JP4490930B2 - Structured document search apparatus and structured document search method - Google Patents

Description

  The present invention relates to a structured document search apparatus and a structured document search method for searching a plurality of structured documents having different document structures managed by a structured document database having a hierarchical logical structure.

  In recent years, structured document databases for storing and retrieving structured document data described in XML (eXtensible Markup Language) or the like have been developed. The query (search) for the structured document database is generally performed by a query language called XQuery (XML Query) which is being standardized by the World Wide Web Consortium (W3C).

  XQuery not only performs filtering by specifying a conditional expression, but also embeds a virtual structured document in the return section that returns the search result, makes the conditional expression a complex nested structure, and adds a functional expression to the conditional expression. It is a language that can be defined, etc., and has a very high query ability with various functions.

  In XQuery, node-level information in DOM (Document Object Model) such as elements and attributes is a search target. For example, Patent Document 1 proposes a technique for searching for node level information of a structured document by the following method.

  First, when a structured document is stored in a database, the data structure of the target document is analyzed, and an index is created by embedding analysis information for the structure (node) in lexical index information or the like. The structure analysis information in this case is information regarding path levels that can be expressed in XPath (XML Path Language) as the same structure information (structure template).

  Next, at the time of search, the search query is analyzed to create a query graph, and after calculating the cost, a query execution plan is created. At this time, by analyzing the query at the time of index processing and obtaining constraints on the structure that each variable must satisfy in advance, the search of the index is limited and the number of intermediate candidates is reduced. Is realized.

  On the other hand, in the field of full-text search, a function for searching for a matching document by specifying a logical condition (such as AND / OR) of a keyword or a function for searching by specifying a character appearance position as a condition is realized. An example of a search condition that uses an appearance position as a condition is a neighborhood search such as “Search for information in which“ web ”and“ title ”exist within 5 characters in the <title> tag”.

  In order to realize the neighborhood search, it is necessary to leave the text position information when processing the AND condition. However, since the XQuery is a node-based model, information on the appearance position is lost when the AND condition is processed, and the neighborhood search cannot be realized.

  Thus, XQuery cannot be said to have sufficient expression capability from the viewpoint of realizing the full-text search function. In order to solve this problem, the W3C has proposed a language standard called XQFT (XQuery-Full-Text) that extends XQuery and combines full-text search functions.

  XQFT can describe a plurality of conditions including a condition of the appearance position of a text for a node to be searched. For example, the example of the neighborhood search described above can be described as “./title ftcontains“ web ”&&“ site ”distance at most 5”. In XQFT, the information to be searched is not at the “Node” level of XQuery but at the “Text Appearance Position” level, so a processing method that takes this into consideration must be realized.

JP 2001-147933 A

  However, in XQFT, search result candidates that take the text appearance position level into consideration are acquired, so the number of intermediate candidates becomes enormous when performing join processing and the like, resulting in a decrease in search processing speed and an explosion in memory capacity. There was a problem that it might invite.

  For example, when “web” appears 100 times and “site” appears 200 times in a certain element, if candidates for each element are simply combined, there are 20000 (100 × 200) intermediate candidates. When the text appearance position level is considered by XQFT, the number of candidates for each element increases, so the number of intermediate candidates obtained by combining these candidates further increases.

  The present invention has been made in view of the above, and an object thereof is to provide a structured document search apparatus and a structured document search method capable of realizing a high-speed full-text search function for a structured document.

  In order to solve the above-described problems and achieve the object, the present invention includes a document having a hierarchical logical structure, which is an element that is actual information corresponding to a structural element that is a unit of the logical structure. Structured document storage means for storing a structured document in association with a document ID for uniquely identifying the structured document, a structure ID for uniquely identifying the structural element, and a feature ID of the element A structure information storage means for storing structure information associated with one feature information; and input of search conditions for the structured document; analysis of the received search conditions; and correspondence to the structure elements of the structured document A hierarchical search condition having a hierarchical structure of nodes that are units of the structured structure, the hierarchy including the node associating the structure ID candidate to be searched with a search key for the structure ID candidate Construction Analyzing means for obtaining a search condition, and for each node of the search condition of the hierarchical structure obtained by the analyzing means, the first feature information associated with the candidate structure ID included in each node is the structure information. Obtaining means for obtaining from the storage means; for each node of the search condition of the hierarchical structure obtained by the analyzing means; calculating means for calculating second feature information representing a feature of the search key included in each node; For each node of the search condition of the hierarchical structure obtained by the analysis means, the second feature information calculated by the calculation means for the corresponding search key among the candidate structure IDs included in each node is: A deletion unit that deletes the structure ID candidate that does not match the first feature information acquired by the acquisition unit; and the floor from which the deletion unit has deleted the structure ID candidate. Based on a structure search condition, the document ID corresponding to the structure ID satisfying the hierarchical structure search condition is obtained, and the structured document corresponding to the obtained document ID is retrieved from the structured document storage means. And a search means.

  Further, the present invention is a document having a hierarchical logical structure, wherein the structured document including an element that is actual information corresponding to a structural element that is a unit of the logical structure, and the structured document are uniquely identified. Structured document storage means for associating and storing a document ID to be identified, structure information for associating structure ID for uniquely identifying the structure element, and first feature information representing the feature of the element Structure information storage means for storing; index storage means for storing an index that associates the document ID; an element ID that uniquely identifies the element; and the structure ID of the structure element corresponding to the element; A search condition of a hierarchical structure that accepts input of search conditions for the structured document, analyzes the accepted search conditions, and has hierarchically structured nodes corresponding to the structural elements of the structured document. , Search target Each of the hierarchical structure search condition obtained by the analysis means, the analysis means for obtaining the hierarchical structure search condition including the node that associates the structure ID candidate and the search key for the structure ID candidate. For the node, an acquisition means for acquiring the first feature information associated with the structure ID candidate included in each node from the structure information storage means, and a search condition of the hierarchical structure obtained by the analysis means For each node, the calculation means for calculating the second feature information representing the characteristics of the search key included in each node, and each node of the hierarchical structure search condition obtained by the analysis means are included in each node. Among the candidate structure IDs, the second feature information calculated by the calculation unit for the corresponding search key is the first feature information acquired by the acquisition unit. A deletion unit that deletes the structure ID candidate that does not conform to the structure ID, and the deletion unit corresponds to the structure ID that satisfies the search condition of the hierarchical structure based on the search condition of the hierarchical structure from which the candidate of the structure ID has been deleted. Index search means for searching the document ID and element ID from the index storage means, and search means for searching the structured document corresponding to the document ID searched by the index search means from the structured document storage means And.

  The present invention is also a structured document search method capable of executing the above apparatus.

  According to the present invention, it is possible to store the feature information representing the feature of the corresponding element for each structural element of the structured document, and to narrow down the number of intermediate candidates by referring to the feature information during the search. Therefore, there is an effect that a high-speed full-text search function for the structured document can be realized.

  Exemplary embodiments of a structured document search apparatus and a structured document search method according to the present invention will be explained below in detail with reference to the accompanying drawings.

(First embodiment)
The structured document search apparatus according to the first embodiment stores feature information indicating the feature of an element corresponding to each structure element of the structured document, and narrows down the number of intermediate candidates by referring to the feature information at the time of search. Thus, a full-text search function for structured documents such as XQFT can be realized at high speed.

  FIG. 1 is a block diagram showing a configuration of a structured document search apparatus 100 according to the first embodiment. As shown in the figure, the structured document search apparatus 100 is connected to a client 300 via a network 200, and includes a communication unit 101, a storage processing unit 110, a search processing unit 120, and a structured document storage unit. 131, a structure template storage unit 132, and an index storage unit 133.

  The client 300 transmits a structured document to be registered (XML document) or a search query for a registered structured document to the structured document search apparatus 100 and receives a search result.

  The network 200 connects the client 300 and the structured document search apparatus 100. For example, any network configuration such as the Internet, a wired LAN (Local Area Network), and a wireless LAN can be applied.

  The communication unit 101 receives a structured document and a search query from the client 300 via the network 200, and transmits a search result to the client 300.

  The structured document storage unit 131 is a storage unit that stores a structured document described in XML. Here, the description format of the structured document will be described. FIG. 2 is an explanatory diagram showing an example of a structured document described in XML.

  In the figure, an example of a structured document in which information related to a patent is described in an XML format is shown. In XML, tags are used to represent the structure of a document. A tag has a start tag and an end tag. By enclosing the components of a structured document with a start tag and an end tag, character strings (text) in the document can be separated and the structure of the text can be any. Can clearly describe whether it belongs to an element.

  In XML, data units defined using tags are called elements. For example, data including a <patent> tag and a </ patent> tag, and data surrounded by both tags constitute one element.

  Further, an attribute for adding additional information such as whether the element can be omitted or can be repeated can be designated for the element. The attribute is set in a format such as “<element name attribute =“ attribute value ”>” in the start tag. In the figure, an example in which the “ID” attribute is designated as the attribute of the “patent” element is shown.

  The start tag is described in a format in which element names are closed with symbols “<” and “>”, and the end tag is described in a format in which element names are closed with symbols “</” and “>”. Between the start tag and the end tag, text representing actual information of the structured document or other elements (child elements) is set. A component that does not include text such as “<patent DB> </ patent DB>” can also be represented as “<patent DB />” as a simple notation.

  The document shown in the figure has an element starting from a “patent” tag as a document root (root), and its child elements include elements starting from “title”, “inventor list”, “effect”, and “keyword list” tags. . In addition, for example, one element (character string) such as “structured document search device” exists in an element starting from a “title” tag.

