as an element here, element determination processing step 120
6a, first check the logical condition for the found element, and if the element is not marked with ¬, execute count-up processing step 1206b and add 1 to the element counter.
Add. Conversely, if the element is marked with ¬, countdown processing step 1206c is executed and 1 is subtracted from the element counter. Here, when the value of the element counter becomes 0, it indicates that the term does not hold, and when it becomes 1 or more, it indicates that the term holds. Conversely, when the term is multiplied by ¬,
When the value of the element counter is 0, it indicates that it is true, and when it is 1 or more, it indicates that it does not hold.

Next, a term establishment determination processing step 1207 is executed. Here, we check whether the term for which the element is specified has changed from not being true to being true, or from being true to not being true. The method for determining whether a term holds true is that if the element counter changes from unfulfilled to true before and after this calculation, the number of satisfied terms has increased by one, so 1 is added to the term counter, and the condition changes from valid to unfulfilled. In this case, the number of valid terms has decreased by one, so 1 is subtracted from the term counter. In this example, the value of the element counter in the initial state is 2 and the value of the term counter is 0. If element Ta is found here,
Since element Ta is multiplied by ¬, 1 is added from the element counter.
, the value of the element counter becomes 1. In this case, since the term is multiplied by ¬, the term has changed from not being true to not being true, so the term counter is not updated. Furthermore, element T
If b is found, element Tb is multiplied by ¬, so 1 is subtracted from the element counter. The value of the element counter becomes 0, which means that the term has changed from not being true to being true, so 1 is added to the term counter. As a result, the value of the term counter becomes 1. As described above, the term establishment determination process is performed.

After the verification information identification processing step 1202, a verification information saving processing step 1210 is executed, and the verification information is output to the output buffer 331. Verification information saving processing step 1210 is executed. This output buffer 3
31 stores collation information for one document, and outputs it as a composite condition determination result 206 when it is found in the logical condition establishment determination process step 1203 that the logical condition is established. By repeatedly performing each of the above processing steps on the collation information stored in the buffer 370, the logical condition determination processing can be realized. Iterative processing step 12
After the end of 00, logical condition determination will not be executed for the last processed document. This is because the input of the document identification information is used to determine the logical conditions of the previously read document. Therefore, the logical condition establishment determination processing step 1203 is executed again here,
Performs logical condition judgment on the last read document.

The above processing procedure will be explained using a specific example. The search condition is the formula “Q=((document[
4C] Understanding) [S] System) [AND] (Document [s]
Search)" will be explained as an example. Each condition judgment program is analyzed by the search control means 101 and a conditional expression separated into each condition is set. Specifically, the logical condition judgment program 330 has The logical condition part of this conditional expression (7-1) “
((Document [4C] Understanding) [S] System) [AND] (
"Document[s]Search)" is set.Here, the identifier of the context condition "(Document[4C]Understanding)[S]System" is set as C
1, and the identifier of the context condition “Document [s] search” is set to C2, the logical condition part of (7-1) is changed to “L1:
C1[AND]C2".L1 is a logical condition identifier.Furthermore, "L1:¬((¬C1)+(¬C
2))” and set in the logical condition determination program 330.

[0094] Now, if the document shown in (7-2) "...This is a retrieval system using document understanding..." is input, as described above, the context condition determination program 320 is the following (11-1
) to (11-8) are output to the buffer 370 as verification information. (D1, 0, 0) (11-1) (T1
,31,32) (11-2)(T2,33,3
4) (11-3) (P1, 31, 34)
(11-4) (T3,39,40) (11-
5) (T4, 41, 44) (11-6) (C1
,30,48) (11-7)(C2,30,4
8) (11-8) (11-1) represents document identification information, D1 represents a document identifier, and the following 2
The term is constant 0. (11-2), (11-3), (1
1-5) and (11-6) are "documents" and "documents", respectively.
Represents the collation term identification information of “understand,” “search,” and “system.” Also, T1, T2, T3, and T
4 represent "document", "understand", "search", and "system verification term identifier, respectively. Also, (11
-4) represents the neighborhood condition identification information of the neighborhood condition "understand document [4C]", and P1 represents this neighborhood condition identifier. Furthermore, (11-7) is a context condition “(document [4C
] understanding) [S] system", C2 represents this context condition identifier, and (11-8)
represents the context condition identification information of the context condition “document [s] search”, and C2 represents this context condition identifier.

The logical condition determination process under the above conditions will be explained step by step with reference to FIGS. 24a, 24b and 25. In the initial state, collation information (11-1) is stored in the buffer 370 as shown in the initial states of FIGS. 24a and 24b.
~(11-8) are stored, and the output buffer 331
And the buffer 370 is in a state where it is cleared to 0. In the logical condition determination processing program 330, the buffer 37
Read these collation information one by one from 0 and set the logical condition "
Li:¬((¬C1)+(¬C2))".

First, as shown in step 1 of FIGS. 24a and 24b, the reading process step 1201 is executed and collation information (11-1), that is, collation information (D1, 0, 0)
is loaded into the program's work area. Next, the verification information identification processing step 1202 is executed, and the verification information (
It is checked whether 11-1) is document identification information or verification term identification information. Verification information (D1, 0, 0) is determined to be document identification information because the last two terms are both 0 (zero). Therefore, logical condition establishment determination processing step 1203 and document identification information determination processing step 120
3a is executed, but since this is the first document identification information, result output processing step 1204 is not executed. Next, the initial setting processing step 1205 is executed, and the output buffer 331
Clear to 0 and initialize the element counter and term counter. In this example, the initial value of the element counter is set to 2, which is the number of elements (¬C1) and (¬C2) on which ¬ is applied. Also, this term ((¬C1) + (¬C2)) is
It is a term multiplied by ¬ that includes an element multiplied by . Therefore, the initial value of the term counter is set to 0 because there is no term with ¬ that does not include an element with ¬ or a term with no ¬ that includes an element with ¬. When the verification information identification processing step 1202 is completed, the verification information saving processing step 1210 is executed next, and the verification information (D1, 0, 0) is transferred to the output buffer 33.
It is stored in 1.

Thereafter, as step 2, the collation information reading processing step 1201 is executed again, and the collation information (11
-2), that is, (T1, 31, 32) is read. Next, a verification information identification processing step 1202 is executed, and it is checked whether the verification information (T1, 31, 32) is verification term identification information. Since the lower 32 bits of the collation information (T1, 31, 32) are not 0 (zero), it is determined to be collation term identification information. Then, in the next verification term identification processing step 1206, it is checked whether this verification term identification information corresponds to an element specified in the logical condition. Since the matching identifier T1 of the matching information (T1, 31, 32) is not specified in this logical condition, processing step 1206
Steps a to 1209 will not be executed. After the verification information identification processing step 1202 is completed, the verification information saving processing step 1210 is executed, and the verification information (T1, 31,
32) is stored in the output buffer 331. Similarly, until step 6, the collation information identifier of the collation information is not specified in this logical condition, so only the collation information save processing step 1210 is executed, and the collation information (T2, 33, 34), (P1
, 31, 34), (T3, 39, 40), and (T4
, 41, 44) are stored in the output buffer 331.

