JPH10198697A - Structured document retrieving device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to retrieve documents at high reproducibility even in the case of retrieving plural documents having respectively different logical structures. SOLUTION: When a retrieval expression 1 is inputted, a retrieval expression generating means 2 rewrites retrieving conditions indicated in the expression to stepwisely moderate conditions and generates a condition eased retrieval expression 4. An accuracy calculating means 3 calculates accuracy indicating the accuracy of a retrieval result based on the expression 4 in accordance with the rewritten condition executed for the formation of each expression 4. A retrieval execution means 5 retrieves structured documents stored in a document storing means 6 based on the inputted retrieval expression 1 and the retrieval expression 4 generated by the means 2. A retrieved result merging means 7 merges retrieved results by arranging them in order from the result having the highest accuracy. Consequently a document of which logical structure is not correctly prepared can also be retrieved and the reproducibility of documents can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a structured document search apparatus for performing a document search for a structured document, and more particularly to a structured document search apparatus for searching a document generated from a plurality of document types.

[0002]

2. Description of the Related Art In a structured document, the content of the document is called a logical structure, and is represented by a tree structure including a plurality of document components such as chapters, sections, and figures. FIG. 19 is a diagram illustrating an example of a logical structure. Such a logical structure 61 may not be created freely, but is created according to a syntax rule called a document type.

FIG. 20 is a diagram showing an example of a document type. In this document type 60, a rectangular node is an element type (element type).
Is defined. The label of the node indicates the name of the element type. Entities of nodes having the same name have the same element type. Therefore, the element type named “section” in FIG. 20 is defined recursively.

A node indicated by an ellipse defines a connection between elements. This node is called a constructor. SEQ nodes are:
This indicates that the instances of the nodes connected to it are generated in that order. The REP node indicates that the instance of the node connected to the REP node is generated one or more times. The OPT node indicates that the instance of the node connected to it may or may not appear. CHO
The node indicates that an instance of any one of the nodes connected to the node is generated. Here, the “instance” indicates a document element generated based on the document type.

[0005] The definition of the document type in FIG. 20 is as follows. An "article" consists of one or more "sections"
"Section" is "heading" and zero or more "paragraphs" or "figure"
And zero or more “nodes”. As aforementioned,
"Sections" may be nested. Logical structure 61 in FIG.
Satisfies the restrictions of the document type 60 in FIG.

FIG. 20 shows an example of a simple document type.
Practical-scale document types are large, and it is not uncommon for the number of element types to be in the hundreds. A document type corresponds to a schema in a database. That is, the document type describes the meaning of the elements of the document and the relationship between the elements. Just as the processing of the database is performed according to the schema, the processing of the structured document is performed based on the document type information. For example, a layout instruction is defined according to a document type, and the document is allocated by inputting the document instance and the layout instruction. As another example, there is an example in which a necessary part is extracted as appropriate from an existing document group, and these are combined to create a new document. At this time, a new part may be inserted if necessary. In such a process, document type information is used for a search process for specifying a necessary part, a verification process for checking whether a newly constructed document is in a desired form, or the like.

To create a structured document based on a document type, a native editor that directly presents a logical structure to a user can be used. To use a native editor, you need to be familiar with the structured document itself and the document type used to create the document.

[0008] If the user is not familiar with structured documents and document types, a text editor or document creation software (WYSIWYW) which displays a screen almost the same as a print image is used.
(G editor), and a method of obtaining a desired logical structure using a converter is widely used. In this case, it is necessary to create a document using a text editor or a WYSIWYG editor in accordance with certain rules.

The creation rules are roughly divided into a method of creating a sentence so as to match a predetermined pattern for extracting a structure, and a specific style (information for designating a style) in a portion to be treated as an element. Two of the specifying methods are used.

In the case of using a pattern, for example, a blank line (continuous line feed) is inserted between elements, a specific element is expressed with indentation, or a predetermined pattern such as a regular expression is used. Methods such as numbering items so as to match or using a character string predetermined by a regular expression or the like are used.

On the other hand, when a style is used, a predetermined style is specified for a paragraph, sentence, phrase, or the like used as an element. By the way, in order to search a document, a full-text search method using a text of the document as a search condition is often used. In a full-text search, conditions regarding terms in a document are defined as AND (and), OR (or), N
Boolean search combined with OT (negation) is common. Simple Boolean searches tend to have good recall but low precision. That is, there is a problem that a search result often includes many documents that the user does not expect.

