JP2006040081A - Information retrieval device, database retrieval preprocessing circuit, and information retrieval method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device and a method, allowing execution of retrieval even when the retrieval execution becomes impossible in a conventional method because resource consumption exceeds an upper limit value usable in the device if based on a normal retrieval designation, in the retrieval of a composite database. <P>SOLUTION: This information retrieval device has: a retrieval expression division part 104 for reading a retrieval expression describing desired information retrieval, and dividing it into a prescribed simple format; a retrieval expression evaluation part 103 for evaluating whether a necessary resource in time of the retrieval by the divided simple-format retrieval expression is within the resource upper limit value of a system or not; and a retrieval processing part for performing the desired information retrieval according to a retrieval instruction list output by the retrieval expression division part. The retrieval expression division part performs the division into the simple format wherein the retrieval expression evaluation part evaluates that it is within the resource upper limit value, and the retrieval processing part 105 performs the information retrieval on the basis of the simple format. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to facilitating data retrieval in a multi-database system, and more particularly to facilitating retrieval using a large amount of resources due to complicated retrieval conditions.

Conventionally, with regard to data search in a multi-database system, many search devices and methods have been proposed that reduce search costs, that is, search time, resources such as memory at the time of search, and narrow down to appropriate targets according to the search intention. Yes.
For example, according to “Database inquiry optimization system” of Patent Document 1, a method for optimizing the cost required for data inquiry of a multi-database system has been proposed. In the method of the above-mentioned patent document 1, as a first stage of optimization, from the data inquiry language such as SQL, first, considering the optimization of the consumption of the entire resource, for example, it is branched deeply to the left. Generate many query trees once. Then, as a second step, considering a reduction in response time, a deep query tree on the left is converted into a balanced tree. At this time, in the above document, a bottom-up approach or a top-down approach is shown as the second stage balanced tree conversion system.

By the way, in recent years, with the increase of multimedia data to be distributed, enlarging data scale stored in a database (DB) and diversification of search objects are predicted. In this case, the amount of resources consumed during the search increases inevitably due to factors such as an increase in data registered in these DBs, application of a search method with a high search processing load, and complicated search expressions. In that case, in a conventional search device that optimizes a search expression, the upper limit value of resources that can be used by the apparatus may be exceeded due to the complexity of the search expression.
JP-A-8-328920

  In the conventional data search apparatus and method, when a desired information is obtained from a composite database composed of a plurality of databases using a search expression such as SQL as an input, various search expressions such as the above-mentioned Patent Document 1 are used. Optimization methods have been considered. However, the search method optimization method is to minimize search time and resource usage. Therefore, there is no mention of a case where the search condition becomes complicated and the resource usage exceeds the available upper limit, and therefore the search may not be performed due to a complicated search expression or a large amount of databases. There is a problem that arises.

The present invention has been made to solve the above-described problem. In the search of a composite database, the resource consumption exceeds the upper limit value that can be used in the apparatus based on the normal search specification. An object of the present invention is to obtain an apparatus and a method that enable execution of a search even when it becomes impossible.
It is another object of the present invention to obtain an apparatus and a method for speeding up the search processing that can be performed while making it possible to search conditions that cannot be searched as they are.

An information retrieval apparatus according to the present invention is an information retrieval apparatus for retrieving desired information from a database.
A retrieval formula that reads the retrieval formula describing the desired information retrieval, and divides the retrieved retrieval formula into a predetermined simple format;
A search expression evaluation unit that evaluates whether the necessary resources at the time of search by the simple search expression divided by the search expression dividing unit are within the resource upper limit value of the system;
A search processing unit that performs the desired information search according to a search command list output by the search expression dividing unit,
The search expression dividing unit is divided into simple formats that the search expression evaluating unit evaluates to be within the resource upper limit value, and the search processing unit performs information search based on the simple format.

  According to the present invention, the search expression dividing unit and the search expression evaluating unit are provided, and the search expression dividing unit divides the search expression evaluation unit into a simple format that evaluates that the value is within the resource upper limit value, and performs a search process. Therefore, there is an effect that the search processing can be performed by making the search possible even in a condition that cannot be searched.

Embodiment 1 FIG.
An apparatus and method capable of executing a search even when the search cannot be executed by the conventional method will be described with reference to the drawings.
The database to be searched by the apparatus according to the present embodiment is a composite DB that collates a plurality of individual DBs that store multimedia data. The multimedia data refers to digitized data such as document data, image data, audio data, video data, and other binary data. Further, the composite DB may be configured on the same device as the individual DB, or may be a form in which different individual DB devices connected via a network are gathered.

FIG. 1 is a diagram showing a configuration of an information search apparatus according to the present embodiment. In the figure, the information retrieval apparatus is composed of the following elements. That is, a search expression input unit 102 that receives an input of the search expression 101 from a user who requests a search, and a search expression evaluation section 103 that evaluates resource consumption at the time of search by the input search expression, which is an important element of the present invention. The search formula dividing unit 104 that divides the search formula and outputs a partial search formula and a search command list, which are also important elements, according to the input search formula and the search command list output from the search formula splitting unit 104 A search processing unit 105 that executes the actual search process, and a search result output unit 106 that outputs the search result 112 to the user who requested the search.
The search processing unit 105 further includes the following elements. That is, a search control unit 107 that controls search processing according to a search expression or a search command list, a search execution unit 109 that executes a search for one or more types of individual DBs 111 associated with a composite DB, and a search execution unit 109 An intermediate result management unit 110 that manages the obtained search results as intermediate results, and an intermediate result totaling unit 108 that executes a totaling (logical operation) process between the intermediate results.
Note that the search execution unit 109 may include a search function for an individual DB, or may call a search function of an individual DB on another device to obtain a search result.
In addition, each unit such as the search expression evaluation unit is described by a software program that realizes the function, and although not illustrated, there is a processor (CPU) and a memory, and a program in which each unit is described. To obtain each function such as search expression evaluation.