  Note that information obtained by extracting the name, hierarchical relationship, number of repetitions, and the like of each tag from such an XML structured document is referred to as structure information. A logical structural unit constituting the structural information of the structured document is called a structural element. In the first embodiment, the above-described elements, attributes, and text are structural elements.

  Next, the data structure of the structured document stored in the structured document storage unit 131 will be described. FIG. 3 is an explanatory diagram showing an example of the data structure of the structured document stored in the structured document storage unit 131.

  This figure shows an example in which the structured document shown in FIG. 2 is represented by a tree-structured data structure. In FIG. 3, an ellipse node is a folder node, a hexagon node is a document node, a single-line rectangle is a tag node, a double-line rectangle is an attribute node, and a rounded rectangle is text Means a node representing

  For example, the subtree below the node representing the “inventor list” tag indicates that it includes two “inventor” elements below the “inventor list” element.

  The structured document storage unit 131 stores such a tree-structured data structure in a table format. FIG. 4 is an explanatory diagram showing an example of the data structure of the structured document represented in a table format.

  As shown in the figure, the structured document in the structured document storage unit 131 corresponds to the document ID, the element ID, and the next brother and the eldest son when the parent-child sibling relationship of the tree structure is expressed in the eldest son-second-sister method. Are stored. The document ID is an identifier for uniquely identifying the structured document. The element ID is an identifier for uniquely identifying a tree-structured element (node) of the structured document identified by the document ID.

  For example, the node with the element ID = 3 in the figure corresponds to the node of the “title” tag in FIG. 2, the node of the “inventor list” tag in FIG. 2 (element ID = 4) as the second brother, and The eldest son has the “text” node (element ID = 5) in FIG.

  The structure template storage unit 132 stores structure information extracted from the structured document in the XML format as described above. The structure template storage unit 132 is referred to when the structure of the structured document stored in the structured document storage unit 131 is analyzed with reference to the structure information.

  FIG. 5 is an explanatory diagram illustrating an example of the data structure of the structure information stored in the structure template storage unit 132 according to the first embodiment. This figure shows an example in which the structure information is represented by a tree structure.

  As shown in the figure, the tree structure of structure information is similar to the tree structure of structured documents. Ellipse nodes representing folders, hexagonal nodes representing documents, single-line square nodes representing tags, and attributes A double-lined square node representing the text and a rounded-cornered square node representing the text are included.

  Further, a template ID (TID) that is an identifier for uniquely identifying a structural element that is each node of the structural information is given to the structural information. The structure information is information obtained by extracting only information representing a structure from a plurality of structured documents. Therefore, for example, even information that can be set in a structured document, such as a node of an “inventor” tag, is collected into one on the structure information.

  The structure template storage unit 132 stores such tree structure information in a table format. FIG. 6 is an explanatory diagram showing an example of the data structure of the structure template storage unit 132 expressed in a table format.

  As shown in the figure, the structure information in the structure template storage unit 132 stores TID, second younger brother, eldest son, minimum vocabulary ID, maximum vocabulary ID, and maximum text size in association with each other.

  In the minimum vocabulary ID and the maximum vocabulary ID, the vocabulary having the minimum and maximum vocabulary ID among the vocabulary appearing in the text element corresponding to the structural element identified by the TID is set. The vocabulary ID is an identifier for uniquely identifying the vocabulary stored in the index storage unit 133 and assigned to the vocabulary in the order of appearance. Details of the index storage unit 133 will be described later.

  The maximum text size refers to the value of the character string length of the text element having the maximum character string length among the text elements corresponding to the structural element identified by the TID.

  As described above, not only the information indicating the parent-child relationship of the structural element (second brother and eldest son) but also the information indicating the characteristic of the text element corresponding to the structural element (characteristic information), the minimum vocabulary ID, the maximum vocabulary ID and the maximum text By storing the size in association with each other, it is possible to narrow down the number of intermediate candidates by referring to the information at the time of search.

  The index storage unit 133 stores an index used for searching the structured document, and includes a statistical information storage unit 133a and a transposed file storage unit 133b.

  The statistical information storage unit 133 a is a storage unit that stores statistical information of vocabulary generated in each structured document stored in the structured document storage unit 131. A vocabulary is a phrase included in text representing actual information of a structured document. In the first embodiment, a vocabulary is a phrase obtained by dividing a text into N grams. N-gram division refers to a method of dividing all N consecutive characters included in a character string as a vocabulary.

  For example, when the character string “XML database” is divided into N grams with N = 3, “XML”, “ML data”, “L data”, “data”, “table”, “tab data”, “base”, “−” It is divided into nine vocabularies, “su” and “su”.

  Note that the method of dividing vocabulary from text is not limited to this, and any method can be applied, such as a method that uses words obtained by morphological analysis as vocabulary as long as it can extract words contained in the text. Can do.

  FIG. 7 is an explanatory diagram illustrating an example of a data structure of statistical information stored in the statistical information storage unit 133a. As shown in the figure, the statistical information storage unit 133a includes, for each divided vocabulary, a vocabulary ID that is an identifier assigned in ascending order in the order of occurrence, the occurrence frequency of the vocabulary in all structured documents, and a transposed file. Stores statistical information that associates numbers.

  In general, a vocabulary with a high appearance frequency is likely to exist in any document, so the value of the vocabulary ID tends to be small. Conversely, for a rare vocabulary, the value of the vocabulary ID is large. Tend.

  The inverted file number is a number for uniquely identifying an inverted file stored in an inverted file storage unit 133b described later. An index for accelerating the retrieval of the structured document is constituted by the lexical statistical information stored in the statistical information storage unit 133a and the transposed file.

  The transposed file storage unit 133b is a storage unit that stores a transposed file that constitutes a part of an index for speeding up the search processing of structured documents. FIG. 8 is an explanatory diagram showing an example of the data structure of the transposed file stored in the transposed file storage unit 133b.

  This figure shows the data structure of one transposed file corresponding to a certain vocabulary stored in the statistical information storage unit 133a. Actually, a transposed file having a structure as shown in the figure is created for each vocabulary and stored in the transposed file storage unit 133b.

  As shown in the figure, the transposed file in the transposed file storage unit 133b stores lexical index information in which TID, document ID, element ID, and occurrence position are associated with each other.

  The occurrence position is information indicating a position where the vocabulary corresponding to the transposed file appears in the element of the structured document identified by the document ID and the element ID.

  The storage processing unit 110 executes processing for storing the structured document received from the client 300 in the structured document storage unit 131. The structure information extraction unit 111, the structure template determination unit 112, the index registration unit 113, , A statistical information update unit 114 and a document registration unit 115 are provided.

  The structure information extraction unit 111 extracts structure information from the structured document received by the communication unit 101 from the client 300. Specifically, the structure information extraction unit 111 parses the received structured document and develops it into a tree structure format such as DOM, and extracts each node of the tree structure as structure information. Each node of the expanded tree structure is stored in the structured document storage unit 131 by the document registration unit 115 described later. The document ID and element ID are given when the structure information extraction unit 111 extracts structure information.

  The structure template determination unit 112 refers to the structure information extracted by the structure information extraction unit 111 and collates with the structure information stored in the structure template storage unit 132, thereby providing each node of the expanded structured document with a tree structure. The corresponding TID is determined. When the structure information extraction unit 111 extracts new structure information that is not included in the structure information already stored in the structure template storage unit 132, the structure template determination unit 112 stores the new structure information in the structure template storage. Stored in the unit 132.

  The index registration unit 113 creates lexical index information for text elements included in the structured document received from the client 300 and registers it in the transposed file storage unit 133b.

  Specifically, the index registration unit 113 includes a TID, a document ID, a TID determined or calculated by the structure template determination unit 112 in a transposed file corresponding to a vocabulary obtained by dividing a character string of a text element by N grams. Vocabulary index information in which the element ID and the occurrence position are associated is newly added to the transposed file.

  The statistical information update unit 114 registers the statistical information of the vocabulary extracted from the text elements included in the structured document received from the client 300 in the statistical information storage unit 133a.

  Specifically, the statistical information update unit 114 updates the statistical information in the statistical information storage unit 133a corresponding to the vocabulary obtained by dividing the character string of the text element into N grams. When a new vocabulary is extracted, the new vocabulary is added to the statistical information in the statistical information storage unit 133a.

  Further, the statistical information update unit 114 performs processing to update these values when the minimum vocabulary ID, the maximum vocabulary ID, and the maximum text size in the structure template storage unit 132 need to be updated. For example, when the structure information as shown in FIG. 6 is registered in the structure template storage unit 132, if the vocabulary with the maximum vocabulary ID = 522 appears as the text element corresponding to the structure element with TID = 4, the maximum vocabulary ID Is updated to 522.

  The document registration unit 115 adds a parent-child sibling relationship to each node developed by the structure information extraction unit 111 into a tree structure, and stores it in the structured document storage unit 131.

  The search processing unit 120 executes a process of searching for a structured document from the structured document storage unit 131 in accordance with the search query received from the client 300, and includes a query analysis unit 121, a constraint adding unit 122, a query planning unit, and the like. 123 and a query execution unit 124.

  The processing of the search processing unit 120 is executed by the method of Patent Document 1 in principle. In the following, a part related to the vocabulary index process in the search process will be described in detail. However, in the actual process, various processes other than the vocabulary index process are executed.

  In the method of Patent Document 1, a query graph, which is a search condition of a hierarchical structure, is created from an internal format obtained by analyzing a query language in accordance with a constraint-dependent plan creation method. Then, with the goal of realizing all the variables included in the created query graph, data representing combinations of possible values (candidate sets) of a variable set called a table are sequentially generated. Here, a processing unit for generating one table is called an operator. The results of each operator are stored in the storage unit as a candidate set.