Therefore, in a document search for a structured document, a method of increasing search accuracy by using a logical structure is widely used. For example, in a chapter that has a character string of “document processing” in the chapter heading, by searching for a figure of “database” in the figure heading, a document database, a document generated based on the data in the database, etc. Diagrams can be searched for databases in the context of document processing.

Conventional techniques using these techniques include:
Japanese Patent Application Laid-Open No. 4-217073 (document search device in a document storage system) is known. In this document search device, a search target can be specified using the parent-child relationship of the nodes in the logical structure of the document, the contents and attributes of the nodes. As a result, if the logical structure is correctly created, the accuracy can be increased without lowering the recall by using the structure.

[0014]

However, in the conventional document retrieval apparatus for structured documents, if the logical structure is not created correctly, the accuracy does not decrease but the recall rate decreases. There was a problem that would.

That is, since the document is created by the user, the logical structure is not always created correctly. In particular, when a document is created using a WYSIWYG editor, a plurality of document elements having different logical structures, which are described so as not to match a predetermined pattern, or have the same or similar screen display. Is sometimes confused. As a result, the document type constraint is satisfied,
A document having a logical structure different from the logical structure to be originally created is created.

For the above reasons, in the conventional document retrieval apparatus for structured documents, the recall is reduced when a large number of documents are retrieved using information of the element type determined by the document type. This applies to all existing document search devices, including technologies other than the prior art exemplified above.

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and has as its object to provide a structured document retrieval apparatus which maintains a high recall even when retrieving a plurality of documents having different logical structures. .

[0018]

According to the present invention, in order to solve the above-mentioned problems, in a structured document search apparatus for searching a structured document, a type of a node in a document structure, a content of the node, and an attribute of the node. When a search expression described by a structural relationship between nodes is input, a search expression generating means for generating a condition relaxation search expression in which the search conditions indicated in the search expression are gradually rewritten into mild conditions And, according to the contents of the rewriting performed to generate the relaxed condition search formula, a certainty calculation means for calculating a certainty indicating the likelihood of the search result by the relaxed condition search formula, and the input search formula and A search execution unit for performing a search based on the relaxed condition search expression; and a search result merging unit for arranging and merging the search results obtained by the search execution unit in descending order of accuracy. Zoka document retrieval apparatus is provided.

According to such a structured document search device,
When a search expression is input, the search expression generation means rewrites the search conditions indicated in the search expression into gradually looser conditions, and generates a relaxed condition search expression. Then, according to the content of the rewriting performed to generate each of the relaxed condition search expressions, the accuracy indicating the likelihood of the search result obtained by the relaxed condition search expression is calculated by the accuracy calculation unit, and the search execution unit calculates A search is performed based on the input search formula and the relaxed condition search formula. The search results are arranged and merged by the search result merging means in ascending order of accuracy.

Further, in a structured document search apparatus for searching a structured document, a type of a node in the document structure,
When a search expression described by the content of the node, the attribute of the node, and the structural relationship between the nodes is input, each search condition in the search expression is set as an individual sub-expression, and the condition is satisfied for each sub-expression. Sub-expression evaluation means for extracting a document part extracted, and a probability calculation for calculating a certainty indicating a degree of certainty of each document part according to a sub-expression in which the document part extracted by the sub-expression evaluation means satisfies a condition. Means and a search result output means for outputting search results in the order of document parts having a high degree of certainty.

According to such a structured document search device,
When a search expression is input, each search condition in the search expression is set as an individual partial expression by the partial expression evaluation means, and a document part satisfying the condition is extracted for each partial expression. Then, the certainty calculating means calculates the certainty indicating the degree of certainty of each document part according to the sub-expression that satisfies the condition of the document part extracted by the sub-expression evaluating means. Then, the search result is output by the search result output unit in the order of the document parts with the highest accuracy.

[0022]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram illustrating the principle of the present invention. First, when a search expression 1 described by the type of a node in the document structure, the content of the node, the attribute of the node, and the structural relationship between the nodes is input, the search expression generation means 2
, Rewrites the search conditions indicated in the search formula gradually into mild conditions, and generates a relaxed condition search formula 4. For example, if "AND" of the structural condition is rewritten to "OR", the condition becomes a relaxed condition search formula. When such a condition relaxation search expression 4 is generated, the accuracy calculation means 3 determines whether the search result by the condition relaxation search expression 4 is correct according to the contents of the rewriting performed to generate each condition relaxation search expression 4. Calculate the likelihood of accuracy.