Furthermore, in step 7, the collation information reading process step 1201 is executed again and the collation information (11-
7), that is, (C1, 30, 48) is read. Next, a verification information identification processing step 1202 is executed, and it is checked whether the verification information (C1, 30, 48) is verification term identification information. Since the lower 32 bits of the collation information (C1, 30, 48) are not 0 (zero), the collation information (C1, 30, 48) is determined to be collation term identification information. Then, in the next verification term identification processing step 1206, it is checked whether the verification term identification information corresponds to an element specified in the logical condition. Since this logical condition specifies the collation identifier C1 of the collation information (C1, 30, 48), processing steps 1206a to 1206a to
1209 will be executed. First, element determination processing step 1206a is executed. In this logical condition, the initial value of the element counter is 2 and the initial value of the term counter is 1. Here, since the element C1 is multiplied by ¬, countdown processing step 1206c is executed. Here, 1 is subtracted from the element counter. As a result, the value of the element counter is updated from 2 to 1. Next, term determination processing step 1207 is executed. Since the value of the element counter at this point is 1, the term changes from not being true to not being true. Therefore, the term counter is not updated. After the verification information identification processing step 1202 is completed, the verification information saving processing step 1210 is executed, and the verification information (C1, 30, 48
) is stored in the output buffer 331.

Finally, in step 8, verification information reading processing step 1201 is executed and verification information (11-8) is executed.
, that is, (C2, 30, 48) is read. Next, a verification information identification processing step 1202 is executed, and it is checked whether the verification information (C2, 30, 48) is verification term identification information. Verification information (C2, 30, 48) is the lower 32
Since the bit is not 0 (zero), it is determined to be verification term identification information. Then, in the next verification term identification processing step 1206, it is checked whether the verification term identification information corresponds to an element specified in the logical condition. This logical condition includes the collation identifier C2 of collation information (C2, 30, 48).
is specified, processing steps 1206a to 1206a-12
09 will be executed. At this point, the value of the element counter is 0, and the value of the term counter is 1. first,
Element determination processing step 1206a is executed. Here, since element C2 is multiplied by ¬, countdown processing step 1206c is executed. Here, 1 is subtracted from the element counter. As a result, the value of the element counter changes from 1 to 0. Next, term determination processing step 1207 is executed. At this time, since ¬ is applied to this clause, a change has occurred from not being true to being true. Therefore, count-up processing step 1208 is executed, and the term counter is incremented by 1 and updated from 0 to 1. Thereafter, the collation information saving processing step 1210 is executed, and the collation information (C2, 30, 48) is stored in the output buffer 331. At this point, the iterative process 1200 ends, but since it is the last document, logical condition satisfaction determination process step 1203 is executed. Since the term counter is 1, this logical condition is satisfied. Therefore, the result output processing step 1204 is executed, and the contents of the output buffer 331 storing the collation information and the logical condition determination result information (L1, 0, 99) are output as the composite condition determination result 206. L1 represents a logical condition identifier, 0 represents document start position information, and 99 represents document end position information.

[0100] As a result of the above logical condition judgment processing, the collation information of (11-1) to (11-9) shown below is finally determined by the expression "Q" shown in the search condition expression (7-1). =((document[
4C] Understanding) [S] System) [AND] (Document [s]
(D1, 0, 0) (11-1) (T1
,31,32) (11-2)(T2,33,3
4) (11-3) (P1, 31, 34)
(11-4) (T3,39,40) (11-
5) (T4, 41, 44) (11-6) (C1
,30,48) (11-7)(C2,30,4
8) (11-8) (L1, 0,99)
(11-9) (11-1) represents document identification information, and D1 represents a document identifier. (11-2), (1
1-3), (11-5), and (11-6) represent collation term identification information of "document", "understanding", "search", and "system", respectively. Also, (11-4)
represents the neighborhood condition identification information of the neighborhood condition “understand document [4C]”. Furthermore, (11-7) represents the context condition identification information of the context condition “(Document [4C] understanding) [S] system”, and (11-8) represents the context condition identification information of the context condition “Document [s] search”. represents information. The last (11-9) is the logical condition “((Document [4C] Understanding) [S] System) [A
ND] (Document [s] search)" represents the logical condition identification information, and L1 represents this logical condition identifier. The above is the logical condition judgment program 3.
10, a method for realizing complex condition determination processing constituted by a context condition determination program 320 and a logical condition determination program 330.

As explained above, the character string matching circuit 200
By configuring the complex condition determination circuit 300,
Complex conditions such as neighborhood conditions, context conditions, and logical conditions can be consistently realized, making it possible to perform fine-grained searches unique to full-text searches. moreover,
For example, three microcomputers each have a neighborhood condition determination program 310, a context condition determination program 320,
By executing the logical condition determination program 330, it becomes possible to operate the programs without synchronizing them. In other words, these programs can perform pipeline processing that starts condition judgment processing in response to collation information stored in each input buffer, making it possible to realize high-speed complex condition judgment processing. .

Next, regarding the second embodiment of the present invention, FIG.
Explain using. In this embodiment, although only one microprocessor operates when only one of the compound conditions is set, the first condition is that pipeline processing of three compound conditions must always be performed.
This method solves the drawbacks of the above embodiments by bypassing the complex condition judgment program when there is a complex condition that is not used in a given search condition, thereby providing efficient complex condition judgment processing. . This embodiment consists of a character string matching circuit 200 and a complex condition judgment circuit 300a. It is configured. In the microcomputer MPUa301, the neighborhood condition determination program 310 is executed in the microcomputer MPUb302.
In the case of the microcomputer MPUc303, the context condition judgment program 320 is the logical condition judgment program 33.
0 is executed. Furthermore, buffers 350, 360, and 370 using first-in, first-out (FIFO) memories are arranged between each MPU, and are used to transfer data between the respective MPUs.

The bypass function of the complex condition determination circuit, which is a feature of this embodiment, will be described. This function uses a multiplexer or selector to bypass the microprocessor that is equipped with a judgment program that corresponds to a complex condition that is not used in the specified search condition expression. This function allows only the existing microprocessor to execute. For example, when the search condition expression "Document [4C] understanding" is specified, only the neighborhood condition is used, so the collation information 205 obtained from the string matching circuit 200 is sent to the neighborhood condition determination program 310 via the buffer 350. The neighborhood condition determination program 310 performs neighborhood condition determination, and collation information is sent via the buffer 360 directly as the composite condition determination result 206. Furthermore, when the search condition expression "(Document [S] understanding) [AND] system" is specified, the context condition and logical condition are used, so the collation information 205 obtained from the string collation circuit 200 is buffered. The collation information is directly input to the context condition determination program 320 via the buffer 350 , and the collation information output from the context condition determination program 320 is input to the logical condition determination program 330 via the buffer 370 .
The verification information output from 0 is sent as the composite condition determination result 206.

A specific implementation method of the bypass function of such a complex condition determination circuit 300a will be explained. This bypass operation is realized by setting multiplexers 390-392 and selectors 380-382. This setting information is given as search information 202 from the search control means 101 in FIG. The multiplexer 390 sends the collation information 205 from the character string collation circuit 200 to the logical condition determination program 330 if a1 is selected, to the context condition determination program 320 if b1 is specified, and to the context condition determination program 320 if c1 is specified. to the neighborhood condition determination program 310, d
If 1 is selected, it will be sent as the composite condition determination result 206 of the composite condition circuit 300a. The multiplexer 391 sends the output of the neighborhood condition judgment program 310 to the context condition judgment program 320 when a2 is specified, to the logical condition judgment program 330 when b2 is specified, and to the logical condition judgment program 330 when c2 is specified. is directly sent out as the output of the complex condition circuit 300a. The multiplexer 392 sends the output of the context condition determination program 320 to the logical condition determination program 330 when a3 is specified, and as the composite condition determination result 206 when b3 is specified.