The resource mentioned here may be a memory having a problem in capacity when used as a buffer by the search processing unit 105 or the search execution unit 109 mounted on the information search apparatus, or temporarily used. Other memories may be used. Further, it may be a disk device for storing temporary files. Alternatively, it may be a CPU that is the basis of the CPU usage rate or a channel that is the basis of the network usage rate. Similarly, a memory, a disk device, a CPU, or a channel of an individual DB search device on a different device may be used.
Other factors that make the search impossible are the specifications of the information search device, such as when the length of the search expression set by the information search device is too long or the number of operators is too large. Furthermore, there are cases where the number of search keywords or the number of unique characters is too large when the individual DB performs a full text search.

In the evaluation of resource consumption at the time of search, the search expression evaluation unit 103 holds resource information at the time of execution of search processing for individual DBs associated with the composite DB and totalization processing of the results. Evaluate the resource consumption during the entire search.
Specifically, for example, when the individual DB is a document search DB, the size of the index read from the disk device onto the memory at the time of the search may be calculated from the number of characters of the search keyword. In many document searches, information on the appearance positions of keywords and characters appearing in a document is stored in the DB as an index. Depending on the frequency of occurrence of the keyword in the document, the size of the corresponding occurrence position information changes, but by checking the total size of the appearance position information of the search keyword in the search formula, whether or not it exceeds the memory capacity Can be evaluated. Therefore, the size information of the appearance position information may be recorded separately from the appearance position information, or the size of the file in which the appearance position information is directly recorded may be checked. Further, a calculation formula for calculating a required memory size such as an index size may be defined in advance using the appearance frequency of the keyword or character, the number of search keywords in the search formula or the number of characters of the keyword as parameters. For example, a calculation formula such as simply taking (the proportionality constant K × appearance frequency of the keyword) as the size of the appearance position information of one search keyword and taking the sum thereof can be considered. In any case, the appearance frequency of the keyword and the size of the appearance position information can be acquired when the document is registered in the DB.

It is possible to search a predetermined percentage, for example, the first 10 percent, extract the appearance frequency corresponding to the search, and multiply it by 10 to see if it overflows. Alternatively, it is also possible to store a record of a search formula and its consumed resource amount that have been input in the past for a certain number of times, and use statistical information of the resource consumption based on these search formulas during the search. For example, the statistical information may include the number of search hits for each search expression, the transition of the consumption of memory and CPU resources during search execution, and the like. In the case of the intermediate result totaling process, it may be estimated from the expected number of search hits.
In addition, when an individual DB 111 that consumes a large amount of resources at the time of search and an individual DB that does not exist coexist, a weight for each DB may be set.
Further, when the resource is the specification of the search execution unit, for example, when the upper limit of the length of the search expression that can be accepted is set, it is only necessary to count the number of characters and the number of bytes. That is, when the upper limit of the length of a search expression that can be accepted at one time by the individual DB search execution unit 109 is determined to be N bytes, a function that acquires the length of a character string such as strlen () By acquiring the length of Expression 101, it can be easily determined whether or not the upper limit of N bytes is exceeded.
In this way, the search expression evaluation unit 103 holds the upper limit information and method for evaluation, and the resource consumption is evaluated based on the information.

FIG. 2 is an operation flowchart showing the flow of search processing performed by the search formula dividing unit 104 and the search formula evaluation unit 103 of the information search apparatus according to the present embodiment.
The operation of the information search apparatus in this embodiment will be described with reference to this figure. When the information retrieval apparatus reads a predetermined retrieval expression describing a desired information retrieval input by the retrieval expression input unit 102 in step S201 (hereinafter, description of the step is omitted), in step S202, the information retrieval apparatus converts the predetermined retrieval expression in the predetermined format. The information is decomposed into a simple format, and the search expression evaluation unit 103 evaluates the consumption amount of the resources used by the information search apparatus when searching with the search expression in the simple format in S203. If the resource consumption during the search is within the range of the upper limit value of resources that can be used by the information search apparatus (hereinafter referred to as the resource upper limit value) (No), the process proceeds to S204. Then, the search expression dividing unit 104 synthesizes the simple formats into an upper simple format. Then, the evaluation of S203 is performed, and thereafter the loop of S204 is repeated until this evaluation exceeds the resource upper limit value.
However, in the evaluation based on the search formula in S202, if the resource consumption at the time of search exceeds the upper limit value of the resources that can be used in the information search apparatus (Yes), the search formula dividing unit 104 in S205 performs the previous S203. For the search expression immediately before exceeding the resource upper limit value, the search expression is reconfigured, for example, divided into a plurality of partial search expressions so that the resource consumption at the time of the search is within the range of the resource upper limit value. If it is confirmed in S203 ′ that this partial search expression is acceptable, this partial search expression is output as a search instruction list in S206. The search processing unit 105 executes search processing according to the search command list. Finally, the search result obtained by this search process is output from the search result output unit 106 to the user, and the process ends.