The search execution means 5 executes a search on the structured document in the document holding means 6 by using the input search expression 1 and the relaxed search expression 4 generated by the search expression generation means 2. . The search result merging unit 7 arranges and merges the search results obtained by the search executing unit 5 in descending order of accuracy.

As a result, a document whose logical structure has not been correctly created can be searched. If you simply look at the search results, the relevance rate will be lower than when searching with the original search formula, but the search results will be those that exactly match the original search formula and those that do not match much Are not arranged in a random order, and are output in the most probable order based on the original search formula, so that the user can examine the search results sequentially from the most probable ones.

FIG. 2 is a block diagram of the first embodiment of the structured document search device. This embodiment includes the following components. In this embodiment,
“Accuracy” used in the description of FIG. 1 is represented by “penalty”. The smaller the value of this penalty, the greater the certainty of the search result.

The search expression input means 11 provides the user with
Provides functions for creating and inputting search formulas. The search formula input means 11 can also input a search formula created by another program. Penalty assigning means 12
Provides a function for a user to assign a reference value for calculating a penalty for each rewrite rule of a search expression. The penalty calculation means 13 calculates the total penalty of each search formula according to the penalty value input by the penalty assignment means 12. The penalty designating means 14 provides the user with a function of inputting the upper limit of the penalty.

The search formula generating means 15 includes a search formula input means 1.
Based on the search formula input in 1, a search formula with mild search conditions is generated according to the search formula rewriting rule.
The search formula holding unit 16 holds the search formula input by the search formula input unit 11 and the search formula generated by the search formula generation unit 15 with the penalty value calculated by the penalty calculation unit 13 added thereto. .

The document holding means 17 stores a plurality of documents having different logical structures. The search execution unit 18 performs a search on the document held in the document holding unit 17 by using the search formula stored in the search formula holding unit 16.
The search result holding unit 19 stores a search result performed by the search expression execution unit 18. At this time, a penalty value set in the search formula is added to each search result. The search result merging unit 20 rearranges the search results stored in the search result holding unit 19 in ascending order of the value of the added penalty. The search result output means 21 displays the search results on the screen of the display device in the order in which the search results have been rearranged by the search result merging means 20.

In order to execute a search in the structured document search apparatus having such a configuration, the penalty assigning means 12 must be set in advance.
Is used to set a penalty for the rule for rewriting the search expression. The penalty to be set is set to a larger value as the degree of loosening the search condition by rewriting the search formula is increased. The input penalty is passed to the penalty calculating means 13 and is stored there. Further, the user sets the upper limit value of the penalty in the penalty specifying means 14.
Can also be entered using. The upper limit of the penalty is not an essential setting item.

After making the above settings, a user who wishes to search for a document inputs an arbitrary search formula to the search formula generation means 15 by the search formula input means 11. Then, the search formula generation unit 15 generates a large number of search formulas in which the search conditions are gradually reduced from the input search formula. The search formula generated by the search formula generating unit 15 is temporarily stored in the search formula holding unit 16. If the upper limit of the penalty is set, a search expression with the penalty exceeding the upper limit is not generated.

The search formulas held in the search formula holding means 16 are sequentially retrieved by the search formula executing means 18, and a search is performed on the structured document stored in the document holding means 17. Then, the search result obtained as a result of evaluating each search expression is held in the search result holding means 19 together with the penalty value added to the search expression.

After all the search expressions are evaluated, the search results held in the search result holding means 19 together with the penalty are sorted and merged by the search result merging means 20 in ascending order of the penalty value. The merged search results are sequentially output by the search result output unit 21.

Next, the contents of the first embodiment will be specifically described. First, an example of a penalty assigned to the rewriting rule will be described. FIG. 3 is a diagram illustrating an example of a penalty assigned to a rewrite rule. This penalty 13a
Are input by the penalty assigning means 12 and held by the penalty calculating means 13. In this example, when the rewrite type is “parent-child → grandchild”, the penalty is “1”.
0 ". If the rewrite type is “ignore type”, the penalty is “30”. If the rewrite type is “descendant AND → OR”, the penalty is “100”. When the rewrite type is "AND
In the case of “→ OR”, the penalty is “200”. If the rewrite type is “Boolean search”, the penalty is “500”.