The selector 380 selects the input of the context condition determination program 320, and when b1 is set in the multiplexer 390, selects X1, that is, the matching result 205 of the character string matching circuit 200, and selects the input of the multiplexer 390.
When a2 is set in 91, Y1, that is, the output of the neighborhood condition determination program 310 is selected. Selector 3
In 81, the input of the logic condition determination program 320 is selected, and when a1 is set in the multiplexer 390, X2 is selected, that is, the collation result 20 of the character string collation circuit 200 is selected.
5 is selected, and when the multiplexer 391 is set to b2, Y2 is selected, that is, the neighborhood condition determination program 310
When multiplexer 392 is set to a3, Z2 is selected, that is, context condition determination program 3
Select 20 outputs. The selector 382 selects the composite condition judgment result 206 and sends it to the multiplexer 390.
When 1 is set, Z3, that is, the matching result 205 of the character string matching circuit 200 is selected, and the multiplexer 39
1 is set to c2, selects Y3, that is, the output of the neighborhood condition judgment program 310, and when multiplexer 392 is set to b3, selects X3, that is, the output of the context condition judgment program 320, and selects it. In other cases, W3 is selected, that is, the output of the logical condition determination program 330 is selected. As described above, the settings of the selectors 380 to 382 are set by the multiplexers 390 to 390.
This will be done according to the settings of 92.

[0106] Depending on the combination of complex conditions, multiplexers are selected as follows. ■No compound condition ─────────Select only d1 ■Neighborhood condition only ─────────Select c1 and c2 ■Context condition only ─────────B1 and Select b3 ■ Logical condition only ────────── Select only a1 ■ Neighborhood condition and context condition ────── Select c1, a2, and b3 ■ Neighborhood condition and logical condition ──── ──Select c1 and b2 ■Context condition and logical condition ──────Select b1 and a3 ■Neighborhood condition, context condition and logical condition ─Select c1, a2, and a3 As described above, search control means Search information 20 from 101
Based on the setting information of the multiplexers 390 to 392 and selectors 380 to 382 given as 2, the neighborhood condition determination program 310 and the context condition determination program 32
By selectively connecting complex condition judgment programs such as 0 and logical condition judgment program 330 using multiplexers 390 to 392 and selectors 380 to 382, one micro Although only a processor is used, the drawback of the first embodiment of always performing pipeline processing of three complex condition judgment programs is solved, and it becomes possible to realize efficient complex condition judgment processing, resulting in high speed processing. A document search device that performs a search can be realized.

Next, regarding the third embodiment of the present invention, FIG.
Explain using. In this embodiment, the complex condition determination circuit 3
In 00b, three complex condition judgment programs, the neighborhood condition judgment program 310, the context condition judgment program 320, and the logical condition judgment program 330, are mounted on one microcomputer, and these compound conditions are switched and executed in order. , the purpose is to realize complex condition judgment processing. Although the complex condition determination circuit 300b of this embodiment is inferior to that of the first embodiment in terms of processing speed, it can be implemented with one microcomputer, so it has the advantage of lowering costs.

[0108] This embodiment consists of a character string matching process 200 and a complex condition judgment circuit 300b, and the complex condition judgment circuit 300b further comprises a buffer 350 for exchanging data with the microcomputer MPUa 301 and the string matching circuit 200. Ru. microcomputer MP
The Ua 301 is equipped with three complex condition determination programs such as a neighborhood condition determination program 310, a context condition determination program 320, and a logical condition determination program 330, and a scheduler 340 that switches between these programs. Here, the buffers 360 and 370 are secured as work areas for the program, but by using them as first-in, first-out (FIFO) memories by the program, the same functionality as in the first embodiment can be obtained. The order in which the complex condition programs are switched in the scheduler 340 is to first execute the neighborhood condition determination program 310, then the context condition determination program 320, then the logical condition determination program 330, and then return to the neighborhood condition determination program 310. The compound condition programs are switched in the correct order. Further, the timing for switching the compound condition program in the scheduler 340 can be set every time n pieces of collation term information are processed or every time n documents are processed. This is determined in consideration of the program switching processing time of the scheduler 340. If you switch frequently, the program switching time will occupy a large proportion of the execution processing time of the complex condition program, so it may be necessary to switch programs when there are hundreds to thousands of collation information or tens to hundreds of documents. is effective. By configuring the complex condition determination circuit 300b as described above, although the processing speed is lower than that of the first embodiment, one microcomputer can process the neighborhood condition determination program 310, the context condition determination program 320, and the logical condition determination program. Since complex condition determination processing such as H.330 can be realized, a low-cost document retrieval device can be realized.

Next, regarding the fourth embodiment of the present invention, FIG.
Explain using. In this embodiment, the complex condition determination circuit 30
In 0c, two complex condition judgment programs, the neighborhood condition judgment program 310 and the context condition judgment program 320, are installed on one microcomputer, and these compound conditions are switched and processed in order, so that the two conditions of neighborhood condition and context are The purpose is to realize complex condition judgment processing. In addition, the logical condition determination program 330 is run on another microcomputer MPUb303.
be installed on. The complex condition determination circuit 300c of this embodiment is as follows:
Although it is inferior to the first embodiment in terms of processing speed, it can be implemented with two microcomputers, so it has the effect of reducing costs. In this embodiment, the character string matching circuit 2
00 and a complex condition judgment circuit 300c, and the complex condition judgment circuit 300c is a microcomputer MPUa30.
5 and a buffer 35 that transfers data between the microcomputer MPUb303 and the character string matching circuit 200.
It consists of 0,370.

The microcomputer MPUa is equipped with two complex condition determination programs, a neighborhood condition determination program 310 and a context condition determination program 320, and a scheduler 341 for switching between these programs. Here, the buffer 360 is secured as a work area for the program, but by using it as a first-in first-out (FIFO) memory by the program, the same function as in the first embodiment can be obtained. The scheduler 341 switches the complex condition programs in such an order that the neighborhood condition determination program 310 is first executed, the context condition determination program 320 is then executed, and then the neighborhood condition determination program 310 is returned to. Further, the timing at which the compound condition program in the scheduler 341 is switched is set such as every time n pieces of collation term information are processed or every time n documents are processed. This is determined in consideration of the program switching processing time of the scheduler 341. If you switch frequently, the program switching time will take up a large proportion of the execution time of the complex condition processing program, so it may be necessary to switch programs when there are hundreds to thousands of matching information or tens to hundreds of documents. is effective. By configuring the complex condition determination circuit 300c as described above, although the processing speed is inferior to that of the first embodiment, two microcomputers can be used to configure the neighborhood condition determination program 310,
Since complex condition determination processing such as the context condition determination program 320 and the logical condition determination program 330 can be implemented, it is possible to provide a relatively high-speed document retrieval device with reduced costs.