  The query analysis unit 121 parses the search query received by the communication unit 101 from the client 300 and creates a query graph as an analysis result. For the creation of the query graph, any conventionally used method such as the method described in Patent Document 1 can be applied.

  FIG. 9 is an explanatory diagram illustrating an example of a search query. The query shown in the figure is obtained by acquiring “patent” document including “structured document” and “XML” in the title element and ID = 3, and retrieving the search result data surrounded by “<search result>” tags. Represents search conditions to be output.

  FIG. 10 is an explanatory diagram illustrating an example of a query graph. The query graph in the figure is a query graph created by the query analysis unit 121 analyzing the search query in FIG. 9.

  As shown in FIG. 10, the query graph is represented by a tree structure including nodes corresponding to each structural element of the structural information. For example, node 2 in the query graph of FIG. 10 indicates that the structural element of the “patent” document corresponds. Further, for example, the node 3 corresponds to the title tag, and the node 4 corresponds to the text element under the title tag.

  When there is a search condition (hereinafter referred to as a search key) for a structural element, the search key is associated with a node corresponding to the structural element to be searched for the search key. For example, in FIG. 10, “ftcontains“ structured document ”&&“ XML ”” is associated with the text element under the title tag corresponding to the node 4 as a search key.

  The constraint adding unit 122 obtains a constraint relationship between nodes in the query graph created by the query analysis unit 121, acquires a candidate for a structural element that satisfies the constraint, and adds the candidate to the query graph. Note that such a structural element candidate corresponds to one of the intermediate candidates.

  As constraints, there are constraints on structures (hereinafter referred to as structure constraints) that must be satisfied by candidate structural elements, and constraints on vocabularies (hereinafter referred to as vocabulary constraints).

  The structure constraint is a constraint set by node level information corresponding to the entire text. For example, a constraint that the node 4 of the query graph must be a text element under the title tag corresponds to the structure constraint. In this case, a structural element with TID = 7 is acquired as a corresponding structural element candidate in the node 4. The structure information is premised on the contents shown in FIGS. 5 and 6 being stored. Similarly, a structural element with TID = 4 is obtained as a candidate for node 2 and a structural element with TID = 5 is obtained as a candidate for node 6.

  In addition, the constraint adding unit 122 can also make a constraint that two certain structural elements exist in the same document (hereinafter referred to as the same document constraint) as the structure constraint.

  Vocabulary constraints are constraints set by vocabulary level information of text. For example, a vocabulary corresponding to a node with TID = 7 corresponds to a vocabulary constraint in which a vocabulary ID must exist between a minimum vocabulary ID = 21 and a maximum vocabulary ID = 1045. The minimum vocabulary ID and the maximum vocabulary ID can be acquired from the structure template storage unit 132 including feature information for each node as shown in FIG.

  In addition, the constraint adding unit 122 can also set a constraint such as a neighborhood search on the condition of the appearance position of a vocabulary as a vocabulary constraint.

  As shown in FIG. 1, the constraint adding unit 122 includes a feature information acquisition unit 122a, a feature information calculation unit 122b, and a TID candidate deletion unit 122c.

  The feature information acquisition unit 122a acquires, from the structure template storage unit 132, the minimum vocabulary ID and the maximum vocabulary ID that are feature information associated with the TID candidates included in each node of the query graph.

  The feature information calculation unit 122b calculates feature information representing the feature of the search key included in each node of the query graph. Specifically, the feature information calculation unit 122b calculates the vocabulary as a result of dividing the search key into N grams as the feature information of the search key.

  The TID candidate deletion unit 122c acquires, from the TID candidates included in each node of the query graph, the vocabulary ID of the vocabulary obtained by dividing the search key that is the feature information calculated by the feature information calculation unit 122b by the feature information acquisition unit 122a. Candidates outside the range of the minimum vocabulary ID and the maximum vocabulary ID are deleted.

  The query planning unit 123 refers to the query graph obtained by the query analysis unit 121 and information on candidate structure elements that satisfy the constraint relationship added by the constraint adding unit 122, and the plan (processing that minimizes the processing cost) Order). As a plan creation method, any conventionally used method such as the method described in Patent Document 1 can be applied. In the N-gram method, in order to determine the processing order at the gram level, the cost for each gram and the cost of the entire search vocabulary are calculated separately.

  11 and 12 are explanatory diagrams showing an example of vocabulary frequency information used in processing cost calculation. FIGS. 11 and 12 show an example in which the processing cost is simply calculated based on the vocabulary frequency.

  In order to narrow down search result candidates at an early stage, the query planning unit 123 creates a plan for executing a search from a vocabulary with a low frequency in principle. For example, when the frequency of the vocabulary extracted from the text “structured document” is a value as shown in FIG. 11, a plan for executing a search from “structured” which is the least frequent (30) vocabulary is created. Is done.

  It is not necessary to search continuously for vocabulary in the same text. For example, the vocabulary frequency extracted from the text “XML” as shown in FIG. 12 and the vocabulary frequency as shown in FIG. 11 are processed in the order of “structured”, “chemical sentence”, and “L”. You may make a plan.

  In addition, a new constraint may be added by executing a certain operator, but in the case of an index obtained by dividing it by the N-gram method, if any vocabulary that does not satisfy the condition exists, At that point, subsequent processing is interrupted.

  The query execution unit 124 executes a query according to the plan created by the query planning unit 123, and includes an index search unit 124a and a search unit 124b.

  The index search unit 124a executes a vocabulary index processing operator that performs lexical index processing.

  The search unit 124b searches for structured documents that satisfy the search conditions by executing the remaining operators.

  Next, a structured document storage process performed by the structured document search device 100 according to the first embodiment configured as described above will be described. FIG. 13 is a flowchart showing the overall flow of structured document storage processing in the first embodiment.

  First, the communication unit 101 receives an XML document that is input data from the client 300 (step S1301).

  Next, the structure information extraction unit 111 parses the input data and expands it into a tree structure format, and extracts structure elements from the expanded tree structure (step S1302).

  Next, the structure template determining unit 112 refers to the tree structure and the structure information stored in the structure template storage unit 132, and determines the structure element ID (TID) corresponding to each node of the tree structure. A determination process is executed (step S1303). Details of the structure template determination process will be described later.

  Next, the index registration unit 113 generates lexical index information and registers it in the transposed file storage unit 133b (step S1304). Information on the TID, document ID, element ID, and occurrence position necessary for the lexical index information registered in the transposed file is acquired or calculated in the structure template determination process.

  Next, the statistical information update unit 114 updates the statistical information in the statistical information storage unit 133a (step S1305). Specifically, the statistical information update unit 114 uses the frequency value of the statistical information stored in the statistical information storage unit 133a as the number of occurrences in the text element for a vocabulary obtained by dividing the character string of the text element into N grams. Update to the value added by minutes.

  Further, when there is a change in the minimum vocabulary ID, the maximum vocabulary ID, and the maximum text size for each TID calculated in the structure template determination process, the statistical information update unit 114 uses the values after the change in the structure template storage unit 132. The feature information (minimum vocabulary ID, maximum vocabulary ID) is updated.

  Next, the document registration unit 115 adds a parent-child sibling relationship to each node of the tree structure, and registers the structured document expressed in a table format in the structured document storage unit 131 (step S1306). For example, when an XML document as shown in FIG. 2 is input and a tree structure as shown in FIG. 3 is obtained by the structure information extraction unit 111, the tree structure is expressed in a tabular format with the information as shown in FIG. The document is registered.

  The above is the overall flow of the structured document storage process. Next, the structure template determination process in step S1303 will be described. FIG. 14 is a flowchart illustrating an overall flow of the structure template determination process according to the first embodiment.

  First, the structure template determination unit 112 executes vocabulary division for a text element among the structure elements extracted by the structure information extraction unit 111 (step S1401). As the vocabulary dividing method, an N-gram method is adopted.

  Next, the structure template determination unit 112 deletes duplicate vocabulary from the divided vocabulary, and adds a vocabulary ID to the new vocabulary (step S1402). In the case of an existing vocabulary, the vocabulary ID is acquired from the statistical information storage unit 133a. The number of occurrences of the vocabulary is stored in a storage unit (not shown) such as a RAM (Random Access Memory) because it is used later when the frequency of the statistical information storage unit 133a is updated.

  Next, the structure template determination unit 112 calculates feature information (maximum vocabulary ID, minimum vocabulary ID, maximum text size) for each TID stored in the structure template storage unit 132 (step S1403). Specifically, the structure template determination unit 112 calculates the maximum value as the maximum vocabulary ID and the minimum value as the minimum vocabulary ID among the vocabulary IDs added or obtained in step S1402. Further, the character string length of the corresponding text element is calculated as the maximum text size.

  Next, the structural template determination unit 112 refers to the structural template storage unit 132, acquires one structural element from the expanded structural elements of the tree structure, and acquires a structural element that matches the structural constraints of the tree structure (step) S1404).

  Next, the structure template determination unit 112 determines whether or not a matching structural element has been acquired (step S1405). If it has not been acquired (step S1405: NO), a new structural element is created (step S1405). S1406).

  Next, the structure template determination unit 112 sets the TID of the created structural element as the TID corresponding to the structural element of the tree structure currently being processed (step S1407).

  If a matching structural element is acquired in step S1405 (step S1405: YES), the structural template determination unit 112 uses the TID of the acquired structural element as the TID corresponding to the structural element of the currently processed tree structure. (Step S1408).