The search instruction list is a list that specifies the execution order of the partial search expression execution and the intermediate result aggregation process so as to be equivalent to the search expression input by the user. FIG. 4 shows an example of a search expression and a search command list. Now, it is assumed that the search formula 401 is input and the resource consumption during the search exceeds the resource upper limit value as a result of the evaluation of the search formula. As a result of dividing the search formula 401 by the search formula dividing unit 104, a search command list 402 is output. In the example of FIG. 4, the search instruction list 402 first executes a partial search expression (A AND B) in the search execution unit, then executes a partial search expression (C AND D) in the search execution unit, and finally It specifies that the aggregation process OR of the intermediate results obtained by the previous two searches is performed. By processing like the search command list 402, the same search result as the search formula 401 can be obtained.
In order to perform simply, as described above, the division may be performed for each logic of the logical operation.
A search result or intermediate result is a set of record information that conforms to a search condition given by a search expression. The record information may include an identifier of a record in the DB, a score obtained by evaluating a record that matches the search condition, data stored in a DB column, and the like.

FIG. 3 is an operation flowchart showing the flow of search processing by the search processing unit 105 of the information search apparatus in the present embodiment. The search processing unit 105 processes a search according to a search expression input from a user, a partial search expression, and a search command list.
For example, when a search command list including a partial search expression is input, the search control unit 107 acquires one search execution command, for example, search (A AND B) from the search command list in S301. In step S <b> 302, if the acquired instruction is a content for executing a partial search expression, the process proceeds to (search), and a search execution instruction based on the partial search expression is issued to the search execution unit 109. In step S303, the search execution unit 109 generates a search plan for executing a search from a partial search formula as necessary, as described in a later embodiment. In S304, a search is performed on the individual DB 111 associated with the composite DB, and the search result is acquired. The acquired search result is stored in the intermediate result management unit 110 as an intermediate result of the search in S305.
If the acquired instruction is execution of totaling of intermediate results in S302, the process proceeds to (totaling), and a plurality of intermediate results are acquired from the intermediate result management unit 110 in S306. In step S307, the intermediate result is summed up. In step S308, the process proceeds to the next partial search expression, and the processing in step S304 or S306 is repeated. The intermediate results obtained in this way are summed up in S309 based on the final search formula. The aggregation of intermediate results may be performed sequentially in S306. Then, it is output to the search result output unit 106.

When the input to the search processing unit 105 is a search expression input from the user, S301 and S302 may be skipped, and the direct search execution unit 109 may execute S304 and subsequent steps, or S306. Further, the search result may be directly output to the search result output unit 106 in step S309 without using the intermediate result management unit 110.
The search execution unit 109 may optimize the partial search expression so that the search time by the partial search expression and the memory usage during the search are minimized in the search plan generation processing step of S303.
In search execution and intermediate result aggregation processing, depending on the results of the previous processing, there are cases where it is not necessary to execute it. In such a case, a search command that does not need to be executed may be skipped. For example, suppose that the following search expression is given, and it is divided into the following three subtrees (01), (02), and (04).
"(((A AND B) OR (C AND D)) AND (E OR F)"
(01) Search (A AND B)
(02) Search (C AND D)
(03) Aggregation OR (aggregate intermediate results of searches 1 and 2)
(04) Search (E OR F)
(05) Aggregation AND (Aggregate intermediate results of Search 4 and Aggregation 3)
At this time, for example, if the execution result of the totalization (03) is false (there is no intermediate result), the search (04) does not have to be executed.

Each processing unit of the search expression input unit 102 to the intermediate result management unit 110 in FIG. 1 may be a program executed by one processor, or may be configured to be executed by a dedicated processor. .
The search formula input by the user may be input directly by the user using an input device such as a keyboard, or a file describing the search formula may be input and read by the search formula input unit 102.
The search formula 101 inputted, the search formula transmitted from the search formula input unit 102 to the search processing unit 105, the partial search formula, and the search command list may be transmitted as information on the memory through a bus (not shown), You may transmit collectively via a temporary file. Further, it may include processing for converting a search condition input by the user through the graphical user interface into an input format of the information search apparatus.
The intermediate result management unit 110 may manage the intermediate result on the memory, or store the intermediate result as a temporary file on an external storage device such as a disk device and use it later. Search results output by the search execution unit 109 and the search processing unit 105 may also be transmitted as information on a memory or may be transmitted via a file.
The search result output unit 106 may output the result to an output device such as a screen or may output it to a file. Further, the search result may be output as information on a memory to another device that edits, processes, and uses the search result.

As described above, the information search apparatus according to the present embodiment evaluates the resource consumption during the search of the search expression input by the user, and executes the search by dividing the search expression when the resource upper limit is exceeded. Is configured to do. Therefore, in the conventional method, even when the search cannot be executed because the resource consumption during the search exceeds the resource upper limit value according to the search expression input from the user, the search can be executed.

Embodiment 2. FIG.
In the information search apparatus in the previous embodiment, the apparatus that divides the search expression and divides the search processing has been described. In this way, if the number of search expression divisions is large, the search time is expected to increase due to search initialization processing, data input / output processing to the intermediate result management unit, and the like. In other words, the smaller the number of search expression divisions, the better the overall search processing efficiency. Here, a description will be given of a method in which the information search apparatus performs division for improving the search processing by the search expression dividing unit.
In general, a logical expression can be expanded into a tree-structured logic. Therefore, in the search expression dividing method according to the present embodiment, the search expression is regarded as a logical expression of a tree structure, and is expanded into a tree structure called a binary tree. A binary tree is a tree structure in which the number of children in all nodes except leaves is 2 or less. When the search expression is expanded into a binary tree, the leaf of the tree corresponds to the search process of one search execution unit, and the internal node corresponds to the total process of the intermediate result totaling unit. Let D be the height of this binary tree.