Next, FIG. 29 shows a fifth embodiment of the present invention.
Explain using. This embodiment overcomes the disadvantage of the fourth embodiment in that pipeline processing of complex conditions is always performed by two microprocessors even when only one of the complex conditions is set. The purpose of this invention is to provide a complex condition judgment circuit 300d with high processing efficiency by bypassing the complex condition judgment program when there is a complex condition that does not exist. This embodiment consists of a character string matching circuit 200 and a complex condition judgment circuit 300d.
00d are buffers 350 and 37 that exchange data between the microcomputer MPUa 305 and microcomputer MPUb 303 and the character string matching circuit 200.
0, and further includes multiplexers 391 and 392 and selectors 380 and 381.

The bypass function of the complex condition determination circuit, which is a feature of this embodiment, will be described. This function uses a multiplexer or selector to bypass the microprocessor on which the complex condition judgment program that is not used in the specified search condition expression is installed. This is a function that only executes. For example, when the search condition expression "Document [4C] understanding" is specified, only the neighborhood condition is used, so the collation information 205 obtained from the character string collation circuit 200 is sent to the neighborhood condition determination program 310 and the neighborhood condition determination program 310 via the buffer 350. The information is input to the context condition determination program 320, and the collation information outputted from the program 320 is directly sent as the composite condition determination result 206 via the buffer 370. In addition, when the search condition expression "understand [AND] system" is specified, since a logical condition is used, the collation information 205 obtained from the character string collation circuit 200 is directly passed through the buffer 350 to determine the logical condition. The collation information inputted to the program 330 and outputted from the logical condition determination program 330 is sent out as the composite condition determination result 206.

A specific method for implementing the bypass function of such a complex condition determination program will be explained. This operation is performed by multiplexers 391 and 392 and selector 3.
This is achieved by setting 80 and 381. This setting information is given as search control information 202 from search control means 101 in FIG. The multiplexer 391 sends the collation information 205 from the character string collation circuit 200 to the logic condition determination program 330 when a2 is specified. Furthermore, when b2 is specified, the matching information 205 from the character string matching circuit 200 is sent to the neighborhood condition determination program 310, and when c2 is specified, it is sent directly as the output of the compound condition circuit 300. It turns out. The multiplexer 392 sends the output of the context condition determination program 320 to the logical condition determination program 330 when a3 is specified, and as the composite condition determination result 206 when b3 is specified.

The selector 380 selects the input of the logic condition determination program 330, and the multiplexer 391
When a2 is set in the multiplexer 392, select X1, that is, the matching result 205 of the character string matching circuit 200, and when a3 is set in the multiplexer 392, select Y1, that is, the output of the context condition determination program 320. become. The selector 381 selects the composite condition judgment result 206, and when c2 is set in the multiplexer 391, X2 is selected, that is, the matching result 20 of the character string matching circuit 200 is selected.
5 is selected, and when the multiplexer 392 is set to b3, Y2 is selected, that is, the context condition determination program 320
Otherwise, select Z2, that is, the output of the logic condition determination program 330. As described above, the settings of selectors 380 and 381 are performed according to the settings of multiplexers 391 and 392.

[0115] Depending on the combination of complex conditions, multiplexers are selected as follows. (1) No compound condition ────────── Select only c2 (2) Only logical condition ────────── Select only a2 (3) Neighborhood condition and context condition ──── ──b2 and b3
Select (4) Neighborhood conditions, context conditions, and logical conditions ─b2 and a3
As described above, the multiplexers 391 and 392 given as the search control information 202 from the search control means 101
Based on the setting information of selectors 381 and 382, complex condition determination programs such as neighborhood condition determination program 310, context condition determination program 320, and logical condition determination program 330 are
By selectively connecting the microprocessors 380 and 381, the fourth embodiment always uses two microprocessors even though only one microprocessor is used when only one of the complex conditions is set. To solve the drawbacks of performing pipeline processing of complex conditions using a microprocessor, to realize efficient complex condition judgment processing, and to realize a relatively low-cost and relatively high-speed document retrieval device. Can be done.

Next, regarding the sixth embodiment of the present invention, FIG.
Explain using. In this embodiment, the complex condition determination circuit 30
In 0e, two complex condition judgment programs, the context condition judgment program 320 and the logical condition judgment program 330, are installed in one microcomputer, and these compound conditions are switched and processed in order.
Two complex condition judgment processes, a context condition and a logical condition, are combined into one
The aim is to realize this using a single microcomputer. Further, the neighborhood condition determination program 310 is placed in another microcomputer. Although the complex condition determination circuit 300e of this embodiment is inferior to the first embodiment in terms of processing speed, since it can be implemented with two microcomputers, it is possible to maintain relatively high performance and reduce costs. .

This embodiment consists of a character string matching circuit 200 and a complex condition judgment circuit 300e.
e is a microcomputer MPUa301, a microcomputer MPUb306, and a character string matching circuit 20.
It is composed of buffers 350 and 360 that exchange data with 0. The microcomputer MPUa 306 includes two complex condition determination programs, a context condition determination program 320 and a logical condition determination program 330.
A scheduler 342 for switching these programs is installed. Here, the buffer 370 is secured in the work area of the program, but by using it as a first-in first-out (FIFO) memory in the program, the same function as in the first embodiment can be obtained. The scheduler 342 first executes the context condition determination program 320, then executes the logical condition determination program 330, and then executes the context condition determination program 330.
The compound condition program is switched in the order of returning to step 20. Further, the timing for switching the compound condition program in the scheduler 342 can be set every time n pieces of collation term information are processed or every time n documents are processed. This is determined in consideration of the program switching processing time of the scheduler 342. If the programs are switched frequently, the program switching time will occupy a large proportion of the execution time of the complex condition processing program.
Alternatively, it is effective to switch programs between dozens and hundreds of documents. As described above, the complex condition determination circuit 300
Although the processing speed is lower than that of the first embodiment, by configuring e, two microcomputers can execute the neighborhood condition determination program 310 and the context condition determination program 320.
, and the logical condition determination program 330, it is possible to realize a relatively high-speed document retrieval device at low cost.

Next, regarding the seventh embodiment of the present invention, FIG.
Explain using. This embodiment solves the problem in the sixth embodiment that pipeline processing of complex conditions is always performed by two microprocessors even when only one of the complex conditions is set. The purpose of this invention is to solve the problem by bypassing the complex condition determination program if there is a complex condition that has not been determined, thereby realizing a complex condition determination circuit 300f with high processing efficiency. This embodiment consists of a character string matching circuit 200 and a complex condition determining circuit 300f.
60, and further includes multiplexers 391 and 392 and selectors 380 and 381.

The bypass function of the complex condition determination circuit, which is a feature of this embodiment, will be described. This function uses a multiplexer or selector to bypass the microprocessor on which a complex condition judgment program that is not used for the specified search condition expression is installed. This is a function that only executes. For example, when the search condition expression "Document [4C] understanding" is specified, only the neighborhood conditions are used, so the collation information 205 obtained from the string matching circuit 200 is sent to the neighborhood condition determination program 310 via the buffer 350. input, and the collation information of this output is sent to the buffer 37.
The composite condition determination result 206 is sent directly via 0. Furthermore, when the search condition expression "understand [AND] system" is specified, since a logical condition is used, the collation information 205 obtained from the character string collation circuit 200 is sent to the buffer 35.
0 directly to the context condition determination program 320, and further input to the logical condition determination program 330 via the buffer 370, and this logical condition determination program 330
The output verification information is sent as the composite condition determination result 206.