  Next, the structure template determination unit 112 determines whether or not all tree structure elements have been processed (step S1409). If all structure elements have not been processed (step S1409: NO), the next step is performed. The process is repeated for the structural element (step S1404).

  If all the structural elements have been processed (step S1409: YES), the structural template determination process ends.

  Next, a structured document search process performed by the structured document search apparatus 100 according to the first embodiment configured as described above will be described. FIG. 15 is a flowchart showing an overall flow of the structured document search process according to the first embodiment.

  First, the communication unit 101 receives a search query from the client 300 (step S1501). Next, the query analysis unit 121 analyzes the received search query and creates a query graph (step S1502).

  Next, the constraint adding unit 122 executes a constraint adding process for adding a constraint to the query graph created by the query analyzing unit 121 (step S1503). Details of the constraint addition processing will be described later.

  Next, the query planning unit 123 refers to the query graph to which the constraint is added by the constraint adding unit 122, and creates a search execution plan that minimizes the processing cost (step S1504).

  Next, the index search unit 124a executes a transposed file scan process that is a process related to the vocabulary index in the search process (step S1505). In the transposed file scan processing, processing for narrowing down index candidates to be searched is executed according to the added constraint. Details of the transposed file scanning process will be described later.

  Next, the search unit 124b executes a process corresponding to the remaining plan other than the transposed file scan process in the search process, and searches for a structured document that satisfies the search condition (step S1506).

  Next, the communication unit 101 transmits a structured document as a search result to the client 300 (step S1507), and the structured document search process is terminated.

  Next, details of the constraint addition processing in step S1503 will be described. FIG. 16 is a flowchart showing an overall flow of the constraint addition processing in the first embodiment.

  First, the constraint adding unit 122 acquires a TID candidate set (hereinafter referred to as a TID candidate set) that satisfies the structure constraints of each node of the query graph (step S1601). For example, since the node 4 of the query graph as shown in FIG. 10 is a text element under the title tag, the structure information TID 7 as shown in FIG. 5 can be acquired as a candidate.

  Note that TID candidates satisfying the structural constraints can be acquired by the processing in step S1601. In steps S1602 to S1610 below, processing is performed to delete candidates that do not satisfy the lexical constraints from the TID candidates.

  Next, the constraint adding unit 122 acquires a TID candidate set Pi for an unprocessed node (step S1602). At this time, the feature information acquisition unit 122a acquires the minimum vocabulary ID and the maximum vocabulary ID associated with each TID candidate from the structure template storage unit 132 in advance.

  Next, the feature information calculation unit 122b divides the search key vocabulary corresponding to Pi (step S1603). For example, the search key “ftcontains“ structured document ”&&“ XML ”” is associated with the node 4 of the query graph of FIG. Therefore, the constraint adding unit 122 divides the “structured document” and “XML” into words. Vocabulary division is performed by the method using N-gram division as described above.

  Next, the constraint adding unit 122 acquires a logical condition corresponding to Pi (step S1604). For example, in the case of node 4 in the query graph of FIG. 10, an AND condition (“&&”) is acquired.

  Next, the constraint adding unit 122 determines whether the acquired logical condition is only an AND condition (step S1605). In the case of only the AND condition (step S1605: YES), among the TID candidates of Pi, any of the vocabulary IDs of the vocabulary divided from the search key is between the minimum vocabulary ID and the maximum vocabulary ID corresponding to the TID. TID candidates that are out of range are acquired (step S1606).

  For example, when the search key is “structured document”, the vocabulary ID of the vocabulary divided from the search key takes a value between 22 and 235 as shown in FIG. Therefore, for example, assuming that the structure information shown in FIG. 6 is set, the value of the vocabulary ID is a range between the minimum vocabulary ID (= 21) and the maximum vocabulary ID (= 1045) of TID7. Fits within. Therefore, TID7 is not acquired in this step. On the other hand, for TID6, there is a vocabulary ID that is outside the range of the minimum vocabulary ID (= 25) and the maximum vocabulary ID (= 103) (vocabulary ID = 22, 178, 235). For this reason, TID6 is acquired as a corresponding TID candidate in this step.

  If it is determined in step S1605 that the AND condition is not the only condition (step S1605: NO), the constraint adding unit 122 determines whether the acquired logical condition is only the OR condition (step S1607). In the case of only the OR condition (step S1607: YES), among the TID candidates of Pi, all the vocabulary IDs of the vocabulary divided from the search key are ranges between the minimum vocabulary ID and the maximum vocabulary ID corresponding to the TID. An outside TID candidate is acquired (step S1608).

  Next, the constraint adding unit 122 determines whether or not a corresponding TID candidate has been acquired in step S1606 or step S1608 (step S1609). When the corresponding TID candidate is acquired (step S1609: YES), the TID candidate deletion unit 122c deletes the corresponding TID candidate from Pi (step S1610).

  In this way, TID candidates can be appropriately narrowed down according to whether the logical condition is an AND condition or an OR condition.

  If it is determined in step S1607 that only the OR condition is not satisfied (step S1607: NO), the position condition determination process after step S1611 is executed without performing the TID candidate deletion process.

  After deleting the TID candidate in step S1610, or when it is determined in step S1607 that only the OR condition is not satisfied, the constraint adding unit 122 specifies a condition related to a position such as neighborhood search (position condition) from the search key corresponding to Pi. ) Is acquired (step S1611).

  Next, the constraint adding unit 122 determines whether or not each TID candidate satisfies the acquired position condition (step S1612). If not satisfied (step S1612: NO), the corresponding TID candidate is deleted from Pi. (Step S1613).

  For example, if the position condition "" XML "&&" database "distance at most 5" is specified, at least 14 characters (= length of the character string in the condition + value of the interval = 9 + 5) are included in the text element. Must be present. On the other hand, for example, if the maximum text size associated with the TID of the structural element corresponding to the text element is less than 14, it can be deleted from Pi because the position condition is not satisfied.

  The maximum text size can be acquired from the structure template storage unit 132. In this way, TID candidates can be narrowed down according to the vocabulary position conditions.

  Next, the constraint adding unit 122 determines whether or not all the nodes have been processed (step S1614). If all the nodes have not been processed (step S1614: NO), the next processing is performed on the nodes. Is repeated (step S1602).

  When all the nodes have been processed (step S1614: YES), the constraint adding unit 122 acquires the relationship between the TIDs that must exist in the same document for the TIDs in all the TID candidate sets, The same document restriction is added (step S1615).

  For example, in the case of the query graph as shown in FIG. 10, TID = 7 and TID = 5 corresponding to the nodes 4 and 6 must exist in the “patent” document (TID = 4), respectively. A constraint is added. In the following, the same document restriction is described in a format such as “TID7 <-> TID5”.

  As described above, in the constraint addition process, the number of TID candidates that are intermediate candidates is determined not only by the structure constraint but also by the vocabulary constraint represented by the feature information representing the vocabulary feature of the text element stored when the structured document is registered. Can be narrowed down. Therefore, a high-speed full-text search function for structured documents can be realized.

  Next, details of the transposed file scanning process in step S1505 will be described. FIG. 17 is a flowchart illustrating an overall flow of the transposed file scan process according to the first embodiment.

  First, the index search unit 124a acquires the structural constraint T added by the constraint adding unit 122 and the set G of vocabulary IDs of the vocabulary (hereinafter referred to as search vocabulary) divided from the search key (step S1701).

  Next, the index search unit 124a acquires a node that executes the query according to the plan created by the query planning unit 123 (step S1702). For example, in the first process, the node corresponding to the search vocabulary with the lowest frequency is acquired.

  Next, the index search unit 124a acquires the transposed file corresponding to the search vocabulary from the transposed file storage unit 133b (step S1703). Specifically, the transposed file number corresponding to the vocabulary ID of the search vocabulary is acquired from the statistical information storage unit 133a, and the transposed file identified by the acquired transposed file number is acquired from the transposed file storage unit 133b.

  Next, the index search unit 124a extracts lexical index information from the transposed file (step S1704). Next, the index search unit 124a determines whether or not the TID of the lexical index information is included in the structure constraint T (step S1705).

  When included in the structure constraint T (step S1705: YES), the index search unit 124a adds the current lexical index information to the index candidate set (step S1706).

  After adding to the candidate set in step S1706 or in step S1705, when it is determined that the TID of the lexical index information is not included in the structure constraint T (step S1705: NO), the index search unit 124a selects all vocabularies. It is determined whether index information has been processed (step S1707).

  If all lexical index information has not been processed (step S1707: NO), the next lexical index information is acquired from the transposed file and the process is repeated (step S1704).

  If all vocabulary index information has been processed (step S1707: YES), the index search unit 124a determines whether all search vocabularies have been processed (step S1708). If not all the search vocabularies have been processed (step S1708: NO), the process is repeated for the next search vocabulary (step S1703).

  If all search vocabularies have been processed (step S1708: YES), the index search unit 124a determines whether all nodes have been processed (step S1709). If all the nodes have not been processed (step S1709: NO), the next node is acquired and the process is repeated (step S1702).

  If all nodes have been processed (step S1709: YES), the transposed file scan process is terminated.

  A TID set is created by eliminating duplication from the TIDs included in the index candidate set obtained by the above-described processing, and the created TID set is a new restriction when executing the next operator processing. Used as a condition.

  Next, a specific example of structured document search processing by the structured document search apparatus 100 according to the first embodiment will be described. FIG. 18 is an explanatory diagram illustrating an example of the structure information stored in the structure template storage unit 132. FIG. 19 is an explanatory diagram showing an example of the input search query.