With such a penalty set, the user inputs a search formula using the search formula input means 11. For example, the following search expression is input. FIG.
Is a diagram showing an example of a search formula. As shown in the figure, a search formula 30 input by the user using the search formula input unit 11 is shown.
Is internally represented by a directed graph.

Nodes 31 and 3 having labels other than AND
3 to 35 are conditions (structures) indicating a structural relationship between a condition (local condition) relating to an element of the document itself and an element specified by the local condition of a node having a label other than AND closest to the node. Condition). The local conditions shown in this example are a condition as to whether or not to be extracted (true or false), a condition as to the type of the node type, and a condition as to the text of the node. The structural condition is specified depending on whether the relationship with the upper node is parent or child or grandchild.

The node 32 whose label is AND indicates a conjunction of conditions relating to the structure. Note that the search expression input means 11
Is always "0".

Hereinafter, the nodes of the directed graph of FIG. 4 and the conditions specified in the nodes are referred to as sub-expressions of the original search formula. When the search formula 30 as shown in FIG. 4 is input, the search formula is rewritten prior to execution of the search process.

FIG. 5 is a flowchart of the rewriting process of the retrieval formula. Hereinafter, the rewriting process of the search formula will be described according to the procedure of FIG. [S1] The penalty calculation means 13 initializes a penalty value. That is, the value is set to “0”.

The following steps S2 to S9
Is an iterative process for the sub-expression of the internal expression of the search expression. [S2] The search expression generation unit 15 checks whether or not there is an applicable rewrite rule in the currently given search expression. If there is an applicable rewrite rule, the process proceeds to step S3; otherwise, the execution ends. [S3] The search formula generation unit 15 selects one applicable rewrite rule. [S4] The search formula generation unit 15 checks whether or not the rewrite rule selected in step S3 is a type ignoring. If the rewrite rule is "ignore type", the process proceeds to step S5, and if not, the process proceeds to step S6. [S5] The search formula generation means 15 rewrites the condition relating to the type of the node other than the extraction target (the extraction target is “false”) to “ANY” indicating an arbitrary type. That is, the condition regarding the type is invalidated. After this processing, go to step S7. [S6] The search formula generation means 15 rewrites the search formula by applying the rewrite rule selected in step S3. Rewrite rules are "parent and child → grandchild", "and descendant AND → OR",
Alternatively, if the condition is “AND → OR of text-related conditions”, the search expression is rewritten according to the rewriting rule. If the rewrite rule is “Boolean search”, the entire search expression is changed to Boolean search. The procedure for the change will be described later (shown in FIG. 10). [S7] The penalty calculation means 13 adds a penalty corresponding to the rewrite rule selected in step S3 to the search formula currently being processed. [S8] The search formula generation unit 15 checks whether the penalty has exceeded a reference value. If the penalty exceeds the reference value, the execution ends, otherwise the process goes to step S9. [S9] The search formula holding means 16 stores the search formula obtained by rewriting the search formula in step S5 or step S6, and goes to step S2.

By performing the above processing,
A plurality of new search formulas are generated from the search formula shown in FIG. Then, a penalty is added to each search formula according to the content of the rewrite. FIGS. 6 to 9 show examples in which the search formula 30 in FIG. 4 is rewritten.

FIG. 6 is a diagram showing a first example of a rewritten search formula. This search expression 30a is the search expression 30 in FIG.
In contrast, a rewrite rule for changing a parent-child relationship to a grandchild relationship is applied to three places. The penalty in this case is 10 × 3 = 30.

FIG. 7 is a diagram showing a second example of the rewritten search formula. This search expression 30b is the search expression 30 in FIG.
The rewriting rule that ignores the type is applied to three places with respect to a. The penalty in this case is (penalty of search formula 30a in FIG. 6) + 30 × 3 = 30 + 90 = 12
0.

FIG. 8 is a diagram showing a third example of the search formula after rewriting. The search expression 30c is the search expression 30 in FIG.
For b, a rewrite rule for changing the AND condition for descendants to OR is applied to one place. The penalty in this case is (the penalty of the retrieval formula 30b in FIG. 7).
+ 100 = 120 + 100 = 220.