A specific method for realizing the bypass function of such a complex condition determination circuit will be described. Such an operation is realized by setting multiplexers 391 and 392 and selectors 380 and 381. This setting information is given as search information 202 from the search control means 101 in FIG. In the multiplexer 391, the character string matching circuit 2
The output of the collation information 205 from 00 is sent to the context condition determination program 320 if a2 is specified, to the neighborhood condition determination program 310 if b2 is specified, and c2
If specified, it will be sent directly as the output of the complex condition circuit 300. The multiplexer 392 sends the output of the neighborhood condition determination program 310 to the context condition determination program 320 when a3 is specified, and as the composite condition determination result 206 when b3 is specified.

The selector 380 selects the input of the context condition determination program 320, and when a2 is set in the multiplexer 391, selects X1, that is, the collation result 205 of the character string collation circuit 200, and selects the input of the context condition determination program 320.
When a3 is set in 92, Y1, that is, the output of the neighborhood condition determination program 310 is selected. selector 3
At 81, the composite condition judgment result 206 is selected, and when c2 is set in the multiplexer 391, X2 is selected, that is, the matching result 205 of the character string matching circuit 200 is selected.
When the multiplexer 392 is set to b3, it selects Y2, that is, the output of the neighborhood condition determination program 310; otherwise, it selects Z2, that is, the output of the logical condition determination program 330.

As described above, selectors 380 and 38
The setting of 1 is performed according to the settings of multiplexers 391 and 392. In other words, the multiplexer is selected as follows based on the combination of complex conditions. (1) No complex conditions ──────────Select only c2 (2) Only neighborhood conditions ─────────b2 and b3
(3) Context condition and logical condition ──────Select only a2 (4) Neighborhood condition, context condition, and logical condition ─b2 and a3
As described above, the multiplexers 391 and 392 given as the search control information 202 from the search control means 101
Based on the setting information of selectors 381 and 382, complex condition determination programs such as neighborhood condition determination program 310, context condition determination program 320, and logical condition determination program 330 are
By selectively connecting the microprocessors 380 and 381, the sixth embodiment always uses two microprocessors even though only one microprocessor is used when only one of the complex conditions is set. This solves the drawbacks of performing pipeline processing of complex conditions using a microprocessor, makes it possible to realize efficient complex condition judgment processing, and provides a relatively high-speed document search device at low cost. .

Next, regarding the eighth embodiment of the present invention, FIG.
Explain using. In the first embodiment, the matching information output from the character string matching circuit 200 includes matching information for context identification character strings that are not subject to neighborhood condition processing, that is, context marker matching information, so the neighborhood condition determination program 3
In No. 10, the neighborhood condition determination process is also performed on the collation information of the context marker, resulting in a problem that the processing speed of the neighborhood condition determination decreases. In this embodiment, as a complex condition determination method that solves the above problem, a buffer 380 is provided that stores only the context marker matching information, and the context condition determination program 320 which bypasses the neighborhood condition determination processing and requires the context marker matching information Complex condition judgment circuit 300g that can perform high-speed complex condition judgment processing by inputting
The purpose is to realize the following.

This embodiment consists of a character string matching circuit 200 and a complex condition judgment circuit 300g.
g is composed of a microcomputer MPUa 301, a microcomputer MPUb 302a, and a microcomputer MPUc 303, buffers 350, 360, 370, and 380 for transferring data between them, a multiplexer 710, and a context marker detector 720. Multiplexer 710 is typically port a
is selected, and the collation information 205 sent from the character string collation circuit 200 is sent to the buffer 350 which is input to the neighborhood condition determination program 310. Further, when a context marker detection signal 721 is sent from a context marker detector 720 (described later), the multiplexer 710 selects port b and sends it to the buffer 380 which becomes the input of a sort merge program 730 (described later). In other words, the context marker matching information is sent to the buffer 380. moreover,
When a document identification information detection signal 722 is sent from a context marker detector 720, which will be described later, multiplexer 710 selects both port a and port b and sends the signal to buffer 350 and buffer 380 at the same time. That is, the document identification information will be sent to both buffer 350 and buffer 380.

[0125] The context marker detector 720 refers to the output of the character string matching circuit 200 and determines whether or not the output is context marker matching information. In other words, if the collation information identifier of the collation information is the same as the collation information identifier of the predetermined context marker, and both the start position information and the end position information are not 0 (zero), it is determined that it is a context marker. do. Further, if both the start position information and the end position information are 0 (zero), it is determined that the information is document identification information. The context marker detector 720 is composed of a comparator for detecting document identification information, two comparators for identifying collation term information of context markers, a register for storing identifiers of context markers, and a register for storing 0 (zero). . First, the comparator for detecting document identification information checks whether the position information of the verification information is 0, and if it is 0, outputs the document identification information detection signal 722. That is, the register storing 0 and the position information of the collation information are compared, and if they are equal, the document identification information detection signal 722 is output. Next, the comparators for identifying the collation term information of the context marker include a comparator that checks whether the context marker is a collation term information identifier, and a comparator that checks whether the collation information that is currently being checked is collation term information. Here, the context marker detection signal 721 is output only when establishment signals are output from both of these comparators. In the comparator that checks whether the collation term information identifier of a context marker is the same as the context marker identifier, if the collation information identifier of the collation information is the same as the context marker identifier, in other words, the comparator compares the collation information identifier of the collation information with the register for storing the context marker identifier, If they are equal, an established signal is output. A comparator that checks whether the information is matching term information compares the position information of the matching information with a register storing 0 if the position information of the matching information is not 0, and outputs an establishment signal if they are not equal. The context marker detector 720 is configured as described above.

When the context marker matching information is input to the context marker detector 720, a context marker detection signal 721 is output to the multiplexer 710. In response, the multiplexer 710 switches the destination of the collation information 205 sent from the character string collation circuit 200 from the buffer 350 to the buffer 380. If it is document identification information, a document identification information detection signal 722 is output to multiplexer 710. Accordingly, multiplexer 7
In step 10, the destination of the collation information 205 sent from the character string collation circuit 200 is set to both the buffer 350 and the buffer 380.

[0127] In the microcomputer 302a, the sort merge program 730 and the context condition determination program 3
20 and a scheduler 342 that controls these are executed. In the sort merge program 730, the buffer 360
The matching information as the output of the neighborhood condition determination program 310 stored in the buffer 380 and the context marker matching information stored in the buffer 380 are merged in ascending order of the end position information. That is, the verification information is read one by one from the buffer 360 and the buffer 380, and the end position information is compared. If the end position information of the verification information read from the buffer 360 is smaller than the end position information of the verification information read from the buffer 380, the buffer 360
The collation information read from 0 is first output to the buffer 390, then the collation information is read from the buffer 360, compared with the collation information read from the buffer 380 earlier, and the smaller one is output to the buffer 390. Conversely, if the end position information of the collation information in the buffer 380 is smaller than the end position information of the collation information in the buffer 360, the buffer 38
Output the verification information of 0 to the buffer 390, then read the verification information from the buffer 380, and similarly output the verification information of 0 to the buffer 390.
60 and outputs the smaller one to the buffer 390. Furthermore, only the document identification information read from the buffer 360 is output to the buffer 390, and the document identification information read from the buffer 380 is not output to the buffer 390. Furthermore, the sort merge performed here is performed for each document.