  Hereinafter, the structured document search process when the structure information as shown in FIGS. 6 and 18 is stored in the structure template storage unit 132 and the search query as shown in FIG. 19 is input will be described.

  The search query shown in FIG. 19 is a query described in accordance with the XQFT language standard. In addition, this query satisfies the restriction that “structured document” and “XML” are included as candidates for the “title” element, and both exist within a distance of two characters or less, and the “inventor” element This means that the search condition is “search for patent documents including“ Tanaka ””.

FIG. 20 is an explanatory diagram illustrating an example of a query graph that is a result of analyzing a search query. The figure shows a query graph as a result of analyzing the search query of FIG. As the structural constraints, the following four constraints (A1 to A4) are added.
A1: Node 6 (TID11, TID19) → Tanaka A2: Node 4 (TID6, TID7, TID15) → XML
A3: Node 4 (TID6, TID7, TID15) → structured document A4: Node 4 and node 6 satisfy the same document constraint

  For example, “Node 6 (TID11, TID19) → Tanaka” means that node 6 must have TID11 or TID19 as a structural element candidate, and “Tanaka” is set as a search key for the corresponding text. It represents that.

  A4 represents the structure constraint of the same document constraint. For example, when the node 4 is TID6, the node 6 must have TID11. Therefore, the same document restriction of A4 can also be expressed as “{TID6, TID7} <−> {TID11}” and “{TID15} <−> {TID19}”.

In addition, the following two constraints (B1, B2) are added as vocabulary constraints.
B1: Node 4 includes all vocabulary IDs corresponding to “structure”, “structured”, “chemical sentence”, “document”, “book”, “XML”, “ML”, “L” B1: node 6 includes vocabulary IDs corresponding to “Tanaka” and “Middle”

  In the case of partial match search such as ftcontains, a vocabulary having a character string length of N grams or less is not included in the conditions, but is added here for convenience of explanation.

  For example, since the search vocabulary for the search key “structured document” is divided as shown in FIG. 11, the distribution of the vocabulary ID is {22, 32, 55, 178, 235}.

  The minimum vocabulary ID of TID6 is 25 from FIG. 6, and since there is a vocabulary ID (22) smaller than 25, TID6 is deleted from the candidates (step S1609: YES, step S1610). Since TID7 and TID15 satisfy the restriction (step S1609: NO), they are not deleted from the candidates.

  Similarly, the vocabulary ID distribution of the search vocabulary for the search key “XML” is {43, 58, 123}, and TID7 and TID15 are not deleted from the candidates because they satisfy the constraints.

  Similarly for node 6, the vocabulary IDs of “Tanaka” and “Middle”, which are the search vocabulary, are compared with the minimum vocabulary ID and the maximum vocabulary ID corresponding to TID11 and TID19. Here, it is assumed that both are left as candidates.

Moreover, the following restrictions (C1) are added as restrictions by position conditions.
C1: Node 4 has “structured document” and “XML” at a distance of two or more characters

  In this case, since the text size of the text corresponding to the node 4 is at least 10 (= “structured document” size + “XML” size + 2 = 5 + 3 + 2) characters, for example, the corresponding maximum text size is 10 If there is a TID candidate that is less than this, it can be excluded at this stage (step S1613).

  FIG. 21 is an explanatory diagram illustrating an example of a query graph in a state where candidates are narrowed down by such processing. Compared with FIG. 20, FIG. 21 shows that TID candidates for the node 4 are narrowed down to TID7 and TID15.

  In this way, after the constraint is added by the constraint adding unit 122, the query planning process by the query planning unit 123 is executed. The query planning unit 123 creates a plan that preferentially processes a search vocabulary with a low frequency.

  FIG. 22 is an explanatory diagram showing the frequency for each search vocabulary. As shown in the figure, since the frequency for the vocabulary “Tanaka” is the lowest (20), the transposed file scan process is executed from the node corresponding to the vocabulary (step S1702).

  As described above, in the structured document search apparatus according to the first embodiment, feature information representing the feature of an element corresponding to each structure element of the structured document is stored, and a transposed file scan is performed with reference to the feature information. It is possible to reduce the number of reading of unnecessary lexical index information at an early stage. As a result, it is possible to reduce the cost of the candidate combination processing required for the search based on the search condition considering the appearance position level of the vocabulary. For this reason, it is possible to execute a full-text search function of a structured document by a query in which position information such as XQFT is combined with designation and structure constraints at high speed.

(Second Embodiment)
The structured document search apparatus according to the second embodiment stores the feature value representing the vocabulary feature included in the corresponding text element in the vocabulary index information, and refers to the number of candidates by referring to the transposed file scan processing. By narrowing down, a higher-speed search is possible.

  In the second embodiment, a transposed file including lexical index information in which feature values calculated from vocabulary included in a text element are associated is stored in the transposed file storage unit 133b. Is different. The feature value is calculated by the structure template determination unit 112 in the structure template determination process.

  Further, the second embodiment is different from the first embodiment in that the index search unit 124a refers to the feature value to narrow down index candidates during the transposed file scan process.

  FIG. 23 is an explanatory diagram illustrating an example of a data structure of an inverted file stored in the inverted file storage unit 133b according to the second embodiment.

  As shown in the figure, the transposed file of the transposed file storage unit 133b stores lexical index information in which TIDs, feature values, document IDs, element IDs, and occurrence positions are associated with each other.

  The feature value is information representing the feature of the element of the structured document identified by the document ID and the element ID, and is calculated by the structure template determination unit 112.

  The structure template determination unit 112 in the second embodiment performs a process of calculating feature values for text elements of the structured document. The feature value is calculated from the minimum vocabulary ID and the maximum vocabulary ID which are the minimum and maximum vocabulary IDs of the vocabulary included in the text. Specifically, a value obtained by dividing the minimum vocabulary ID and the maximum vocabulary ID by a conversion coefficient α (for example, 10) is calculated as a feature value.

  For example, when the minimum vocabulary ID of the vocabulary included in the text is 22 and the maximum vocabulary ID is 301, the result of dividing each by α = 10 (2, 30) is calculated as the feature value of the text. Is done.

  Note that the feature value calculation method is not limited to this, and can be calculated by any method as long as the information represents the feature of the text. For example, a method of using character code values such as alphanumeric characters and symbols, or a method of preparing a plurality of character string patterns in advance and designating which pattern is included by a pattern number can be applied.

  The index search unit 124a according to the second embodiment performs processing for narrowing down index candidates with reference to feature values when processing a vocabulary index. Specifically, the index search unit 124a performs a conversion process similar to that used when calculating the feature value on the vocabulary ID of the vocabulary included in the search key, and converts the vocabulary ID within the range of the feature value of the candidate vocabulary index information. If the vocabulary ID value is not included, the lexical index information is excluded from index candidates. Note that the feature value range refers to a range from 2 to 30 when the feature value is (2, 30), for example.

  Next, the structure template determination process by the structured document search apparatus according to the second embodiment configured as described above will be described. FIG. 24 is a flowchart illustrating an overall flow of the structure template determination process according to the second embodiment.

  The second embodiment is different from the first embodiment in that the process of calculating the feature value of the text element is inserted in step S2404. The other processes are the same as those in the first embodiment, and a description thereof will be omitted.

  In step S2404, the structure template determination unit 112 calculates a text feature value from the maximum vocabulary ID and the minimum vocabulary ID calculated in step S2403 (step S2404).

  FIG. 25 is an explanatory diagram for explaining the calculated feature values. The figure shows the feature values calculated for the text “XML database”.

  For the sake of explanation, this figure shows a part of the information in the statistical information storage unit 133a (vocabulary and vocabulary ID) and the information in the transposed file storage unit 133b (occurrence position, feature value, document ID, element ID, TID). Represents information associated with.

  In the example shown in the figure, the vocabulary ID = 12 of the vocabulary “data” is the minimum vocabulary ID, and the vocabulary ID = 147 of the vocabulary “table” is the maximum vocabulary ID. Therefore, a value (1, 14) obtained by dividing the minimum vocabulary ID and the maximum vocabulary ID by dividing by the conversion coefficient α = 10 is calculated as the feature value.

  The feature value calculated in this way is finally stored in the transposed file of the transposed file storage unit 133b. In the above example, nine lexical index information having the same document ID, element ID, TID, and characteristic value but different occurrence positions are newly stored in the transposed file.

  For a vocabulary that occurs a plurality of times in the same text element, the feature value may be calculated by dividing the distribution of vocabulary IDs into a plurality of sections. FIG. 26 is an explanatory diagram for explaining the feature values calculated in such a configuration. This figure shows the feature values calculated for the text “Relationship between documents”.

  In this example, since the vocabulary “document” occurs twice in the same text, the lexical ID distribution (22, 22, 26, 55, 103, 178, 235, 301) is divided into two sections (22, 22, 26, 55) and (103, 178, 235, 301). The characteristic values (2, 5) and (10, 30) calculated in each of the divided sections are stored in association with the vocabulary “document”.

  In this case, unlike FIG. 25, the feature value is (2, 5) or (10, 30) for the vocabulary “document”, and the feature value is (2, 30) for other vocabularies. One lexical index information is stored in the transposed file.

  As described above, since a more detailed feature value can be assigned to a vocabulary having a high occurrence frequency and a high processing cost, an effect of improving the accuracy of narrowing down by the feature value can be obtained.

  Next, a transposed file scan process by the structured document search apparatus according to the second embodiment configured as described above will be described. FIG. 27 is a flowchart illustrating an overall flow of the transposed file scan process according to the second embodiment.

  The node acquisition process, the transposed file acquisition process, and the vocabulary index information acquisition process from step S2701 to step S2705 are the same processes as steps S1701 to S1705 in the structured document search apparatus according to the second embodiment. The description is omitted.

  If it is determined in step S2705 that the TID of the lexical index information is included in the structure constraint T (step S2705: YES), the index search unit 124a compares the feature value range in the lexical index information with the set G The value of the vocabulary ID is converted (step S2706). Specifically, the vocabulary ID value of the search vocabulary is divided by the conversion coefficient α used when calculating the feature value. For example, if the vocabulary ID of the search vocabulary is 100, the value of the vocabulary ID is converted to 10 by dividing by the conversion coefficient α = 10.

  Next, the index search unit 124a determines whether or not the converted vocabulary ID value is included in the range of feature values in the vocabulary index information (step S2707). For example, if the feature value is (2, 5) and the converted vocabulary ID value is 10, it is determined that it is not included in the range (between 2 and 5).

  If it is determined that the converted vocabulary ID value is included in the range of feature values in the vocabulary index information (step S2707: YES), the index search unit 124a sets the current vocabulary index information as an index candidate set. It adds (step S2708).

  Note that it may be configured to determine whether or not the determination by the feature value is possible according to the processing cost in consideration of the cost of the conversion processing, the relationship between the number of candidates, and the like, or one value (minimum value) of the feature values Alternatively, only the maximum value may be used.

  Since the end determination process from step S2709 to step S2711 is the same as that from step S1707 to step S1709 in the structured document search apparatus according to the second embodiment, the description thereof is omitted.

  As described above, in the inverted file scan process, the feature value of the text registered in the inverted file is referred to when the index is created, and the value corresponding to the feature value calculated from the vocabulary included in the search condition matches the feature value of the index. If not, the index can be excluded from the candidates. For this reason, the number of candidates at the time of lexical index processing can be further narrowed down, and a high-speed full-text search function for structured documents can be realized.

  Here, a specific example of data processed in the transposed file scanning process will be described. FIG. 28 is an explanatory diagram of an example of candidates processed in the transposed file scan process. In the figure, {TID7, TID11} (representing that TID7 and TID11 are TIDs of structural elements satisfying the structure constraint) is acquired as the structure constraint T, and {135, 292, 356} is acquired as the vocabulary ID set G. (Step S2701).

  It is assumed that “XML” is designated as the search key. Therefore, the vocabulary ID set G includes vocabulary IDs corresponding to the vocabulary “XML”, the vocabulary “ML”, and the vocabulary “L”, respectively.

  Further, in the figure, it is assumed that a transposed file including lexical index information of five TIDs is stored in the transposed file storage unit 133b as a transposed file corresponding to the vocabulary “XML”. In the figure, the element ID and the occurrence position are omitted from the lexical index information.

  Under the premise as described above, the TID of the transposed file is narrowed down to three due to the structure constraint condition (step S2704). Moreover, since the conversion values of the vocabulary IDs in the set G are 13, 29, and 35, respectively (step S2705), the vocabulary index information of document ID = 200 in which these conversion values are out of the feature value range is excluded. (Step S2707).

  As described above, the candidates are TID7 and TID11 according to the structure constraint alone, but the candidates can be narrowed down to only the index corresponding to TID7 by referring to the feature value.

  Next, an example in which the restriction is further narrowed down by repeating the process will be described. 29 and 30 are explanatory diagrams illustrating another example of candidates processed in the transposed file scanning process.

  In this example, it is assumed that structure information as illustrated in FIG. 18 is stored in the structure template storage unit 132.

  In this state, it is assumed that the candidates for TID7 and TID15 are narrowed down to only candidates for TID15 by scanning the transposed file 1 (FIG. 29), and then the structure constraints {TID11, TID19} are processed (FIG. 30).

  In this case, since TID 19 is the only TID satisfying the same document restriction as TID 15 in the structure information as shown in FIG. Therefore, in the transposed file 2, only the vocabulary index information of the document ID = 62 having TID19 needs to be a candidate.

  Next, a specific example of structured document search processing by the structured document search device according to the second embodiment will be described.

  Here, a description will be given on the assumption of a specific example of the structured document search processing described with reference to FIGS. 18 to 22 in the first embodiment. That is, it is assumed that a query graph as shown in FIG. 21 is generated by the processing described with reference to FIGS. 18 to 22, and the frequency of the search vocabulary in the query graph has a value as shown in FIG.

  FIG. 31 is an explanatory diagram showing an example of candidates processed in the transposed file scanning process at this time. Assume that the index search unit 124a performs the transposed file scan process using [TID11, TID19] as a structure constraint, and as a result, only TID11 remains as a candidate set.

  Although there are three candidates for TID11, the candidate for document ID = 200 is deleted from the candidates by comparison of feature values. If the vocabulary ID of “Tanaka” is 136, the value of the converted vocabulary ID is 13 (step S1706), and is not included in the range of 80 to 90 that is the feature value.

  Since TID19 is excluded from the structure constraint, TID15 is also excluded from the candidates in accordance with the same document constraint “{TID15} <−> {TID19}”. Therefore, in the node 4 to be processed after this, only the TID 7 excluding the TID 15 needs to be extracted as a structural constraint.

  FIG. 32 is an explanatory diagram showing an example of candidates processed in the transposed file scanning process at this time.

  The node 4 executes processing for the vocabulary “building” with the lowest frequency. Also in this case, only one candidate corresponding to the document ID = 43 is extracted as a result of narrowing down by the feature value. “Structured” is one of the vocabularies constituting “structured document”, and in this case, it is left in the form including the index to the position information and used for the next processing.

  As described above, in the structured document search apparatus according to the second embodiment, the feature value representing the feature of the vocabulary included in the text element is stored in the vocabulary index information, and the number of candidates is referred to during the transposed file scanning process. Can be narrowed down. For this reason, the full-text search function of a structured document by a query combining position information such as XQFT with designation and structure constraints can be executed at higher speed.

(Third embodiment)
The structured document search apparatus according to the third embodiment defines a division rule based on feature information for each structure element, and when the feature information of the element value corresponding to the structure element matches the division rule, The structure template storage unit 132 is updated by creating a new structural element by dividing the element.

  The third embodiment is different from the second embodiment in that a division rule for each structural element is set in the structural template storage unit 132. In addition, in the structured document registration process, the structure template determination unit 112 refers to the division rule and divides the structure element stored in the structure template storage unit 132 to create a new structure element. This is different from the second embodiment.

  FIG. 33 is an explanatory diagram illustrating an example of the data structure of the structure information stored in the structure template storage unit 132 according to the third embodiment. This figure shows an example in which the structure information is represented in a table format.

  As shown in the figure, in the third embodiment, the structure information stores the maximum number of generated vocabularies in the element, the number of times that the text size threshold is exceeded, the reference information, and the division method in further correspondence. is doing. A division rule is designated by the reference information and the division method.

  The maximum number of vocabulary generated in an element refers to the maximum value of the number of occurrences of each vocabulary generated in the corresponding text element. The number of times that the text size threshold is exceeded means the number of times that a predetermined text size threshold (for example, 100 characters) is exceeded.

  In the reference information, information to be applied for division determination is set. For example, the minimum vocabulary ID, the maximum vocabulary ID, the maximum text size, the maximum number of generated vocabularies in the element, the number of times exceeding the threshold of the text size, etc., which are characteristic information for each structural element, can be set. In addition to this, it is also possible to use frequently used vocabulary as a criterion. The figure shows an example in which a frequently used vocabulary with a vocabulary ID = 22 or 400 is extracted and set as reference information.

  In addition, a condition that the information set in the reference information satisfies is set as the division condition. When the information set in the reference information satisfies the division condition, the corresponding structural element is divided. The figure shows an example in which the division condition is set such that the number of occurrences of the vocabulary with the vocabulary ID = 22 or 400 defined as the frequent vocabulary exceeds a predetermined threshold of 30.

  Note that it is desirable to use feature information with a small calculation amount as the reference information in order to reduce the processing load. Further, in the initial stage, it may be configured not to set the reference information or the division condition, or may be configured to set a default value.

  In addition to this, for example, the number of times that the text size threshold (100) is exceeded may be used as the reference information, and the structural element may be divided on the basis of the number of times that the number exceeds 10 as a division condition. In this case, as a division method, for example, a method of dividing a text element having a text size of 100 or more and a text element having a text size of 99 or less into two structural elements is applied.

  In this way, for example, when a certain vocabulary occurs frequently in the same text element, it is determined that there is a possibility of causing performance degradation in the query processing, and the structural element of the structural template is divided in advance at the time of registration. be able to. For this reason, candidates can be narrowed down early at the stage of the transposed file scan process.

  In addition to the TID determination and feature value calculation functions, the structure template determination unit 112 according to the third embodiment determines whether the TID satisfies the division condition, and divides the TID when the division condition is satisfied. A process of determining the divided TID as the corresponding TID is performed.

  That is, the structure template determination unit 112 may divide the structure element according to the feature information of the corresponding text, instead of determining the TID based only on the structure information. Then, the structure template determination unit 112 determines the structure element newly created by the division as a structure element corresponding to the tree structure node of the structured document.

  Next, template determination processing by the structured document search apparatus according to the third embodiment configured as described above will be described. FIG. 34 is a flowchart showing the overall flow of the template determination process in the third embodiment.