FIG. 9 is a diagram showing a fourth example of the rewritten search formula. This search formula 30d is the same as the search formula 30 in FIG.
For c, a rewrite rule for changing the AND of the text-related condition to OR is applied to one place. The penalty in this case is (penalty of search formula 30c in FIG. 8) + 200 = 220 + 200 = 420.

FIGS. 6 to 8 show examples in which a tree-structured search formula is rewritten to a tree-structured search formula. In these rewrites, search conditions in each node are rewritten according to a rewrite rule. I just need. On the other hand, rewriting to Boolean search requires the following procedure.

FIG. 10 is a flowchart of a procedure for rewriting the entire search expression into a boolean search. This procedure is a process performed by the retrieval formula generation means 15. The input of this procedure is a node of the retrieval formula represented by a tree structure, and the output is obtained by converting a subtree of the retrieval formula having the node as a vertex. Is a boolean search expression. In this procedure, the output retrieval formula is represented by a character string. [S11] It is determined whether or not the node is a condition for the element.
If the condition is related to an element, the process proceeds to step S12; otherwise, the process proceeds to step S13. [S12] The condition relating to the text content specified in the node is set as a search expression, and the process proceeds to step S14. [S13] The search expression is made empty, and the process proceeds to step S14. [S14] It is determined whether the node has a child node.
If a child node exists, the process proceeds to step S15; otherwise, the process proceeds to step S23.

The following steps S15 to S2
2 is a repetition process for a child node of the node. [S15] It is determined whether there is any child node that has not been subjected to the conversion process. If there is an unprocessed child node, step S16
Otherwise, go to step S23. [S16] One unprocessed child node is selected. [S17] Using the node selected in step S16 as an argument, a rewrite procedure for Boolean search is recursively called. [S18] It is determined whether the search expression is empty. If the search expression is empty, go to step S19; otherwise, go to step S20. [S19] The search formula (character string) obtained as a result of the procedure call in step S17 is used as a search formula, and the process proceeds to step S15. [S20] It is determined whether the node given as an argument of this procedure is an OR node. If it is an OR node, go to step S22, otherwise go to step S21. [S21] The search expression and the search expression obtained as a result of the procedure call in step S17 are connected by AND, and the process proceeds to step S1.
Go to 5. [S22] The search formula and the search formula obtained as a result of the procedure call in step S17 are connected by OR, and the process proceeds to step S15. [S23] The execution of the procedure is terminated using the search expression as a return value.

By such processing, a Boolean search formula can be obtained from a tree-structured search formula. An example in which the search expression of FIG. 4 is directly rewritten to a Boolean search expression will be described below.

[0049]

[Equation 1] (SGML AND conference) AND "Yuri Rubinski"
AND Publishing The penalty for this boolean search formula is (penalty for search formula 30 in FIG. 4) + 500 = 0 + 500 = 500.

When a search is performed based on each of a plurality of tree-structured search expressions and a Boolean search expression as described above, a penalty of the search expression is added to each search result. Then, all search results are arranged in ascending penalty order. At this time, if one document is detected by a plurality of search formulas, a smaller value penalty is adopted as the search result. The arranged search results are displayed on the screen of the display device.

FIG. 11 is a diagram showing a display example of a search result according to the first embodiment. In the drawing, a document obtained as a search result is displayed in a search result display screen 21a. These documents are listed with the penalty, document name, author, and date created. A scroll bar is provided on the screen. By operating the scroll bar, a document to be displayed can be scrolled, and search results with small penalties can be sequentially displayed on the screen. If the upper limit of the penalty is set, there is no search result exceeding the upper limit.

Although the strictness of the search condition is indicated by the penalty in FIG. 5, a score may be calculated so that the stricter the search condition, the higher the score.
For example, when the penalty upper limit value is specified by the penalty specifying means 14, a value obtained by subtracting the penalty value from the upper limit value can be used as the score.

In this way, it is possible to browse the search results based on the search formula in which the search conditions are gradually relaxed from the input search formula. Therefore, even when a search formula created based on a correct logical structure is input, a document whose logical structure has not been created correctly can be detected, and the recall is improved. In addition, since a search expression having a low degree of loosening condition is preferentially displayed, even if the number of search results increases, browsing of the user does not become inconvenient.