By performing these processes, buffer 3
90, the collation information of the buffers 360 and 380 is sorted and merged based on the end position information, and the first
The same verification information as that output by the neighborhood condition determination program 310 in the embodiment is stored. The program switching order in the scheduler 343 is such that the sort-merge program 730 is first executed, the context condition determination program 320 is then executed, and the programs are switched back to the sort-merge program 730. Further, the timing at which the program in the scheduler 343 is switched may be set every time n pieces of collation term information are processed or every time n documents are processed. This is determined in consideration of the program switching processing time of the scheduler 343. If the program is switched frequently, the program switching time will occupy a large proportion of the program execution processing time, so it may be difficult to switch programs when there are hundreds to thousands of matching information or tens to hundreds of documents. Effective. By realizing the complex condition determination circuit 300g as described above, it is possible to bypass the neighborhood condition determination program 310 with the context marker collation information that is not necessary for the neighborhood condition determination process, and the document processing speed is faster than in the first embodiment. A search device can be realized.

Next, regarding the ninth embodiment of the present invention, FIG.
Explain using. This embodiment, like the eighth embodiment,
Since the matching information output from the character string matching circuit 200 includes matching information of context markers that are not subject to neighborhood condition processing, the neighborhood condition determination program 310 also performs neighborhood condition determination processing on the matching information of context markers. In order to solve the problem of the first embodiment that the processing speed of neighborhood condition judgment is slow, a buffer is provided to store only the context marker matching information, bypassing the neighborhood condition judgment processing and storing the context marker matching information. The purpose is to realize a complex condition determination circuit 300h that can perform high-speed complex condition determination processing by inputting only the required context condition determination program 320.

This embodiment differs from the eighth embodiment in that a context marker character string matching circuit 2 is provided exclusively for context marker detection.
This is the point where 00a was provided. In the eighth embodiment, since all search terms including context markers are set in the string matching circuit 200, there is a problem that the number of search terms set in the string matching circuit 200 increases. Furthermore, there may be cases where the number of search terms exceeds the number allowed by the character string matching circuit 200. Furthermore, whereas the search term for a context marker only needs to be set once, if the search term for a context marker is reset each time a search condition is given as in the eighth embodiment, the search information 202 is created. There also arises the problem that the time and setting time become longer.

The present embodiment is composed of a character string matching circuit 200, a character string matching circuit for context marker 200a, and a complex condition determining circuit 300h. Complex condition determination circuit 300
h is composed of a microcomputer MPUa301, a microcomputer MPUb302a, a microcomputer MPUc303, and buffers 350, 360, 370, and 380 for exchanging data between them. The configuration of the context marker string matching circuit 200a is the same as the string matching circuit 200, and the context marker string matching circuit 200a is set with a context marker as a search term, and the string matching circuit 200 is set with a context marker other than the context marker. Set search terms. In addition, the search control means 1
In 01, search information 202 is set in the character string matching circuit 200 every time a search condition is given, but the character string matching circuit 200a for context marker is set only once when the present search device is started up.

The processing procedure of this embodiment will be explained using a specific example. First, the operations of the character string matching circuit 200 and the context marker character string matching circuit 200a will be explained using a specific example. (7-1)
The formula “Q=((Document [4C] Understanding) [S] System) [AN
D] (Document [S] search)" is used as an example. Each compound condition determination program is analyzed by the search control means 101, and conditional expressions separated into each condition are set. In this example, search The control means 101 sets the four search terms T1: document, T2: understanding, T3: search, and T4: system in the string matching circuit 200, and performs string matching for context markers. "S1:." is set as a search term in the circuit 200a.

[0133] Now, if the document shown in (7-2) ``...'' is a search system using document understanding is input, the string matching circuit 200 outputs the following matching result. Information (13-1) to (13-5) are verification information 2
It is output to the buffer 350 as 05. (D1, 0, 0) (13-1) (T1
,31,32) (13-2)(T2,33,3
4) (13-3) (T3, 39, 40)
(13-4) (T4,41,44) (13-
5) In addition, the following context marker matching information (12-1) to (12
-3) is output to the buffer 380 as verification information 205a. (D1, 0, 0) (12-1) (S1
,30,30) (12-2)(S1,48,4
8) (12-3) The above buffers 360, 38
Similar to the eighth embodiment, the matching information of 0 is processed by the neighborhood condition determination program 310, the sort merge program 730,
Processed by a context condition determination program 320 and a logical condition determination program 330. By providing the character string matching circuit 200a dedicated to context marker detection as described above, it becomes possible to bypass the neighborhood condition judgment program 310 with the context marker matching information that is not necessary for the neighborhood condition judgment process. Since it takes less time to create and set the search information 202 regarding context markers than in the embodiment, a faster document retrieval device than in the first embodiment can be realized.

Next, FIG. 3 shows a tenth embodiment of the present invention.
4 will be used for explanation. In this embodiment, in the complex condition determination circuit 300 of the first embodiment, the microcomputer MPUa
301, since a special memory called first-in first-out (FIFO) memory is used for the buffers 360 and 370 used to transfer data between the microcomputer MPUb302 and the microcomputer MPUc303, the cost per memory capacity is low. The problem is that it costs a lot of money. The present embodiment aims to realize a low-cost complex condition determination circuit 300i by using a general memory instead.

[0135] This embodiment uses a microcomputer MPUa.
301, microcomputer MPUb302, microcomputer MPUc303, buffer 350, bus 630, and shared memory 620. In addition, in the microcomputer MPUa301, the neighborhood condition determination program 310 is
At 302, a context condition determination program 320 is executed, and at the microcomputer MPUc303, a logic condition determination program 330 is executed.

The shared memory 620 is used to transfer data between the neighborhood condition determination program 310, the context condition determination program 320, and the logical condition determination program 330. That is, the buffer 360a in the shared memory 620 is used to exchange data between the neighborhood condition determination program 310 and the context condition determination program 320, and the buffer 370a is used to exchange data between the context condition determination program 320 and the logical condition determination program 330. each used. The buffers 360a and 370a are used as first-in, first-out (FIFO) memories in the neighborhood condition determination program 310, the context condition determination program 320, and the logical condition determination program 330, respectively.
It is possible to obtain the same function as the buffer 370.

[0137] The operation of the complex condition determination process of this embodiment will be explained. The matching result of the character string matching circuit 200 is sent to the buffer 350 as matching information 205. The collation information stored in the buffer 350 is processed by the neighborhood condition determination program 310, and the determination result is stored in the buffer 360a in the shared memory 620. Next, when the collation information is stored in the buffer 360a, the context condition determination program 320 is executed, and the determination result of the context condition determination program 320 is stored in the buffer 370a in the shared memory 620. Furthermore, when the collation information is stored in the buffer 370a, the logical condition determination program 330 is executed, and the determination result of the logical condition determination program 330 is sent out as the composite condition determination result 206. As described above, the complex condition determination circuit 30
By realizing 0i, low-cost ordinary memory can be used in buffers 360a and 370a instead of FIFO memory.
This makes it possible to realize a low-cost document retrieval device.