FIG. 5 is an operation flowchart showing the search expression dividing process by the search expression dividing unit of the information search apparatus in the present embodiment.
In the figure, when a search expression is input, the search expression dividing unit 104b expands the search expression into a binary tree in step S501. In S502, a subtree having a height i of 1 is constructed. In step S503, the search expression evaluation unit 103 evaluates the resource consumption during the search for the subtree constructed in the previous step. If the resource consumption during the search by the subtree does not exceed the resource upper limit in S503 (No), a subtree having a height i + 1 is constructed in the next S504. In step S503, the tree is raised until the upper limit is reached.
If the resource consumption at the time of searching by the subtree exceeds the resource upper limit in S503 (Yes), in S505, the partial search expression by the subtree having the highest height immediately before becoming Yes is checked in S503 and a new part is obtained. Whether or not the upper limit value of the resource is exceeded by the subtree having the search expression is checked in S503 ′, and the partial search expression and the intermediate result totaling process (root) are output as a search instruction list in S506. That is, the processing of S503 and S504 is repeated from the height i = 1 (that is, the leaf-only subtree) to the maximum D. When the division of the binary tree developed in S501 is completed, the search command list is output in S506 and the process is terminated.

  In this way, the search expression is expanded into a binary tree in S501, but there are a plurality of methods for expanding the search expression into a binary tree. For example, the search expression 1101 in FIG. 13 can be expanded as a tree 1102 or a tree 1103. However, in the case of a tree deep on the left (or right) like the tree 1102, there is a possibility that the number of divisions of the search expression increases. For example, when the partial tree 1104 is divided as one partial search expression, the leaves C and D are single partial search expressions, and the division efficiency is not good. On the other hand, in the case of a balanced tree (a tree having a minimum height) such as the tree 1103, there is no problem like the tree 1102. Therefore, when expanding the search expression to a binary tree, it expands to a balanced tree such as the tree 1103.

The operation of the search expression dividing process in the present embodiment will be described with reference to FIG. As an example in that case, the search formula 401 in FIG. 4 is used.
1) When a search expression 601 is input, the search expression 601 is expanded into a binary tree 602 in S501. Here, A to D are leaves, which are search conditions for individual DBs associated with the composite DB. Internal nodes E to G are intermediate result totaling processes. In this example, the aggregation of intermediate results is an AND or OR logical operation. As an example, assume that the resource upper limit value is 100, and the resource consumption amount when searching for A to D alone is 30. In order to make the explanation easy to understand, it is assumed that the aggregation processing of intermediate results does not consume resources. In addition, the resource consumption during the search using the subtree is the sum of the resource consumption during the search for the leaf alone. The height of the tree 602 is 3.
2) In S502, a partial tree having a height of 1, that is, tree A, tree B, tree C, and tree D is constructed.

3) In S503, the resource consumption during search of the subtree constructed in S502 is evaluated. As a result of the evaluation, both are 30 and are within the range of the resource upper limit (100).
4) In S504, a subtree having a height of 2 is constructed. That is, a tree ABE and a tree CDE are constructed.
5) In S503, the resource consumption during the search of the subtree constructed in S502 is evaluated. The evaluation results are all 60 and are within the range of the resource upper limit.
6) In S504, a subtree of height 3 is constructed. That is, the tree ABCDEFG is constructed.
7) In S503, the resource consumption during the search of the subtree constructed in S502 is evaluated. The result of the evaluation is 120, which exceeds the resource upper limit (100).
8) In S505, a search expression corresponding to two height-2 subtrees in the tree ABCDEFG is constructed. This clearly has a rating of 60 and is within the resource range. In step S506, the search expression information corresponding to the height 2 subtree and the leaf G information are output as a search instruction list. That is, the partial search expression (A AND B), the partial search expression (C AND D), and the intermediate result totaling process OR are set as a search instruction list (search instruction list 402 in FIG. 4).

  The information retrieval apparatus according to the present embodiment constructs a long partial tree from the leaf to the bottom up after expanding the retrieval formula into a binary tree as described above, and when the resource consumption exceeds the resource upper limit value, Since the configuration is such that the previous longest subtree is divided, the resource consumption during the search by the partial search expression can be maximized within the range of the resource upper limit in the search expression dividing process. As a result, the number of search expression divisions can be reduced.

Embodiment 3 FIG.
For the same purpose as in the previous embodiment, an information search apparatus that performs other operations will be described.
The operation of the information search apparatus in this embodiment will be described with reference to the flowchart of FIG. In the figure, when a search expression is input, the search expression dividing unit 104c of the information search apparatus expands the search expression into a binary tree in step S701, and further scans internal nodes of the binary tree to obtain the same label. Join nodes that have a parent-child relationship. By the processing of S701, in the binary tree, the number of children of the internal node is always 2 or less, but the internal node has 2 or more children. Let D be the height of this tree. After S702, a subtree is constructed in a bottom-up manner, similar to the processing flow described in the second embodiment. In S702, a subtree having a height of 1 is constructed. In step S <b> 703, the search expression evaluation unit 103 evaluates the resource consumption during the search for the subtree constructed in step S <b> 702. If the resource consumption during the search by the subtree does not exceed the resource upper limit in S703 (No), a subtree having a height i + 1 is constructed in S704. The steps S703 and S704 are repeated until the maximum height D is reached.
If the resource consumption at the time of retrieval by the subtree exceeds the resource upper limit at S703 (Yes), the division by the subtree immediately before it is reconsidered at S705, and the division by each subtree is the resource consumption at the time of retrieval. Is the maximum in the range of the resource upper limit, that is, the subtree having the minimum number of divisions is constructed. In step S706, a search instruction list is output as information on the total processing logic and the partial search expression corresponding to the divided subtree of height i, and the division ends.