In the first embodiment, the search conditions are gradually reduced by rewriting the search conditions of the input search formula. Document parts (individual elements of the structured document) are extracted based on the formula, and a document part having a high degree of satisfying the condition of the search formula can be output as a search result. Such an example will be described below as a second embodiment.

FIG. 12 is a block diagram of a second embodiment of the structured document retrieval apparatus. In this embodiment, the likelihood of the search result is represented by “score”.
This score indicates that the greater the value, the greater the likelihood of the search result. The search formula input means 41 provides a user with a function of creating and inputting a search formula. The search formula holding unit 42 holds the search formula input by the search formula input unit 41.

The document holding means 43 stores a plurality of documents having different logical structures. The sub-expression evaluation means 44
The sub-expressions of the retrieval formula stored in the retrieval formula holding unit 42 are sequentially extracted, and the documents stored in the document retaining unit 43 are evaluated based on the partial formula. That is, the document part satisfying the condition indicated by the extracted sub-expression is stored in the document holding unit 43.
From the list as candidates. On the other hand, score assigning means 45
Provides a user with a function for assigning a reference value for calculating a score for each type of sub-expression.
The score calculation means 46 obtains the scores of the candidates extracted by the sub-expression evaluation means 44 according to the value of the score input by the score allocation means 45.

The candidate holding means 47 includes a sub-expression evaluation means 44
Adds the value of the score to the obtained candidate and holds it. The score specifying means 48 provides the user with a function of inputting the lower limit of the score. The search result output unit 49 displays, on the screen of the display device, document parts having a score greater than the lower limit specified by the score designating unit 48 as search results, in descending order of the score.

It should be noted that the sub-expression evaluation means 44, the score calculation means 46, and the candidate holding means 47 in the figure constitute a retrieval expression evaluation means 40. To execute a search in such a structured document search device, a user assigns a value that is a reference for calculating a score for each type of sub-expression using the score assigning unit 45 in advance. The score to be assigned is set to a larger value as the degree of narrowing down the search target by the search condition is larger. The input score is passed to the score calculation means 46, where it is stored. Also,
The user can also input the lower limit value of the score using the score specifying means 48. Note that the lower limit of the score is not an essential setting item.

After making the above settings, the user who wishes to search for a document can use the search formula input means 41 to input the search formula created by himself or a predetermined program into the search formula holding means 42. To enter. The search formula input by the search formula input unit 41 is held by the search formula holding unit 42. The search expression held in the search expression holding means 42 is
The sub-expression evaluation unit 44 evaluates the document held in the document holding unit 43 for each sub-expression. The candidates obtained as a result of evaluating the sub-expressions of the search expression by the sub-expression evaluation unit 44 are stored in the candidate storage unit 47 together with the score calculated by the score calculation unit 46 for each sub-expression. When a candidate document part is obtained at the stage of evaluating the sub-expression, if the part does not yet exist in the candidate holding unit 47, the score calculation unit 46 determines the score of the sub-expression as the candidate score. Assign a score. If the obtained candidate already exists in the candidate holding unit 47, the score calculating unit 46 increases the score of the candidate by the score of the sub-expression.

After all the sub-expressions have been evaluated, the search results held in the candidate holding means 47 together with the scores are output by the search result output means 49 in the order of the scores. Next, the contents of the second embodiment will be specifically described. First, an example of a score assigned to a condition for evaluating a document will be described.

FIG. 13 is a diagram showing an example of scores assigned to conditions for evaluating a document. This score 46a
Is input by the score allocating unit 45 and the score calculating unit 4
6 holds. In the example of the figure, when the condition is “type”, the score is “100”. When the condition is “parent-child relationship”, the score is “100”. If the condition is “grandchild relationship”, the score is “100”. If the condition is “condition for text”, the score is “1”.
00 ”. If the condition is “type AND”, the score is “200”. If the condition is “AND of condition related to text”, the score is “200”.

With such a score set, the user inputs a search formula using the search formula input means 41.
For example, a search formula as shown in FIG. 4 is input. Then, the search formula evaluation unit 40 evaluates the search formula for the document in the document holding unit 43, and determines a document with a high evaluation as a search result.