Finally, an eleventh embodiment of the present invention will be explained using FIG. 35. In the complex condition determination circuit 300 of the first embodiment, the microcomputer MPUa301,
Because a special memory called first-in first-out (FIFO) memory is used for the buffer 360 and buffer 370 used to transfer data between the microcomputer MPUb302 and the microcomputer MPUc303, the cost per memory capacity is high. There is a problem. In this example, this FI
The purpose is to realize a low-cost complex condition determination circuit 300j by using a general memory instead of the FO memory. In addition, in the tenth embodiment, the shared memory 620 is accessed by three microprocessors in a time-sharing manner, so if each processor accesses the memory the same number of times, the number of memory accesses will triple and the processing speed will increase. There is a problem that it becomes a memory access bottleneck. In contrast, this embodiment aims to solve this problem by providing two buffers using general memory between each processor.

[0139] This embodiment uses a microcomputer MPUa.
301, microcomputer MPUb302, microcomputer MPUc303, buffer 350, buffer 360b, and buffer 370b. In addition, in the microcomputer MPUa301, the neighborhood condition determination program 310
The Ub302 executes a context condition determination program 320, and the microcomputer MPUc303 executes a logical condition determination program 330.

The buffer 360b is used to exchange data between the neighborhood condition determination program 310 and the context condition determination program 320, and the buffer 370b is used to exchange data between the context condition determination program 320 and the logical condition determination program 330. The buffer 360b includes a multiplexer 630, a selector 631, a memory 622,
623, buses 640, 641, and communication memory 624
It consists of Buffer 370b is also buffer 36
It is configured similarly to 0b. The memories 622 and 623 that constitute the two-sided buffer are used by the neighborhood condition determination program 31.
0 to the context condition determination program 320. While the neighborhood condition determination program 310 is outputting the collation information to the memory 622, the context condition determination program 320 reads from the memory 623 the collation information previously output by the neighborhood condition determination program 310. Further, while the neighborhood condition determination program 310 is outputting collation information to the memory 623, the context condition determination program 32
0 is the previous neighborhood condition determination program 3 from the memory 622.
Read the verification information output by 10. Communication memory 624
is used to exchange control information for switching the buffers in the memory 622 and memory 623.

The multiplexer 630 switches the switching signal 63
When 0a is set to 0, port a is selected, and the collation information output from the neighborhood condition determination program 310 is stored in the memory 622.
is stored in When the switching signal 630a is set to 1, port b is selected, and the verification information output from the neighborhood condition determination program 310 is stored in the memory 623. When the selector 631 sets the switching signal 631a to 0, port x is selected, and collation information is read from the memory 622 by the context condition determination program 320. When the switching signal 631a is set to 1, port y is selected, and the collation information from the memory 622 is sent to the context condition determination program 32.
Read by 0.

A method of controlling the operation of the buffer 360b using the two-sided buffer method will be described below. First, the neighborhood condition determination program 310 sets the switching signal 630a to 0.
is sent to the multiplexer 630, that is, the collation information output from the neighborhood condition determination program 310 is sent to the memory 62.
Output to 2. When the neighborhood condition determination program 310 has finished writing a predetermined amount of collation information into the memory 622, the context condition determination program 320 passes through the communication memory 624.
to inform that the memory 622 is available for use. The context condition determination program 320 receives this and sets 0 to the selector 631 as the switching signal 631a. That is,
Memory 62 as an input to the context condition determination program 320
2 is selected, the collation information in the memory 622 is read and context condition determination processing is performed.

Next, the neighborhood condition determination program 310 sends 1 to the multiplexer 630 as a switching signal 630a. That is, the collation information output from the neighborhood condition determination program 310 is output to the memory 623. When the neighborhood condition determination program 310 finishes writing a predetermined amount of collation information to the memory 623, the information is written to the context condition determination program 320 via the communication memory 624.
inform you that it can be used. The context condition determination program 320 receives this and sends a switching signal 6 to the selector 631.
31a is set to 1. That is, the memory 623 is selected as an input to the context condition determination program 320, and the collation information in the memory 623 is read and context condition determination processing is performed.

After that, the neighborhood condition determination program 31 is executed again.
0 as the switching signal 630a to the multiplexer 6
Send it to 30. At this time, the neighborhood condition determination program 310
writes the collation information to the memory 622, the neighborhood condition determination program 310 waits to write the collation information to the memory 622 until the context condition determination program 320 reports that the reading of the memory 622 is completed. Therefore, in the context condition determination program 320, the memory 62
When the reading of 2 is completed, via the communication memory 624,
It is notified that reading of the memory 622 has been completed. By controlling the switching of the buffers in the memory 622 and the memory 623 in this way, the neighborhood condition determination program 3
A two-sided buffer system is realized in which the context condition determination program 320 and the context condition determination program 320 do not access the same memory. A buffer 37 used for exchanging data between the context condition determination program 320 and the logical condition determination program 330
0b can also use the same buffer as the buffer 360b. As described above, by realizing FIFO
Use low-cost regular memory as a buffer instead of memory3
By configuring the complex condition determination circuit 300j using 60b and 370b, a low-cost and high-speed document search device can be realized.

[0145]

[Effects of the Invention] As described above, according to the document retrieval method and device provided by the present invention, it is possible to easily determine complex conditions such as neighborhood conditions, context conditions, and logical conditions.
In addition, it is possible to perform judgment processing at high speed, and it is possible to provide a document search device that can quickly perform detailed searches unique to full-text searches.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 2 is an explanatory diagram of a document search device.

FIG. 3 is an explanatory diagram of an example of determination of complex conditions.

FIG. 4 is an explanatory diagram of collation information.

FIG. 5 is an explanatory diagram of an example of determination of complex conditions.

FIG. 6 is an explanatory diagram of an example of determination of complex conditions.

FIG. 7 is an explanatory diagram of an example of determination of a complex condition.

FIG. 8 is an explanatory diagram of a document identifier detection circuit.

FIG. 9 is an explanatory diagram of a search term length table.

FIG. 10 is an explanatory diagram of a position information adding circuit.

FIG. 11 is an explanatory diagram of document identification information.

FIG. 12 is an explanatory diagram of verification term identification information.

FIG. 13 is an explanatory diagram of collation information.

FIG. 14 is an explanatory diagram of collation information.

FIG. 15 is an explanatory diagram of neighborhood condition determination processing.

FIG. 16a is a part of an explanatory diagram of neighborhood condition determination processing.

FIG. 16b is another part of an explanatory diagram of the neighborhood condition determination process.

FIG. 17 is an explanatory diagram of neighborhood condition determination processing.

FIG. 18 is an explanatory diagram of context condition determination processing.

FIG. 19 is an explanatory diagram of context condition determination processing.

FIG. 20a is a part of an explanatory diagram of context condition determination processing.

FIG. 20b is another part of an explanatory diagram of context condition determination processing.

FIG. 21a is a part of an explanatory diagram of context condition determination processing.

FIG. 21b is another part of an explanatory diagram of context condition determination processing.

FIG. 22 is an explanatory diagram of context condition determination processing.

FIG. 23 is an explanatory diagram of logical condition determination processing.

FIG. 24a is a part of an explanatory diagram of logical condition determination processing.

FIG. 24b is another part of an explanatory diagram of logical condition determination processing.

FIG. 25 is an explanatory diagram of logical condition determination processing.

FIG. 26 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 27 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 28 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 29 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 30 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 31 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 32 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 33 is an explanatory diagram of a complex condition determination circuit using the present invention.

FIG. 34 is an explanatory diagram of a buffer between processors.

FIG. 35 is an explanatory diagram of a buffer between processors.

[Explanation of symbols]

101 Search control means 104 Storage device control means 105 String storage means 200 String matching means 300 Complex condition determination means 210 Document identifier detection circuit 220 Term comparator 230 Character number counter 240 Location information addition circuit 250 Search term length table 310 Neighborhood condition determination program 320 Context condition determination program 330 Argument condition judgment program 350 buffer 360 buffer 370 Buffer 380 buffer

Claims (16)

[Claims] [Claim 1] In a document search method that searches for documents that include a search term specified in a search condition expression in a document database stored as character codes, when the specified search term is matched in a document, A character string matching step that outputs document identification information including a document identifier, the identifier of the search word that was matched, and the matching position in the document as matching information; It consists of a complex condition judgment step that judges the search condition regarding the positional relationship between the search terms specified in the search condition expression, creates collation information of the judgment result indicating that the search condition is met, and outputs it as the search result. A document retrieval method characterized by: [Claim 2] The document retrieval method according to claim 1, characterized in that the compound condition determination step includes a neighborhood condition determination step of determining a proximity distance condition between search terms specified in the search condition expression. Document search method. [Claim 3] As the complex condition determination step in the document search method according to claim 1, a co-occurrence condition of the search terms specified in the search condition expression in the same phrase, the same sentence, or the same paragraph is determined. A document retrieval method comprising a step of determining a context condition. 4. A document characterized in that the document search method according to claim 1 has a logical condition determining step for determining a logical condition between search terms specified in a search condition expression as the complex condition determining step. retrieval method. 5. In the document retrieval method according to claim 1, the complex condition determination processing step includes a neighborhood condition determination step of determining a proximity distance condition between search terms specified in the search condition expression; a context condition determination step for determining co-occurrence conditions of the search term specified in the same phrase, the same sentence, or the same paragraph; and a logical condition between the search terms specified in the search condition expression. A document retrieval method characterized by comprising a logical condition determination step of determining. [Claim 6] As the compound condition determination step in the document search method according to claim 2, when the document data is Japanese, a determination is made regarding the proximity distance condition expressed by the number of characters between the search words specified in the search condition expression. A document retrieval method comprising a neighborhood condition determination step. [Claim 7] As the complex condition determination step in the document search method according to claim 2, when the document data is in English, a determination is made regarding the proximity distance condition expressed by the number of words between the search terms specified in the search condition expression. A document retrieval method comprising the step of determining a word spacing condition. [Claim 8] In the document retrieval method according to claim 5, in the character string matching step, when a specified search term is matched in a document, a document identifier including an identifier of the document, an identifier of the matched search term, and The first and last character positions of the matching search term in the document are output as matching information, and if a context condition is specified and a string identifying the context is matched, the context identifier matched with the document's identifier. Outputs the identifier of the character string and the start and end positions of the matching context identifying character string in the document as matching information, as well as a neighborhood condition judgment step, a context condition judgment step, and a logical condition judgment step that constitute the compound condition judgment step. In the step, the proximity condition determination step determines the proximity distance condition expressed as the number of characters between the search terms specified in the search condition expression based on the matching information output in the string matching step, and The first character position of the search word located at In the determination step, the co-occurrence conditions of the search terms specified in the search condition expression in the same phrase, the same sentence, or the same paragraph are determined based on the collation information output in the neighborhood condition determination step, and the conditions are determined. The first character position of the context identification character string located at the front and the last character position of the context identification character string located at the back that match are used as matching information, and these are added to the matching information output in the neighborhood condition determination step. In the logical condition determination step, the logical conditions between the search terms specified in the search condition expression are determined based on the collation information output in the context condition determination step, and the document units that match the conditions are determined. A document search method characterized by outputting collation information as final search result information. [Claim 9] In a document search device that searches a document database stored as character codes for documents that include a search term specified in a search condition expression, when the specified search term is matched in a document, A character string matching means that outputs document identification information including a document identifier, an identifier of a search word that has been matched, and a matching position in the document as matching information; It is comprised of a compound condition determining means that determines a search condition regarding the positional relationship between the search terms specified in the search condition expression, creates collation information of the determination result indicating that the search condition is met, and outputs it as a search result. A document retrieval device characterized by: 10. The document retrieval device according to claim 9, characterized in that the compound condition determining means includes a neighborhood condition determining means for determining a proximity distance condition between search terms specified in the search condition expression. Document search device. [Claim 11] The compound condition determination means in the document retrieval device according to claim 9 determines whether the search terms specified in the search condition expression co-occur in the same phrase, the same sentence, or the same paragraph. A document retrieval device comprising a context condition determining means. 12. The document retrieval device according to claim 9, further comprising a logical condition determining means for determining a logical condition between search terms specified in a search condition expression, as the compound condition determining means. Search device. 13. The compound condition determination processing means in the document retrieval device according to claim 9, comprising: a proximity condition determination means for determining a proximity distance condition expressed by the number of characters between search words specified in the search condition expression; a context condition determining means for determining co-occurrence conditions of the search term specified in the search condition expression in the same phrase, same sentence, or same paragraph; and a context condition determination means for determining the co-occurrence condition of the search term specified in the search condition expression; A document retrieval device characterized by comprising a logical condition determining means for determining logical conditions between. 14. The compound condition determination means in the document search device according to claim 10, when the document data is Japanese text, determines the proximity distance condition expressed by the number of characters between search words specified in the search condition expression. A document retrieval device characterized in that it is provided with a neighborhood condition determination means for determining a neighborhood condition. 15. The compound condition determination means in the document search device according to claim 10, when the document data is in English, determines the proximity distance condition expressed by the number of words between the search terms specified in the search condition expression. A document retrieval device characterized in that it has a word spacing condition determination means. 16. The document retrieval device according to claim 13, when the specified search term is matched in a document, the character string matching means includes an identifier of the search term matched with an identifier of the document, and an identifier of the search term matched with the identifier of the document. Outputs the first character position and last character position of the matching search term in The neighboring condition determining means constituting the compound condition determining means is configured to output an identifier and the start and end positions of the matching context identification character string in the document as matching information, Based on the matching information, the proximity distance condition expressed as the number of characters between the search words specified in the search condition expression is determined, and the first character position of the search word located before the search word that matches the condition and the search word located after the search condition are determined. The context condition determining means constituting the compound condition determining means is configured to output the position of the last character of a word as verification information of the determination result, adding this to the verification information output by the character string verification means. , based on the collation information output by the neighborhood condition determining means, determine whether the search terms specified in the search condition expression co-occur in the same phrase, the same sentence, or the same paragraph, and match the condition. The first character position of the context identification character string located at the front and the last character position of the context identification character string located at the rear are used as verification information, and these are added to the verification information output by the neighborhood condition determination means and output. The logical condition determining means constituting the compound condition determining means determines the logical conditions between the search terms specified in the search condition expression based on the matching information output by the context condition determining means. What is claimed is: 1. A document retrieval device characterized in that it is configured to perform the following steps and output collation information for each document that meets conditions as final retrieval result information.