  In the process of S504 of FIG. 5 in the previous second embodiment, the tree of height i can be uniquely divided into two subtrees of height i-1. However, there are many methods for dividing the tree of height i into the tree of height i−1 in the processing of S705 of FIG. 7 in the present embodiment. An algorithm for obtaining an approximate solution in polynomial time has been devised for the node combination optimization problem that maximizes the resource consumption during subtree search within the range of the resource upper limit, such as the subtree division processing in S704. And can be processed at high speed.

FIG. 8 is a diagram illustrating an example of search expression division processing according to the present embodiment.
Assume that a search expression 801 is input. At this time, in the method of dividing the search expression into the binary tree 802 shown in the second embodiment, the binary tree 802 includes three subtrees ABG, subtree CDH, and subtree EFI. Divided into subtrees. The search command list consists of the following five processes.
(1) Search (A OR B) (Processing of subtree with G as root)
(2) Retrieval (C ORD) (Processing of subtree rooted in H)
(3) Intermediate result total OR (internal node J processing)
(4) Search (E OR F) (Processing of subtree rooted at I)
(5) Intermediate result total OR (internal node K processing)

The flow of search expression division processing in this embodiment is as follows.
11) When the search expression 801 is input, the search expression 801 is expanded into the binary tree 802 in S701. Here, A through E are leaves, which are search conditions for individual DBs associated with the composite DB. The internal nodes G to K are the intermediate result aggregation process. Similar to the second embodiment, in this example, the aggregation of intermediate results is an AND or OR logical operation. Assume that the resource upper limit value is 100, and the resource consumption of the search processing from A to E is as shown in FIG. Furthermore, a tree 803 is obtained by combining internal nodes G to K having the same label in the tree 802 and having a parent-child relationship.
12) In S702, a subtree having a height of 1 is constructed. That is, tree A to tree F.
13) In S703, the resource consumption during search of the subtree constructed in S702 is evaluated. As a result of the evaluation, the resource consumption is 30 or 40, both of which are within the range of the resource upper limit (100).

14) In S704, build a subtree of height 2. That is, tree ABCDEFK.
15) In S703, the resource consumption during search of the subtree constructed in S704 is evaluated. The result of the evaluation is 200, which exceeds the resource upper limit.
16) In step S705, the tree ABCDEFG has a height of 2 in this case, and since there is no point in setting the height to 1, the tree ABCDEFG is divided into two or more various subtrees. Here, for example, when the leaves of the tree 803 are selected in order from the left and divided into partial trees, the tree can be divided into a partial tree including the leaf AB, a partial tree including the leaf CDE, and three partial trees 803 including the leaf K. Alternatively, when the combination of leaves is changed as in the tree 804, the tree can be finally divided into two subtrees, a subtree ACDG and a subtree BEFH, as in the tree 805.
17) Since the division of the tree 803 is completed, the search command list is output in S706 and the process is terminated.
At this time, the search instruction list includes the following three processes.
(11) Search (A OR C OR D) (Processing of subtree rooted in G)
(12) Retrieval (B OR E OR F) (Processing of subtree with H as root)
(13) Intermediate result totaling OR (processing of internal node K)

  As described above, the information search device in the present embodiment expands the search expression into a binary tree, further joins nodes in the parent-child relationship having the same label, and constructs a subtree from the leaf to the bottom up, Since the subtree is divided when the resource consumption exceeds the resource upper limit value, the number of divisions can be further reduced in the search expression division processing compared to the method in the previous embodiment.

Embodiment 4 FIG.
A special form of search expression is described. In the present embodiment, the search formula dividing unit 104d of the information search apparatus transforms the search formula into a tree structure of DNF (Disjunctive Normal Form) format. DNF is a form of logical expression, also called product-sum standard form, which is a logical expression in which logical variables are combined with a logical operator AND and logical terms are combined with a logical operator OR. Can be expressed as An arbitrary logical expression can be transformed into the DNF format.

FIG. 9 is a flowchart showing the operation of the search expression dividing process by the search expression dividing unit 104d of the information search apparatus in the present embodiment. When a search expression is input, the search expression dividing unit 104d transforms the search expression into a DNF tree in S901. Subsequent to S902, a subtree is constructed bottom-up in the same manner as the processing flow described in the third embodiment. That is, in S902, a subtree having a height i is constructed. In step S903, the search expression evaluation unit 103 evaluates the resource consumption during the search for the subtree constructed in step S902. If the resource consumption during the search by the subtree does not exceed the resource upper limit in S903 (No), a subtree having a height i + 1 is constructed in S904. In this way, a tree is constructed in which the height is sequentially increased in S903 and S904.
If the resource consumption during the search by the subtree exceeds the resource upper limit in S903 (Yes), the resource consumption is maximized in the range of the resource upper limit when searching for the subtree of height i in S905. That is, it is divided into two or more height i subtrees so that the number of divisions is minimized. In this way, S503 to S505 are repeated from the height i = 1 to a maximum of 3 in this case, and information on the route (intermediate result totaling process) and information on the partial search expression corresponding to the divided subtree of height i are obtained. In step S906, a search command list is created and output.
As described above, the optimal solution can be easily found by converting the logical trees of the search formulas shown in FIGS.

The operation of the search expression dividing process in this embodiment will be described with reference to FIG.
Assume that a search expression 1001 is input. At this time, in the method shown in the previous embodiment, after the search expression is expanded into a binary tree, it is further transformed into a tree 1002 in which nodes in the parent-child relationship having the same label are combined. Here, A to E are leaves. The internal nodes J to N are intermediate result totaling processes. Similar to Embodiment 2, in this example, the aggregation of intermediate results is an AND or OR logical operation. Assume that the resource consumption of the search processing with the resource upper limit value being 100 and A to H is as shown in FIG. In this example, the resource consumption of the intermediate result totaling process is determined as follows.
AND processing: equal to the largest resource consumption of the child OR processing: equal to the sum of the resource consumption of the children In the method in the previous embodiment, the tree 1002 is a subtree ABCDL, subtree EFJ , Subtree GIK, and subtree H are divided into four subtrees. The search command list at this time is composed of the following seven processes.
(21) Search (A AND B AND C AND D)
(22) Search (E OR F)
(23) Search (G OR I)
(24) Search (H)
(25) Intermediate result totaling OR (processing of internal node K)
(26) Intermediate result totaling AND (processing of internal node M)
(27) Intermediate result totaling OR (processing of internal node N)

The operation of the search expression dividing process in the present embodiment is as follows.
21) When the retrieval formula 1001 is input, the retrieval formula 1001 is transformed into a tree 1003 in the DNF format in S901.
22) In S902, a subtree having a height of 1 is constructed. That is, tree A or tree H.
23) In step S903, the resource consumption during search of the subtree constructed in step S902 is evaluated. As a result of the evaluation, both are within the range of the resource upper limit (100).
24) In S904, a subtree having a height of 2 is constructed. That is, they are tree ABCDL, tree EIL, tree EGL, tree EHL, tree FGL, tree FHL, and tree FIL.
25) In S903, the resource consumption during the search of the subtree constructed in S904 is evaluated. The evaluation results are: tree ABCDL, tree EIL, tree FIL are 50, tree EGL, tree FGL are 40, tree EHL, and tree FHL are all within the range of the resource upper limit.
26) In step S904, a tree having a height of 3 is constructed. That is, trees A to N.
27) Again in S903, the resource consumption during search of the tree of height 3 constructed in S904 is evaluated. As a result of the evaluation, the resource consumption is 270, which exceeds the resource upper limit value.
28) In S905, the trees A to N are divided into two or more subtrees. As shown in the tree 1004, it can be divided into three subtrees: a subtree 1005, a subtree 1006, and a subtree 1007.
29) In S903 ′, when these individual subtrees are evaluated, both the subtree 1005 and the subtree 1006 become 100, which is within the resource upper limit. Therefore, since the division of the tree 1003 has been completed, the search command list is output in S906 and the process ends.

At this time, the search command list includes the following four processes.
(31) Search ((A AND B AND C AND D) OR (E AND I))
(32) Search ((E AND G) OR (E AND H) OR (F AND G))
(33) Search ((F AND H) OR (F AND I))
(34) Intermediate result total OR (root node processing)
Thus, as in Embodiment 3, the node combination optimization problem that maximizes the resource consumption during subtree search within the upper limit of the resource, such as the subtree division processing in S905, is an approximate solution. An algorithm obtained in polynomial time has been devised and can be processed at high speed.

  The information retrieval apparatus according to the present embodiment constructs a partial tree from the leaf to the bottom up after transforming the retrieval formula into a DNF-format tree as described above, and when the resource consumption exceeds the resource upper limit value, In the search expression dividing process, the resource consumption during the search by the partial search expression can be maximized within the range of the resource upper limit. As a result, the number of search expression divisions can be reduced.

  As a modification of the present embodiment, a search expression dividing method by another search expression dividing unit 104e will be described. That is, the search formula is transformed into a tree structure of CNF (Conjunctive Normal Form) format. CNF is one of the forms of logical expressions, also called the sum-of-products standard form, and is a logical expression in the form of connecting clauses that combine logical variables with the logical operator OR with the logical operator AND. It can be expressed as three trees. Arbitrary logical expressions can be transformed into CNFs.

The operation of the search expression dividing process by the search expression dividing unit 104e is equivalent to the process of S901 in FIG. 9 that “transforms the search expression into CNF”.
In the information retrieval by the retrieval formula dividing unit 104e, after the retrieval formula is transformed into a CNF format tree, a partial tree is constructed from the leaf to the bottom up, and the partial tree is divided when the resource consumption exceeds the resource upper limit value. Configure as follows.
Therefore, in the search expression dividing process, the resource consumption during the search by the partial search expression can be maximized within the upper limit value, and the number of search expression divisions can be reduced.

Embodiment 5. FIG.
In each of the previous embodiments, the information search apparatus internally constructs a subtree, generates a search expression that consumes the maximum resource within the resource upper limit value, and executes the search. In this embodiment, the search formula divided by the information search apparatus is output to a screen or a log file.
FIG. 12 is a diagram illustrating a configuration example of the information search apparatus according to the present embodiment. In this configuration, a search expression output unit 1205 that outputs a partial search expression output from the search expression dividing unit to an output device such as a screen or a log file 1208 is added to the information search apparatus in each of the previous embodiments. It has become. The other search expression input unit 1202, search expression evaluation unit 1203, search expression dividing unit 1204, search processing unit 1206, and search result output unit 1207 are the same elements as those in the previous embodiment.
The information output by the search expression output unit 1205 to an output device such as a screen or the log file 1208 includes a partial search expression output from the search expression dividing unit 1204, a search command list, and a flag indicating whether the search expression has been divided. When there are a plurality of information and search expression division methods, information such as the division method may be used. Further, the search formula input to the search formula input unit 1202 may be output as it is.

FIG. 11 is a diagram showing an operation of search processing by the information search apparatus in the present embodiment.
The operation will be described with reference to the drawings. When the search expression input unit 1202 receives the input of the search expression in S1301, the consumption amount of the resource at the time of the search by the search expression input in the search expression evaluation unit 1203 is evaluated in S1302. If the resource consumption at the time of search is within the resource upper limit value due to the evaluation of the search expression (No), the details are not shown in the figure, but as described in the previous embodiments, the subtree A partial search expression is created by sequentially increasing the height, and the process proceeds to S1304 with the subtree having the highest height within the resource upper limit. That is, in the evaluation of the search expression in S1302, when the resource consumption amount at the time of search exceeds the upper limit value that can be used (Yes), in S1303, the search expression dividing unit 1204 sets a plurality of resource consumption amounts within the upper limit value. Divide into partial search expressions.
In step S1304, the partial search expression obtained by dividing the search expression and the search command list are output to the screen, the log file, etc. 1208. In S1205, the search processing unit 1206 executes search processing according to the input search formula or the partial search formula and search command list divided in S1303. Finally, in step S1306, the search result obtained by the search processing in step S1305 is output to the user from the search result output unit 1207, and the process ends.
Note that the partial search expression output process of S1304 may be executed at any stage as long as the operation is executed after S1303.

  As described above, the information search apparatus according to the present embodiment outputs information related to the search expression divided by the search expression dividing unit on a screen display or the like, so that the user refers to the search expression division status as necessary. Can do.

In the above embodiments, each element such as the search expression dividing unit and the search expression evaluating unit has been described as having a dedicated function. However, it is also possible to use a general-purpose computer to perform an equivalent operation using a program having the functions shown in each drawing such as FIGS. 2, 3, 5, 7, 9 described in each embodiment.
Even if it does in this way, the same outstanding effect as an information search method is acquired.

Embodiment 6 FIG.
Hereinafter, a flow of a general search expression dividing process will be described with reference to FIG. The flow of processing after expanding the search expression into a logical tree is equivalent to S702 in FIG.
31) When a search expression 1401 is input, the search expression 1401 is expanded into a logical tree 1402. In this example, the aggregation of intermediate results is an AND or OR logical operation. Assume that the resource upper limit value is 100, and the resource consumption of the A to N search processing is as shown in FIG. Also, since the resource consumption of the subtree is simple, it is the sum of the resource consumption of the root of the subtree.
32) Build a subtree of height 1. That is, the subtrees A to N. When the resource consumption during the search of the subtree is evaluated, all are less than 100 and are within the range of the resource upper limit.
33) Build a subtree of height 2 That is, there are subtrees 1406, 1407, and 1408. When the resource consumption at the time of searching these subtrees is evaluated, all of them are less than 100, and therefore are within the range of the resource upper limit.
34) Build a height-3 subtree. That is, the subtrees 1409 and 1410.
When the resource consumption during retrieval of these subtrees is evaluated, the resource consumption of the subtree 1409 is 30, which is within the range of the resource upper limit. On the other hand, the resource consumption of the subtree 1410 is 170, which exceeds the resource upper limit.
35) Since the resource consumption at the time of searching the subtree 1410 exceeds the resource upper limit, the subtree is divided. As an example of optimal division, the subtrees 1412 and 1413 are divided. After the division, the divided subtree search formula and the root label information of the height 3 subtree are output to the search instruction list. At this time, the search command list is as follows.
(41) Search (subtree 1412)
(42) Search (partial tree 1413)
(43) Intermediate result aggregation AND (aggregation of execution results of subtrees 1412 and 1413)
36) Build a height-4 subtree. That is, it is a subtree 1411. When the resource consumption during the search of the subtree 1411 is evaluated, it is 60, which is within the range of the resource upper limit.
37) Build a subtree of height 5. That is, although the tree 1405 is rooted at the node T, the partial tree rooted at the node S has already been divided, and it is clear that the resource consumption when searching the tree 1405 exceeds the resource upper limit. In such a case, a search command for the subtree 1411, a search command for the subtree rooted at the node S, and information on the label of the root T are output to the search command list. That is, the search command list is as follows.
(51) Search (subtree 1411)
(52) Search (subtree 1412)
(53) Search (subtree 1413)
(54) Intermediate result aggregation AND (aggregation of execution results of subtrees 1412 and 1413)
(55) Intermediate result aggregation OR (aggregation of execution results of retrieval of 1 and aggregation of 4)

It is a figure which shows the structure of the information search device in Embodiment 1 of this invention. FIG. 6 is an operation flowchart of search processing by the information search apparatus in the first embodiment. FIG. 7 is an operation flowchart of search processing by a search processing unit in the first embodiment. It is a figure which shows the example of the search type which the information search device in Embodiment 1 and 2 of this invention handles. FIG. 10 is an operation flowchart of search expression construction and division processing performed by the information search apparatus in the second embodiment. FIG. 10 is a diagram for explaining operations of search expression construction and division processing in the second embodiment. It is an operation | movement flowchart of the search formula construction | assembly and division | segmentation process which the information search device in Embodiment 3 of this invention performs. FIG. 10 is a diagram for explaining operations of search expression construction and division processing in the third embodiment. It is an operation | movement flowchart of the search formula construction | assembly and division | segmentation process which the information search device in Embodiment 4 of this invention performs. FIG. 20 is a diagram for explaining operations of search expression construction and division processing in the fourth embodiment. It is an operation | movement flowchart of the search formula construction | assembly and division | segmentation process which the information search device in Embodiment 5 of this invention performs. FIG. 10 is a diagram showing a configuration of an information search device in a fifth embodiment. FIG. 10 is a diagram illustrating division of a search expression that is a comparison target in the second embodiment. FIG. 25 is a diagram showing a flow of general search expression dividing processing in the sixth embodiment.

Explanation of symbols

  102 Search Formula Input Unit, 103 Search Formula Evaluation Unit, 104, 104b, 104c, 104d, 104e Search Formula Dividing Unit, 105 Search Processing Unit, 106 Search Result Output Unit, 107 Search Control Unit, 108 Intermediate Result Counting Unit, 109 Search Execution unit, 110 intermediate result management unit, 111 database (DB), 1205 search expression output unit, S201 search expression reading step describing desired information search, S202 decomposition step into predetermined simplest format, S203 decomposition (synthesis) ) Evaluating whether the resource exceeds the upper limit at the time of searching in the simple format, S204 Simple format compositing step, S205 Reconstructing the search formula in the simple format immediately before, S206 Partial search formula and intermediate tabulation processing Retrieval command list output step, S501 Retrieval expression into binary tree expansion S502, subtree construction step of height i = 1, S503 step of evaluating whether the resource exceeds the upper limit at the time of retrieval by subtree description, subtree construction step of S504 i + 1, S505 the subtree just before is divided Subtree construction step with search formula, S701 Binary tree expansion of search formula and parent-child node join step with the same label, S702 subtree construction step with height i = 1, S703 Resource limit when searching by subtree description A step of evaluating whether the value is exceeded, a subtree construction step of S704 i + 1, a subtree construction step having a search expression obtained by dividing the immediately preceding subtree, S901, a step of transforming the search expression into DNF, S902 height i = 1 Subtree construction step, S903 Does the resource exceed the upper limit when searching by subtree description? Evaluating step, S904 i + 1 partial tree construction step, the partial tree construction step with S905 immediately before search expression subtree was split.

Claims (9)

In an information retrieval apparatus for retrieving desired information from a database,
A search expression dividing unit that reads the search expression describing the desired information search and divides the read search expression into a predetermined simple format;
A search expression evaluation unit that evaluates whether the necessary resources at the time of search by the simple search expression divided by the search expression dividing unit are within the resource upper limit value of the system;
A search processing unit that performs the desired information search according to a search command list output by the search expression dividing unit,
The search expression dividing unit divides the search expression evaluation unit into a simple form within evaluation that is evaluated as being within the resource upper limit value, and the search processing unit performs information search based on the simple form within evaluation. An information retrieval apparatus characterized by that.   The search expression dividing unit expands to a binary tree as a predetermined simple format, and when the search expression evaluation unit evaluates that the resource tree is within the upper limit of the resource when searching by the binary tree, the height of the binary tree is set. The above-mentioned evaluation by the search expression evaluation unit at the time of searching with the raised binary tree is repeated, and the raised binary tree that can obtain an evaluation close to the resource upper limit is output as a search instruction list. The information search apparatus according to claim 1, wherein   3. The information retrieval apparatus according to claim 2, wherein the retrieval formula dividing unit first creates a partial tree having a height of 1 when expanding into a binary tree.   3. The information search apparatus according to claim 2, wherein the search expression dividing unit creates a partial tree by combining parent and child nodes having the same label when expanding into a binary tree.   The search expression dividing unit is described by a product-sum type or a sum-product type logical expression as a predetermined simple form, and when the search expression evaluation unit evaluates that it is within the resource upper limit at the time of the search using the logical expression, By increasing the combined height of the expressions and repeating the evaluation by the search expression evaluation unit at the time of searching with the increased logical expressions, an evaluation close to the resource upper limit is obtained, and the increased logical expressions are used as a search instruction list. The information retrieval apparatus according to claim 1, wherein the information retrieval apparatus outputs the information.   2. The information search apparatus according to claim 1, further comprising a search expression output unit that displays and outputs the search instruction list output by the search expression dividing unit, and displays and outputs the search instruction list. In order to retrieve the desired information from the database,
A search expression dividing unit that reads the search expression describing the desired information search and divides the read search expression into a predetermined simple format;
A search expression evaluation section that evaluates whether the necessary resources at the time of search by the simple search expression divided by the search expression dividing section are within the resource upper limit value of the system,
The search expression dividing unit divides the search expression evaluation unit into a simple form within evaluation that is evaluated as being within the resource upper limit value, and outputs the result as a search expression for database search. circuit. In an information retrieval method for retrieving desired information from a database,
A search expression dividing step of reading a search expression describing the desired information search and dividing the read search expression into a predetermined simple format;
A search expression evaluation step for evaluating whether the necessary resources at the time of search by the simple form search expression divided by the search expression dividing step are within the resource upper limit value of the system;
The search expression dividing step repeats the synthesis until it becomes a simple form within the maximum evaluation which is a simple form search expression of the longest length evaluated as being within the resource upper limit value in the search expression evaluation step,
An information search method comprising: a search processing step for performing the desired information search according to a search command list described in the simplified form within the maximum evaluation obtained in the search formula dividing step.   The search formula dividing step is expanded into a binary tree as a predetermined simple format, and if the search formula evaluation step evaluates that it is within the resource upper limit at the time of search by the binary tree, the height of the above binary tree And increase the binary tree description to obtain the evaluation close to the upper limit of the resource as a search instruction list by repeating the evaluation in the search expression evaluation step at the time of searching with the increased binary tree description. The information retrieval method according to claim 8, wherein:

Images