FIG. 14 is a flowchart of a procedure for evaluating a retrieval expression. In this procedure, scores are given to search result candidates at the stage of evaluating search conditions. [S31] The sub-expression evaluation unit 44 checks whether there is any unprocessed node among the nodes of the internal expression of the search expression. If there is an unprocessed node, the procedure goes to step S32. Otherwise, go to step S37. [S32] The sub-expression evaluation means 44 selects one unprocessed node from the nodes of the internal expression of the search expression. [S33] The sub-expression evaluation unit 44 checks whether there is any unprocessed search condition among the search conditions of the node selected in step S32. If there is an unprocessed condition, the process proceeds to step S34; otherwise, the process proceeds to step S31. [S34] The sub-expression evaluation means 44 selects one unprocessed search condition among the search conditions of the node selected in step S32. [S35] The sub-expression evaluation means 44 evaluates the search condition selected in step S34. [S36] The score calculation means 46 adds a score corresponding to the search condition evaluated in step S35 to each candidate obtained as a result of evaluating the search condition in step S35. After that, it goes to step S33. [S37] The sub-expression evaluation unit 44 checks whether or not there is an unevaluated structural condition. If there is an unevaluated structural condition, the process proceeds to step S38; otherwise, the process proceeds to step S41. [S38] The sub-expression evaluation unit 44 selects one of the structural conditions that has not been evaluated. [S39] The sub-expression evaluation means 44 evaluates the structural condition selected in step S38. [S40] The score calculation means 46 adds a score corresponding to the condition evaluated in step S39 to each candidate obtained in step S39. [S41] The candidate holding unit 47 excludes non-extraction targets from the candidates and holds the remaining candidates. That is, only the candidate whose extraction target is “true” is held.

By performing the above processing,
Search results are obtained from the search formula shown in FIG. And
A score is added to each search expression. Since only the node 31 of type "Section" is to be extracted in the search formula of FIG. 4, the retrieved result is only the document part of type "Section". FIGS. 15 to 17 show search results obtained by the search formula 30 in FIG. 4 together with scores.

FIG. 15 is a diagram showing a first example of a search result. In the document part 50, the relations and contents of the respective nodes 51 to 54 satisfy all the conditions of the search formula 30 in FIG. Therefore, the score has the maximum value (= 1500).

FIG. 16 is a diagram showing a second example of the search result. This document part 50a includes three nodes 51 and 5
Only the conditions regarding 2,54 are satisfied. for that reason,
The score is lower than the example of FIG.
It is.

FIG. 17 is a diagram showing a third example of the search result. This document part 50b is composed of two nodes 51 and 52.
Only the conditions regarding are satisfied. Therefore, the score is lower than the example of FIG. 16, and the score is 700.

The retrieval results thus retrieved are displayed on the screen of the display device in the order of the scores. FIG.
It is a figure showing the example of a display of the search result by an embodiment.
In the figure, a document obtained as a search result is displayed in a search result display screen 49a. These documents are displayed with their scores, document names, authors, and creation dates. By operating a scroll bar on this screen, a document to be displayed on the screen can be scrolled, and search results having a small score can be sequentially displayed on the screen.
At this time, if the upper limit of the score is determined, the search results exceeding the upper limit are not displayed.

It is to be noted that, in addition to the above two embodiments, the following modified examples are also conceivable. In the above embodiment,
Although the score and penalty are fixed values, they may be variable depending on the state of the database. An example of a method for changing these values according to the state of the database will be described below.

For example, in the case of a condition relating to a text, the score or penalty is lowered for words and phrases that appear more frequently in the database. Further, in the case of the condition relating to the type, the score or the penalty is reduced as the words appear more frequently in the database.

As described above, in order to calculate a penalty or a score according to the frequency of appearance of a word, the penalty calculating means 13 (shown in FIG. 2) or the score calculating means 46
(Shown in FIG. 12), when a condition relating to text is specified, it is sufficient to have means for determining the degree of frequent occurrence of a phrase in the condition relating to text, and means for determining a penalty or score according to the degree. .

When the parent-child relationship or the grandchild relationship is inevitably established, the score when the condition regarding the parent-child relationship or the grandchild relationship is matched is set to 0. In this case,
Since a parent-child relationship or a grandchild relationship always holds, no matter what the value of the penalty, it has no practical meaning. Whether the parent-child relationship or the grandchild relationship is inevitably established can be determined by looking at the document type. A certain type T1
Therefore, when another type T2 can be reached by tracing only the SEQ or REP node, if T1 exists, T2 always exists (and vice versa), and they are in a parent-child relationship. When it is possible to reach another type T2 from one type T1 by tracing only the SEQ, REP, or type node, T1
, T2 always exists (and vice versa), and they are in a grandchild relationship.