  The vocabulary division processing and vocabulary ID addition processing from step S3401 to step S3402 are the same as the processing from step S2401 to step S2402 in the structured document search apparatus according to the second embodiment, and thus description thereof is omitted.

  After adding the vocabulary ID, the structure template determining unit 112 stores the feature information for each TID stored in the structure template storage unit 132 (maximum vocabulary ID, minimum vocabulary ID, maximum text size, maximum number of vocabulary in elements, text size The number of times that the threshold is exceeded is calculated (step S3403). For example, the appearance function of the vocabulary with the highest number of appearances among the vocabularies in the text element is calculated as the maximum number of occurrences in the element.

  Since the feature value calculation process and the structural element acquisition process from step S3404 to step S3406 are the same as the process from step S2404 to step S2406 in the structured document search apparatus according to the second embodiment, the description thereof is omitted.

  If a structural element is not acquired in step S3406 (step S3406: NO), the structural template determination unit 112 newly creates a structural element, sets default reference information and division conditions, and stores the structural template. Registered in the unit 132 (step S3407).

  Next, the structure template determination unit 112 sets the TID of the created structure element as a TID corresponding to the structure element of the tree structure currently being processed (step S3408).

  If a structural element is acquired in step S3406 (step S3406: YES), the structural template determination unit 112 selects the most suitable structural element according to the feature information corresponding to the structural element (step S3409). This process is a process executed to select one of the structural elements when a plurality of structural elements are acquired. Therefore, this process is not executed when only one structural element is acquired.

  For example, it is assumed that structure information as shown in FIGS. 18 and 33 is stored in the structure template storage unit 132, and TID6 and TID7 are acquired as the structure elements corresponding to the text elements under the title tag. In this case, for example, if “document” is included in the text element, since the vocabulary ID of “document” is 22, TID7 including the vocabulary ID between the minimum vocabulary ID and the maximum vocabulary ID is most suitable. Selected as a structural element.

  Next, the structure template determination unit 112 refers to the structure template storage unit 132 and acquires reference information and division conditions corresponding to the selected structure element (step S3410).

  Next, the structure template determination unit 112 refers to the acquired reference information, the division condition, and the feature information calculated in step S3403, and determines whether the structural element satisfies the division condition (step S3411).

  If the division condition is not satisfied (step S3411: NO), the TID of the structural element selected in step S3409 is set as the TID corresponding to the structural element of the tree structure currently processed (step S3412).

  When the division condition is satisfied (step S3411: YES), the structure template determination unit 112 newly creates a structure element, sets default reference information and division conditions, and registers them in the structure template storage unit 132 (step S3411). S3407).

  Since the end confirmation process in step S3413 is the same process as step S2410 in the structured document search apparatus according to the second embodiment, the description thereof is omitted.

  Next, a specific example of structured document registration processing by the structured document search device according to the third embodiment will be described.

  FIG. 35 is an explanatory diagram illustrating an example of the structure information stored in the structure template storage unit 132. FIGS. 36 and 37 are explanatory diagrams showing examples of structured documents to be registered.

  Here, an example of registering a structured document as shown in FIGS. 36 and 37 on the assumption that the structure information as shown in FIGS. 33 and 35 is stored in the structure template storage unit 132 will be described. In the following description, the processing for the title element will be described.

  For the title element, two structural element candidates are acquired from the structural information as shown in FIG. 35 (step S3405, step S3406: YES). Therefore, the structure template determination unit 112 selects the most suitable structural element according to the feature information corresponding to the structural element (step S3409).

  In this example, since the vocabulary “document” (vocabulary ID = 22) is included in the text element, TID7 including the vocabulary ID between the minimum vocabulary ID and the maximum vocabulary ID is selected as the most suitable structural element.

  Next, the structure template determination unit 112 determines whether or not a structural element division condition is satisfied (step S3411). Here, since the text element includes two vocabulary “documents” with the vocabulary ID = 22, which is a frequent vocabulary, it is assumed that the division condition that the number of appearances is 30 or more is satisfied (step S3411: YES).

  In this case, the structure template determination unit 112 divides TID7 and newly creates a structure element (step S3407). In the division, only the structural element corresponding to the text element under the title tag may be divided, or a new structural element including a parent element may be created and divided.

  FIG. 38 is an explanatory diagram showing an example of the structure information stored in the structure template storage unit 132 after the division. This figure shows an example in which structural elements are divided including a patent document which is a parent element.

  As described above, in the structured document search apparatus according to the third embodiment, the division rule based on the feature information is determined for each structure element, and the feature information of the element value corresponding to the structure element matches the division rule. In addition, the structural template storage unit can be updated by dividing the structural element to create a new structural element. As a result, the number of processing candidates increases, and a portion that can become a bottleneck of search processing can be detected in advance to divide the structural elements. For this reason, candidates can be narrowed down early at the stage of the transposed file scan process.

  As described above, the structured document search apparatus and the structured document search method according to the present invention manage a plurality of structured documents having different document structures in a structured document database having a hierarchical logical structure, and It is suitable for a structured document search apparatus and a structured document search method that perform a search specified by using position information as a condition, such as full-text search processing.

It is a block diagram which shows the structure of the structured document search apparatus concerning 1st Embodiment. It is explanatory drawing which shows an example of the structured document described by XML. It is explanatory drawing which shows an example of the data structure of the structured document stored in the structured document storage part. It is explanatory drawing which shows an example of the data structure of the structured document represented with the table format. It is explanatory drawing which shows an example of the data structure of the structure information stored in the structure template memory | storage part in 1st Embodiment. It is explanatory drawing which shows an example of the data structure of the structure template memory | storage part represented with the table format. It is explanatory drawing which shows an example of the data structure of the statistical information stored in the statistical information storage part. It is explanatory drawing which shows an example of the data structure of the transposed file stored in the transposed file storage part. It is explanatory drawing which shows an example of a search query. It is explanatory drawing which shows an example of a query graph. It is explanatory drawing which showed an example of the frequency information of the vocabulary used by processing cost calculation. It is explanatory drawing which showed an example of the frequency information of the vocabulary used by processing cost calculation. It is a flowchart which shows the whole flow of the structured document storage process in 1st Embodiment. It is a flowchart which shows the whole flow of the structure template determination process in 1st Embodiment. It is a flowchart which shows the whole flow of the structured document search process in 1st Embodiment. It is a flowchart which shows the whole flow of the restriction | limiting addition process in 1st Embodiment. It is a flowchart which shows the whole flow of the transposition file scan process in 1st Embodiment. It is explanatory drawing which shows an example of the structure information memorize | stored in the structure template memory | storage part. It is explanatory drawing which shows an example of the input search query. It is explanatory drawing which shows an example of the query graph which is the result of having analyzed the search query. It is explanatory drawing which shows an example of a query graph. It is explanatory drawing which showed the frequency for every search vocabulary. It is explanatory drawing which shows an example of the data structure of the transposition file stored in the transposition file memory | storage part in 2nd Embodiment. It is a flowchart which shows the whole flow of the structure template determination process in 2nd Embodiment. It is explanatory drawing for demonstrating the calculated feature value. It is explanatory drawing for demonstrating a feature value. It is a flowchart which shows the flow of the whole transposition file scan process in 2nd Embodiment. It is explanatory drawing which shows an example of the candidate processed by a transposition file scanning process. It is explanatory drawing which shows another example of the candidate processed by a transposition file scanning process. It is explanatory drawing which shows another example of the candidate processed by a transposition file scanning process. It is explanatory drawing which shows an example of the candidate processed by a transposition file scanning process. It is explanatory drawing which shows an example of the candidate processed by a transposition file scanning process. It is explanatory drawing which shows an example of the data structure of the structure information stored in the structure template memory | storage part in 3rd Embodiment. It is a flowchart which shows the whole flow of the template determination process in 3rd Embodiment. It is explanatory drawing which shows an example of the structure information stored in the structure template memory | storage part. It is explanatory drawing which shows an example of the structured document to register. It is explanatory drawing which shows an example of the structured document to register. It is explanatory drawing which shows an example of the structure information stored in the structure template memory | storage part after a division | segmentation.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Structured document search apparatus 101 Communication part 110 Storage processing part 111 Structure information extraction part 112 Structure template determination part 113 Index registration part 114 Statistical information update part 115 Document registration part 120 Search processing part 121 Query analysis part 122 Constraint addition part 122a Feature Information acquisition unit 122b Feature information calculation unit 122c TID candidate deletion unit 123 Query planning unit 124 Query execution unit 124a Index search unit 124b Search unit 131 Structured document storage unit 132 Structure template storage unit 133 Index storage unit 133a Statistical information storage unit 133b Transpose File storage unit 200 Network 300 Client

Claims (14)

A document having a hierarchical logical structure, which includes an element that is actual information corresponding to a structural element that is a unit of the logical structure; a document ID that uniquely identifies the structured document; Structured document storage means for storing
A structure information storage unit that stores structure information in which a structure ID that uniquely identifies the structure element is associated with first feature information that represents a feature of the element;
A search condition of a hierarchical structure that accepts input of search conditions for the structured document, analyzes the accepted search conditions, and has hierarchically structured nodes corresponding to the structural elements of the structured document. Analyzing means for obtaining a search condition for the hierarchical structure including the node in which the structure ID candidate to be searched and the search key for the structure ID candidate are associated with each other;
Acquisition means for acquiring, from the structure information storage means, the first feature information associated with the candidate structure ID included in each node for each node of the search condition of the hierarchical structure obtained by the analysis means; ,
Calculating means for calculating second feature information representing characteristics of the search key included in each node for each node of the search condition of the hierarchical structure obtained by the analyzing means;
For each node of the search condition of the hierarchical structure obtained by the analysis means, the second feature information calculated by the calculation means for the corresponding search key among the candidate structure IDs included in each node is: A deletion unit that deletes the candidate of the structure ID that does not match the first feature information acquired by the acquisition unit;
Based on the search condition of the hierarchical structure in which the deletion means deletes the candidate for the structure ID, the document ID corresponding to the structure ID satisfying the search condition of the hierarchical structure is obtained, and the document ID corresponding to the obtained document ID is obtained. Search means for searching the structured document from the structured document storage means;
A structured document retrieval apparatus characterized by comprising: A document having a hierarchical logical structure, which includes an element that is actual information corresponding to a structural element that is a unit of the logical structure; a document ID that uniquely identifies the structured document; Structured document storage means for storing
A structure information storage unit that stores structure information in which a structure ID that uniquely identifies the structure element is associated with first feature information that represents a feature of the element;
Index storage means for storing an index that associates the document ID, an element ID that uniquely identifies the element, and the structure ID of the structural element corresponding to the element;
A search condition of a hierarchical structure that accepts input of search conditions for the structured document, analyzes the accepted search conditions, and has hierarchically structured nodes corresponding to the structural elements of the structured document. Analyzing means for obtaining a search condition for the hierarchical structure including the node in which the structure ID candidate to be searched and the search key for the structure ID candidate are associated with each other;
Acquisition means for acquiring, from the structure information storage means, the first feature information associated with the candidate structure ID included in each node for each node of the search condition of the hierarchical structure obtained by the analysis means; ,
Calculating means for calculating second feature information representing characteristics of the search key included in each node for each node of the search condition of the hierarchical structure obtained by the analyzing means;
For each node of the search condition of the hierarchical structure obtained by the analysis means, the second feature information calculated by the calculation means for the corresponding search key among the candidate structure IDs included in each node is: A deletion unit that deletes the candidate of the structure ID that does not match the first feature information acquired by the acquisition unit;
Based on the search condition of the hierarchical structure in which the deletion means deletes the candidate structure ID, the document ID and the element ID corresponding to the structure ID satisfying the search condition of the hierarchical structure are searched from the index storage means. Index search means to perform,
Search means for searching the structured document storage means for the structured document corresponding to the document ID searched by the index search means;
A structured document retrieval apparatus characterized by comprising: The index storage means includes, for each vocabulary that is a character string included in the element, a vocabulary ID that uniquely identifies the vocabulary, and an occurrence frequency of the vocabulary in all structured documents of the structured document storage means And storing the index further corresponding to
The structure information storage means associates the minimum vocabulary ID, which is the minimum value of the vocabulary ID associated with the vocabulary included in the element, and the maximum vocabulary ID, which is the maximum value, as the first feature information. Memorize structural information,
The calculation means calculates the vocabulary ID of the vocabulary included in the search key of each node as the second feature information for each node of the hierarchical structure search condition obtained by the analysis means,
For each node of the hierarchical structure search condition obtained by the analyzing means, the deleting means is the vocabulary ID calculated by the calculating means for the corresponding search key among the candidate structure IDs included in each node. 3. The structure ID candidate not included between the minimum vocabulary ID and the maximum vocabulary ID, which is the first feature information acquired by the acquisition unit, is deleted. Structured document retrieval device.   The index storage means includes a vocabulary ID that uniquely identifies the vocabulary for each vocabulary that is all consecutive plural characters included in the element, and all structured documents in the structured document storage means of the vocabulary. 4. The structured document search apparatus according to claim 3, wherein the index further correlates with the occurrence frequency within the index is stored.   The index storage means includes, for each vocabulary that is a word obtained by morphological analysis of the element, a vocabulary ID that uniquely identifies the vocabulary, and in all structured documents of the structured document storage means of the vocabulary The structured document search apparatus according to claim 3, wherein the index further storing the occurrence frequency is stored. The index storage means stores an index in which the document ID, the element ID, the structure ID of the structural element corresponding to the element, and third characteristic information representing the characteristic of the element are associated with each other. ,
Second acquisition means for acquiring the third feature information associated with the document ID and the element ID searched by the index search means from the index storage means;
The second feature information calculated by the calculation unit for the search key included in the node of the search condition of the hierarchical structure used in the search from the document ID and the element ID searched by the index search unit A second deletion unit that deletes the document ID and the element ID that do not match the third feature information acquired by the second acquisition unit,
3. The structured document according to claim 2, wherein the retrieval unit retrieves the structured document corresponding to the document ID deleted by the second deletion unit from the structured document storage unit. Search device. The index storage means includes, for each vocabulary that is a character string included in the element, a vocabulary ID that uniquely identifies the vocabulary, and an occurrence frequency of the vocabulary in all structured documents of the structured document storage means The document ID, the element ID, the structure ID of the structural element corresponding to the element, and a minimum vocabulary ID that is a minimum value of the vocabulary ID associated with the vocabulary included in the element; Storing the index associating with the third feature information calculated based on the maximum vocabulary ID which is the maximum value;
The calculation means calculates the second feature information for each node of the search condition of the hierarchical structure obtained by the analysis means based on the vocabulary ID of the vocabulary included in the search key of each node,
The second deleting means includes the calculating means for the search key included in the search condition node of the hierarchical structure used for searching from the document ID and the element ID searched by the index searching means. 7. The structured document according to claim 6, wherein the calculated vocabulary ID deletes the document ID and the element ID that do not match the third feature information acquired by the second acquisition unit. Search device.   The index storage means includes the document ID, the element ID, the structure ID of the structural element corresponding to the element, the appearance position of the vocabulary included in the element, and the element for each appearance position. 8. The structured document search device according to claim 7, wherein an index that associates the third feature information that represents the feature is stored. The structure information storage means stores the structure information further associated with a condition for dividing the structure element based on the first feature information,
Receiving means for receiving input of the structured document;
Extracting means for extracting a hierarchical structure including at least one structural element from the structured document received by the receiving means;
Structure information determining means for determining the structure ID corresponding to each structure element included in the hierarchical structure extracted by the extracting means;
Whether the first feature information of the element corresponding to the structural element included in the hierarchical structure extracted by the extracting unit satisfies the condition corresponding to the structure ID determined by the structural information determining unit Determining means for determining
When the determination means determines that the condition is satisfied, the structure information corresponding to the structure ID determined by the structure information determination means is divided to create new structure information, and the structure information storage means 3. The structured document search apparatus according to claim 2, further comprising a structure information dividing unit for registration.   The structure information storage means uses the maximum character string length that is the maximum value of the character string length of the element corresponding to the structure element as the first feature information, and the maximum character string length exceeds a predetermined threshold value. 10. The structured document search apparatus according to claim 9, wherein the structure information further correlates with a condition for dividing the structure element when the structure element is divided.   The structure information storage means uses, as the first feature information, a maximum vocabulary number that is a maximum value of the vocabulary number in the element corresponding to the structural element, and the maximum vocabulary number exceeds a predetermined threshold value. 10. The structured document search apparatus according to claim 9, wherein the structure information further correlates with a condition for dividing the structure element when the structure element is divided.   The structure information storage means uses the number of times that the character string length of the element corresponding to the structure element exceeds a predetermined first threshold as the first feature information, and sets the number of times to the second 10. The structured document search apparatus according to claim 9, wherein the structure information is further associated with a condition for dividing the structure element when the threshold value of the structure element is exceeded. The index storage means includes, for each vocabulary that is a character string included in the element, a vocabulary ID that uniquely identifies the vocabulary, and an occurrence frequency of the vocabulary in all structured documents of the structured document storage means And storing the index further corresponding to
The structural information storage means includes the vocabulary stored in the index storage means, the vocabulary having an appearance frequency greater than a predetermined first threshold value appearing in the element corresponding to the structural element. 10. The structured document search device according to claim 9, wherein the structure information further stores a condition for dividing the structure element when the number of times exceeds a predetermined second threshold. The structured document search device
A document having a hierarchical logical structure, and receiving an input of a search condition of a structured document including an element that is actual information corresponding to a structural element that is a unit of the logical structure, and analyzes the received search condition. , A hierarchical search condition having nodes that are units of a structure corresponding to the structural element of the structured document, the structure uniquely identifying the structural element of the structured document to be searched An analysis step for obtaining a search condition for the hierarchical structure including the node that associates an ID candidate and a search key for the structure ID candidate;
For each node of the search condition of the hierarchical structure obtained by the analysis step, from the structure information storage means for storing the structure information in which the structure ID is associated with the first feature information representing the feature of the element. An acquisition step of acquiring the first feature information associated with the candidate structure ID included in the node;
A calculation step of calculating second feature information representing a feature of the search key included in each node for each node of the search condition of the hierarchical structure obtained in the analysis step;
For each node of the search condition of the hierarchical structure obtained in the analysis step, the second feature information calculated by the calculation step for the corresponding search key among the structure ID candidates included in each node is: A deletion step of deleting the candidate of the structure ID that does not match the first feature information acquired by the acquisition step;
An index that stores an index that associates a structured document, a document ID that uniquely identifies the structured document, an element ID that uniquely identifies the element, and the structure ID of the structural element corresponding to the element The document ID and the element ID corresponding to the structure ID satisfying the hierarchical structure search condition are searched from the storage means based on the hierarchical structure search condition in which the deletion step deletes the structure ID candidate. An index search step;
A retrieval step for retrieving the structured document corresponding to the document ID retrieved by the index retrieval step from a structured document storage means for storing the structured document in association with the document ID;
A structured document retrieval method characterized by executing :

Images