[0073]

As described above, in the first aspect of the present invention, a condition relaxation search formula in which search conditions are gradually relaxed is generated based on an input search formula, and a search with moderate conditions is also performed. As a result, a document for which a logical structure has not been created correctly can be searched, and the recall is improved. In addition, since the search results are output in the most probable order in light of the original search formula, the user can examine the search results sequentially from the most probable ones.

Further, in the second invention, the document parts are extracted by the sub-expression of the input search expression, and are output as the search results in descending order of the degree satisfying the conditions of the input search expression. In the same manner as in the first aspect, a document whose logical structure has not been correctly created can be searched, and the recall is improved.

[Brief description of the drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a block diagram of a first embodiment of a structured document search device.

FIG. 3 is a diagram showing an example of a penalty assigned to a rewrite rule.

FIG. 4 is a diagram showing an example of a search formula.

FIG. 5 is a flowchart of a search formula rewriting process.

FIG. 6 is a diagram illustrating a first example of a search expression after rewriting.

FIG. 7 is a diagram illustrating a second example of a rewritten search formula.

FIG. 8 is a diagram illustrating a third example of a rewritten search formula.

FIG. 9 is a diagram illustrating a fourth example of a rewritten search expression.

FIG. 10 is a flowchart of a procedure for rewriting the entire search expression to a Boolean search.

FIG. 11 is a diagram illustrating a display example of a search result according to the first embodiment.

FIG. 12 is a block diagram of a structured document search device according to a second embodiment.

FIG. 13 is a diagram illustrating an example of a score assigned to a condition for evaluating a document.

FIG. 14 is a flowchart of a procedure for evaluating a search expression.

FIG. 15 is a diagram showing a first example of a search result.

FIG. 16 is a diagram showing a second example of a search result.

FIG. 17 is a diagram showing a third example of a search result.

FIG. 18 is a diagram illustrating a display example of a search result according to the second embodiment.

FIG. 19 is a diagram illustrating an example of a logical structure.

FIG. 20 is a diagram illustrating an example of a document type.

[Explanation of symbols]

 1 search formula 2 search formula generation means 3 accuracy calculation means 4 relaxed condition search formula 5 search execution means 6 document holding means 7 search result merging means

Claims (6)

[Claims] 1. A structured document search apparatus for a structured document as a search target, wherein a search expression described by a node type, a node content, a node attribute, and a structural relationship between nodes in a document structure is provided. When input, a search expression generating means for generating a condition relaxation search expression in which the search conditions indicated in the search expression are gradually rewritten into mild conditions, and rewriting performed to generate the condition relaxation search expression According to the content of, a certainty calculation means for calculating the certainty indicating the likelihood of the search result by the relaxed search formula, and a search execution means for performing a search based on the input search formula and the relaxed search formula And a search result merging means for arranging and merging the search results obtained by the search executing means in descending order of accuracy. 2. The structured document search device according to claim 1, wherein said search formula generation means generates only a condition relaxation search formula having a higher accuracy than a predetermined limit value. 3. A reliability assigning means for assigning a reference accuracy for each rewriting rule, wherein the accuracy calculating means calculates the accuracy of the condition relaxation search formula based on the reference accuracy assigned by the accuracy assigning means. 2. The document search device according to claim 1, wherein: 4. A structured document search apparatus for a structured document as a search target, wherein a search expression described by a node type, a node content, a node attribute, and a structural relationship between nodes in a document structure is used. When input, each search condition in the search expression is set as an individual sub-expression, and a sub-expression evaluation unit that retrieves a document part satisfying each condition for each sub-expression; Based on what sub-expression conditions the document part satisfies, the accuracy calculation means that calculates the degree of certainty of each document part, and the search results are output in the order of the document parts with the highest accuracy And a search result output unit. 5. The structured document search device according to claim 4, wherein said search result output means sets only a document part having a higher accuracy than a predetermined limit value as a search result. 6. A reliability assigning means for assigning a reference accuracy for each type of sub-expression, wherein the accuracy calculating means calculates the accuracy of each document part based on the reference accuracy assigned by the accuracy assigning means. 5. The structured document search device according to claim 4, wherein: