JP2008084192A - Structured document retrieval device, structured document retrieval method and structured document retrieval program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a structured document retrieval device adapted to appropriately acquire and output a retrieval result. <P>SOLUTION: The device comprises a receiving part 101 for receiving input of a retrieval keyword; a retrieval part 102 for retrieving a structured document containing the received retrieval keyword from a document management device; a calculation part 106 for calculating statistic information related to appearance of a structural element that is a unit of a logical structure contained in the retrieved structured document; a generation part 113 for generating a feature vector based on the calculated statistic information; a structure extraction part 107 for selecting a structural element characterizing the structured document based on the feature vector, and extracting a similar partial structure that is a similar constituting element between retrieved structured documents from the selected structural element; a document extraction part 108 for extracting a partial document within the structured document corresponding to the extracted similar partial structure; and an output part 111 for outputting the extracted partial document for each structural element. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a structured document search apparatus, a structured document search method, and a structured document search program for searching a structured document that matches a search keyword from a plurality of structured documents having different hierarchical structures.

  In recent years, structured documents have been widely used to facilitate the management and processing of document data on computers. For example, a hierarchically structured structured document can be defined such that the title of the document is surrounded by a tag “title” and the entire document is surrounded by a tag “document”.

  As conventions for describing such structured documents, standardization of structured document conventions such as SGML (Standard Generalized Mark-up Language) and XML (eXtensible Markup Language) has been promoted and widely used.

  In the retrieval of structured documents, it is possible to perform a retrieval specifying a structure as a retrieval range together with a keyword. For example, it is possible to specify a search condition for searching for documents in which the search keyword “XML” is included in the tag “headline”. Therefore, a structured document in which the search keyword “XML” is included only in the character string other than the tag “headline” can be excluded from the search target.

  In other words, in the conventional plain text document keyword search, all documents including keywords are detected, but in the structured document keyword search, only documents whose keywords are included in the specified element range are detected. Is possible.

  On the other hand, a technique for outputting a document search result as a structured document has also been proposed. For example, Patent Document 1 proposes a technique for detecting elements in a structured document including a designated search keyword and enumerating only the detected elements as search results. According to the method of Patent Document 1, only elements including a keyword are output as search results, and elements that do not include a keyword can be omitted. Further, according to the method of Patent Document 1, it is possible to extract related elements in one structured document as a search result, and to format and output the extracted elements as one element.

JP 2005-267344 A

  However, in the method of Patent Document 1, since the documents are shaped in units of documents, there is a problem that it may not be easy to compare when confirming search results output for a plurality of documents. This is a problem caused by not considering the relevance of elements between a plurality of documents as search results.

  When considering the relevance of elements between a plurality of documents, it is necessary to pay attention to the fact that there is a structured document that can take any tag name or structure, such as XML. For example, in XML, there may be documents that include the same elements but differ in the hierarchical relationship between elements, and documents that represent elements having the same semantic content but different element names. Even in such a case, it is desirable to retrieve elements that are related between the structured documents, and to format and output them so that they can be compared.

  The present invention has been made in view of the above, and is a structured document search apparatus and structure capable of appropriately acquiring and outputting a search result even for a structured document having an indefinite structure such as XML. An object of the present invention is to provide a structured document retrieval method and a structured document retrieval program.

  In order to solve the above-described problems and achieve the object, the present invention is a structured document retrieval apparatus connected to a document management apparatus that stores a structured document having a hierarchical logical structure via a network. Receiving means for receiving input of a search keyword; search means for searching a plurality of the structured documents including the received search keyword from the document management apparatus; and the plurality of structured documents included in the searched plurality of structured documents. A calculation unit that calculates statistical information regarding the appearance of a structural element that is a unit of a logical structure; a generation unit that generates feature vectors of the plurality of structured documents searched based on the calculated statistical information; The structural elements that characterize the plurality of the structured documents that have been retrieved are selected based on the feature vectors that have been retrieved. Structure extraction means for extracting similar partial structures that are similar structural elements between structured documents, document extraction means for extracting partial documents in the structured document corresponding to the extracted similar partial structures, and extraction Output means for outputting the partial document for each structural element.

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

  According to the present invention, it is possible to extract a similar structural element between a plurality of documents based on statistical information, and to search and output a partial document corresponding to the extracted structural element. For this reason, there is an effect that a search result can be appropriately acquired and output even for a structured document having an indefinite structure such as XML.

  Exemplary embodiments of a structured document search device, a structured document search method, and a structured document search program 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 calculates statistical information of the structural elements of the searched structured document, and obtains similar structural elements between a plurality of structured documents based on the calculated statistical information. Extracting, acquiring partial documents corresponding to the extracted structural elements, and outputting them in a list format.

  FIG. 1 is a block diagram showing a configuration of a structured document search apparatus 100 according to the first embodiment. The structured document search apparatus 100 is connected to a document management apparatus 200 that manages structured documents via a network 300.

  The network 300 connects the document management apparatus 200 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 document management apparatus 200 includes a structured document storage unit 221 that stores a structured document to be searched. Note that the document management apparatus 200 is not limited to one, and can search for structured documents stored in a plurality of document management apparatuses 200.

  The structured document storage unit 221 stores a plurality of structured documents. The structured document storage unit 221 can be configured by any commonly used storage medium such as an HDD (Hard Disk Drive), an optical disk, a memory card, and a RAM (Random Access Memory).

  Here, the structured document will be described. FIG. 2 is an explanatory diagram illustrating an example of a structured document. This figure shows an example of a structured document in which XML is described as a structured document convention. Note that the structured document is not limited to a document that uses XML as a structured document convention, and may be another structured document such as SGML or HTML.

  As shown in the figure, in XML, a “tag” is added to a character string representing actual document information. In the example shown in the figure, “<new score>”, “</ new score>”, “<artist>”, “<title>”, and the like represent tags. With this tag, the document data can be expressed in a hierarchical logical structure. In XML, a tag is expressed by surrounding a name representing the contents of the tag with a symbol “<” and a symbol “>”.

  A tag surrounded by the symbols “<” and “>” is called a start tag, and a tag surrounded by the symbols “</” and the symbol “>” is called an end tag. A unit in which a character string is enclosed by a start tag and an end tag is referred to as a structural element (hereinafter simply referred to as an element) that is one unit constituting a logical structure. A name representing the contents of the tag is called an element name. By further enclosing elements with tags, the document can be structured hierarchically. An element corresponding to the innermost start tag and end tag sandwiching the character string is referred to as a text element.

  In the example shown in the figure, a character string “Limited Time” is surrounded by “<Artist>” and “</ Artist>”. Thereby, the element name “artist” is given to this character string. Further, the artist element is surrounded by “<new score>” and “</ new score>” tags, and is structured as an element included in the new score element. Further, since the artist element is the innermost tag of the character string “Limited Time”, the artist element is a text element.

  FIG. 3 is an explanatory diagram showing an example of another structured document having a structure different from that in FIG. The structured document storage unit 221 can simultaneously store structured documents having different element names, structures, structure depths, or numbers of elements in addition to structured documents as shown in FIGS. . The structured document storage unit 221 can store structured documents in any storage format as long as the search unit 102 can search.

  The outline of the structured document storage format in the present embodiment will be described below. 4 and 5 are explanatory diagrams showing the logical structure of the structured document shown in FIGS. 2 and 3, respectively.

  As shown in FIGS. 4 and 5, each element is expressed as a tree-structured hierarchical structure. Each element is indicated by a circle, and a character string (text data) included in the text element is indicated by a rectangle. A character string in a circle represents an element name.

  For example, since “Artist” in FIG. 4 is an element, it is represented by a circle and an element name (artist) is displayed. In addition, “Limited Time” is text data in an artist element that is a text element, and thus is represented by a rectangle. Since “Limited Time” is text data included in the artist element, it is connected and represented by a line of vertical relations.

  On the right shoulder of each element and text element, an ID used in each process described later is displayed. The ID value represents an element ID and an element name ID in order from the left. The element ID means an identifier that uniquely identifies an element in the document. The element name ID means an identifier that uniquely identifies an element name in all structured documents.

  The artist element in FIG. 4 has an element ID of 2 and an element name ID of 2. The artist element in FIG. 5 is different from FIG. 4 because the element ID is 4, but the element name ID has the same value (2) because it has the same element name as the artist element in FIG.

  A small circular value on the right shoulder of the text data is a text data ID that uniquely identifies the text data within the document. In FIG. 4, for example, the text data ID of the text data “Limited Time” is 1.

  As described above, the structured document storage unit 221 stores the structured document by giving an ID for identifying each element to each element of the tree structure of the structured document.

  Returning to FIG. 1, the structured document search apparatus 100 includes a statistical information storage unit 122, a reception unit 101, a search unit 102, a first determination unit 103, a determination unit 104, a change unit 105, and a calculation unit 106. A generation unit 113, a structure extraction unit 107, a document extraction unit 108, a second determination unit 109, an addition unit 110, and an output unit 111.

  The statistical information storage unit 122 stores statistical information regarding each element included in the structured document. Specifically, the statistical information storage unit 122 stores, as statistical information, the appearance frequency of each element included in the structured document in the structured document stored in the document management apparatus 200.

  FIG. 6 is an explanatory diagram illustrating an example of a data structure of statistical information stored in the statistical information storage unit 122. As shown in the figure, the statistical information stores the element name ID of each element and the appearance frequency in the structured document stored in the document management apparatus 200 in association with each other.

  The statistical information stored in the statistical information storage unit 122 is referred to when the calculation unit 106, which will be described later, calculates statistical information of each element of the searched structured document.

  The accepting unit 101 accepts a search keyword input by a user. The receiving unit 101 may be configured to receive a search keyword input from an input device such as a keyboard, or may be configured to receive a search keyword input from an external device via the network 300.

  The search unit 102 searches for a structured document including the search keyword from the structured document storage unit 221 in the document management apparatus 200 using the search keyword received by the receiving unit 101 as a search condition. The search unit 102 can detect not only the case where the search keyword is included in the text data of the text element but also the case where the keyword is included in the element name.

  Note that the search unit 102 may be configured to search only text data. The search unit 102 may be configured to search a structured document including a keyword in which the meaning of the search keyword is expanded with reference to a synonym dictionary or the like (not shown). For example, it can be realized by making it possible to specify that the search is to be performed by expanding the meaning of the search keyword by setting the user at the time of search or by setting in advance.

  FIG. 7 is an explanatory diagram showing an example of a search result by the search unit 102. This figure shows an example of a search result when “Limited Time” and “CD” are input as search keywords.

  For example, in the structured document of FIG. 2, the first keyword “Limited Time” is included in the artist element, and the second keyword “CD” is included in the media element. Since the document includes all the search keywords, the document in FIG. 2 is searched as a conforming document that matches the search keyword.

  In the structured document of FIG. 3, the first keyword is included in the artist element and the second keyword is included multiple times as the element name included in the genre element. Since all the search keywords are included in the document, the document in FIG. 3 is also searched as a conforming document.

  In FIG. 7, for each search keyword, a search that associates a document ID that uniquely identifies a conforming structured document, a text element flag that indicates whether or not the text element is included, an element ID, and an element name ID. An example of the result is shown. The text element flag is set to T (True) when the element including the search keyword is a text element, and set to F (False) otherwise. This figure shows an example in which the document ID of the structured document in FIG. 2 is 1 and the document ID of the structured document in FIG. 3 is 2.

  As described above, the search unit 102 creates information for identifying the document, the detected element, and the element name of the document in which all keywords are detected, and passes the information to the next process as a search result. At this time, when a text element is detected, the text data ID of the text data included in the detected text element may be passed.

  In FIG. 7, information on text elements in which the first keyword “Limited Time” is detected is listed one by one in each document, and the text element flag is “T”. The element in which the second keyword “CD” is detected is a text element when the document ID is 1, and “T” is similarly set in the text element flag. When the document ID is 2, since elements with element IDs 3 and 8 are detected, two are listed. In addition, since the elements with element ID = 3 and 8 are not text elements, both “F” are set in the text element flag.

  Returning to FIG. 1, the first determination unit 103 determines whether there is an element name having a synonym relationship between structured documents for each element included in the searched structured document. . Specifically, the first determination unit 103 refers to a synonym dictionary (not shown) and determines whether there is an element name having a synonym relationship in each structured document. Note that the element name determined to be a synonym is changed by the changing unit 105 described later so as to be unified with any element name.

The determination unit 104 determines a representative element name that is the element name of the element having the highest degree of matching that represents the degree of matching with the search keyword from the text elements including the search keyword in the text data. Specifically, the determination unit 104 determines an element name of an element that satisfies the following conditions as a representative element name.
Condition 1: Data length / keyword length (= N) is minimum (N> = 1)
Condition 2: Appearance frequency of search keyword is maximum

  The representative element name determined in this way is referred to by the changing unit 105 described later when the element name of an element that includes the search keyword in the element name is changed to the representative element name.

  The changing unit 105 performs element name normalization processing that unifies similar element names for the element names that the first determining unit 103 determines to have a synonym relationship. Specifically, the changing unit 105 changes one element name to the other element name having a synonym relation.

  Further, the changing unit 105 performs a matching element name changing process for changing the element name of an element including the search keyword in the element name. Specifically, the changing unit 105 first changes the element name of the element including the search keyword in the element name to the representative element name determined by the determining unit 104 for the search keyword. Then, the changing unit 105 adds the element name before the change as text data corresponding to the representative element name, and changes the element of the representative element name to a text element.

  The calculation unit 106 calculates statistical information of each element of the searched structured document. Specifically, the calculation unit 106 acquires the corresponding appearance frequency for each element in the searched structured document from the statistical information storage unit 122 and sets it as statistical information. When normalization processing is performed, the calculation unit 106 calculates the statistical information by adding the appearance frequency of the element whose element name has been changed to the appearance frequency of the element name to be changed. Further, when the changing unit 105 executes the matching element name changing process, the calculating unit 106 adds the appearance frequency of the element whose element name has been changed to the representative element name to the appearance frequency of the representative element name to obtain statistical information. calculate.

  The generating unit 113 refers to the statistical information calculated by the calculating unit 106, and further generates a feature vector for each document, taking into account the element that has detected the search keyword. A feature vector is information representing elements that characterize a structured document. In the present embodiment, the generation unit 113 generates a feature vector by extracting an element whose appearance frequency is higher than a predetermined threshold value (threshold value regarding the appearance frequency) and adding an element in which a search keyword is detected. Details of the feature vector will be described later.

  The structure extraction unit 107 extracts similar elements (similar partial structures) between the searched structured documents from the generated feature vector. That is, the structure extraction unit 107 calculates the similarity between feature vectors of a plurality of searched structured documents using a vector space method or the like. Specifically, assuming that the two feature vectors are A and B, respectively, the structure extraction unit 107 calculates the similarity by A × B / | A || B |. A × B represents the inner product of the vector A and the vector B. Note that the method of calculating the similarity is not limited to this, and any method can be applied as long as the similarity between the structures of the two structured documents is calculated.

  Then, the structure extraction unit 107 extracts each element included in the structured document whose calculated similarity is larger than a predetermined threshold (threshold for similarity) as a similar partial structure.

  The document extraction unit 108 performs processing for extracting and classifying a matching partial document from a structured document as a search result for each similar partial structure. Specifically, the document extraction unit 108 acquires a structured document including the extracted similar partial structure, traces the structure from the element to which the search keyword matches the acquired structured document, and includes it in the similar partial structure Detected element as a partial document.

  The second determination unit 109 determines whether there is a search keyword that is not included in the extracted partial document. Specifically, the second determination unit 109 determines whether there is an element that is not included in the partial document extracted by the document extraction unit 108 among the elements that match the search keyword.

  The adding unit 110 performs a process of integrating search result elements with respect to partial documents classified by being extracted for each similar partial structure, and adjusting the size of the partial documents. Specifically, the adding unit 110 adds a partial document corresponding to the element when the second determining unit 109 determines that there is an element that is matched with the search keyword and is not included in the partial document. To do. When the size of the extracted partial document is larger than a predetermined threshold value (threshold value regarding the partial document size), the size of the partial document is reduced by deleting elements that do not include the search keyword.

  The output unit 111 outputs the integrated and size-adjusted partial documents in an optimum format for each classification based on similar partial structures. Specifically, the output unit 111 outputs text data of elements corresponding to similar partial structures in a tabular format.

  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. 8 is a flowchart showing an overall flow of the structured document search process according to the first embodiment.

  First, the reception unit 101 receives a search keyword input by a user (step S801). Next, the search unit 102 executes a search process using the received search keyword (step S802). Specifically, the search unit 102 searches the structured document storage unit 221 of the document management apparatus 200 for a structured document including the accepted search keyword.

  Next, a structural statistical information calculation process for calculating statistical information of each element is performed for each of the searched structured documents (step S803). Details of the structure statistical information calculation process will be described later.

  Next, the structure extraction unit 107 executes a similar partial structure extraction process for extracting a similar partial structure from the searched structured document (step S804). Details of the similar partial structure extraction processing will be described later.

  Next, the document extraction unit 108 executes a partial document extraction process for extracting a partial document corresponding to the similar partial structure (step S805). Details of the partial document extraction processing will be described later.

  Next, partial document integration processing for integrating necessary partial documents and adjusting the size is executed (step S806). Details of the partial document integration processing will be described later.

  Finally, the output unit 111 displays the integrated partial document (step S807) and ends the structured document search process.

  Next, details of the structural statistical information calculation processing in step S803 will be described. FIG. 9 is a flowchart showing an overall flow of the structural statistical information calculation processing in the first embodiment.

  First, the first determination unit 103 acquires a structured document including an element that matches the search keyword (step S901). Next, an element name normalization process is executed. In the normalization process, first, the first determination unit 103 determines whether there is an element having a synonym relationship between the plurality of acquired structured documents (step S902). At this time, the first determination unit 103 refers to a synonym dictionary (not shown) or the like to detect other element names that have a synonym or synonym relationship with a certain element name.

  For example, the first determination unit 103 determines that the element name of the “title” element of the structured document of FIG. 2 and the element name of the “title” element of the structured document of FIG. .

  If there is an element having a synonym relationship (step S902: YES), the changing unit 105 performs normalization to change one element name to the other element name (step S903). In the above example, the changing unit 105 changes, for example, the element name “title” of the “title” element to “title”. At the same time, the changing unit 105 changes the element name ID of the “title” element from 11 to 3. That is, in FIG. 5, the element ID of the “title” element is 6 and the element name ID is 11. However, after the change, the element ID is 6 and the element name ID is 3.

  The element name to be unified can be determined by any method such as a method of changing to an element name having a high appearance frequency. The element name normalization process may be omitted.

  When there is no element having a synonym relationship (step S902: NO), or after executing the normalization process (step S903), the matching element name change process is executed. As a previous stage of the matching element name changing process, a representative element name determining process by the determining unit 104 is executed. In the representative element name determination process, first, the determination unit 104 determines whether or not the search keyword is searched as an element name (step S904).

  If a search is made (step S904: YES), the determination unit 104 determines a representative element name for each search keyword (step S905). Specifically, the determination unit 104 sets an element name of an element having a condition that best meets the above-described condition 1 and condition 2 as a representative element name.

  For example, when the search keyword “Limited Time” is input, for the structured document (document ID = 1) in FIG. 2, the text data completely matches the search keyword, so the data length / keyword length = 1. The minimum. If the frequency of “Limited Time” appears most frequently in the “Artist” element as in the document with document ID = 1, the representative element name is determined to be “Artist”.

  Similarly, when the search keyword “CD” is input, since the data length / keyword length = 1 in the document with the document ID = 1, “CD” appears most frequently in the “media” element. , “Media” is determined as the representative element name.

  Next, the changing unit 105 executes a matching element name changing process. Specifically, the changing unit 105 changes the element name to the representative element name determined for the search keyword for each element whose element name matches the search keyword. Further, an element name that matches the search keyword is added to the element as text data (step S906).

  For example, when the search keyword “CD” is input, in the structured document of FIG. 3, the search keyword is included in the element name of the “CD” element, so that the element is searched as a matching element. FIG. 10 is an explanatory diagram showing an example of a result of executing the matching element name changing process for this example. As shown in FIG. 5, in FIG. 5, the two “CD” elements that were child elements of the genre become “media” elements, and the element name ID is also changed to 5. Further, text data including a character string “CD” is added as a child element of each changed “media” element. As shown in the figure, the added text data is the element name “CD” before the change.

  Note that, in the case where the keyword search is not performed on the element name, the compatible element name changing process may be omitted.

  Returning to FIG. 9, when it is determined in step S904 that the search keyword has not been searched as an element name (step S904: NO), or after execution of the matching element name changing process in step S906, the calculation unit 106 displays the element name. The statistics of the appearance frequency for each are calculated (step S907).

  Specifically, the calculation unit 106 acquires the appearance frequency of the element name ID included in the document that matches the search keyword from the statistical information storage unit 122, and for the element that has undergone the normalization process or the matching element name change process The appearance frequency statistics are calculated by adding the appearance frequency of the element before the change to the appearance frequency of the element after the change.

  FIG. 11 is an explanatory diagram illustrating an example of calculated statistical information. As shown in the figure, the calculation unit 106 outputs an element name ID frequency table in which the calculated frequencies are represented in a table format. This figure shows an example calculated based on the frequency acquired from the statistical information storage unit 122 shown in FIG. Further, for example, the “title” element (element name ID = 11) of the structured document shown in FIG. 5 has been changed to element name ID = 3 by the normalization process, so the frequency of element name ID = 11 ( 10) is added to the frequency (39) of the element name ID = 3, and the value (49) is set in the element name ID frequency table. At the same time, the element with the element name ID = 11 is deleted from the element name ID frequency table.

  Similarly, the “CD” element (element name ID = 10) of the structured document shown in FIG. 5 has been changed to element name ID = 5 by the matching element name change process, and therefore the frequency of element name ID = 10. A value (45) obtained by adding (12) to the frequency (33) of the element name ID = 5 is set in the element name ID frequency table. At the same time, the element with the element name ID = 10 is deleted from the element name ID frequency table.

  If other element name IDs exist in other documents suitable for the search, they are also listed and the appearance frequency is calculated. In the high frequency column, the rank is set for the element name ID having the higher frequency number. Furthermore, in the keyword hit column, 1 is set for the element name ID of the element that matches the search keyword, and 0 is set for the other element name IDs. The calculation unit 106 passes the element name ID frequency table created in this way to the next processing.

  Next, details of the similar partial structure extraction processing in step S804 will be described. FIG. 12 is a flowchart showing the overall flow of the similar partial structure extraction process in the first embodiment.

  First, the structure extraction unit 107 acquires a structured document including matching elements of search keywords (step S1201). Next, the generation unit 113 generates a feature vector of an element having a high appearance frequency for each acquired structured document (step S1202).

  FIG. 13 is an explanatory diagram illustrating an example of the generated feature vector. In the figure, element names of elements constituting the structured document are listed, and an example of a feature vector in which the presence / absence of these elements is expressed by 1 (with elements) and 0 (without elements) is shown for each document. The feature vector includes a predetermined number of high-frequency elements from the element name ID frequency table passed from the calculation unit 106 and elements that match the search keyword.

  In the example of FIG. 13, from the element name ID frequency table as shown in FIG. 11, the “artist” element with element name ID = 2, the “title” element with element name ID = 3, and the element name ID as the high frequency elements. = 3 “Date” elements are selected, and a “Media” element is selected as an element to which the keyword is matched. Although the addition of an element that matches the search keyword can be omitted, more appropriate search results can be acquired and output.

  The column of document ID = 1 represents a feature vector obtained by extracting the presence or absence of an element corresponding to the element name ID selected from FIG. 11 for the structure of the document of FIG. Similarly, feature vectors are extracted for the structures of the other documents (document ID = 2, 3,... In FIG. 13).

  Returning to FIG. 12, the structure extraction unit 107 calculates the similarity between structured documents from the feature vector (step S1203). Specifically, as described above, assuming that the feature vectors for the two structured documents are A and B, respectively, the structure extraction unit 107 calculates the similarity by A × B / | A || B | .

  The lower part of the table of FIG. 13 shows the result of calculating the similarity between each document calculated from the feature vector. Here, the similarity is calculated using a feature vector in which 1 is present when there is an element and -1 when there is no element.

  It can be determined that structured documents having a high degree of similarity are similar to each other. In the example shown in the figure, it is determined that the document IDs = 1, 2, and 3 are similar to each other.

  Next, the structure extraction unit 107 extracts the sum of elements included in the structured document with a high degree of similarity as a similar partial structure (step S1204). In the example of FIG. 13, all elements shown in FIG. 13 are selected as similar partial structures as the sum of the elements of document ID = 1, 2, and 3.

  FIG. 14 is an explanatory diagram showing an example of the extracted similar partial structure. This figure shows similar partial structures extracted from document ID = 1, 2, 3 in FIG. In the figure, an element with an element name ID = 0 is a parent element, and elements with an element name ID = 2 to 5 (an “artist” element, a “title” element, a “date” element, and a “media” element) are child elements. An example in which the included structure is extracted as a similar partial structure is shown.

  At this time, it is desirable that the similar partial structure includes an element in which the search keyword is detected. In the example of FIG. 13, since the search keyword is detected in the “artist” element and the “media” element with the document ID = 1, an element in which the search keyword is detected is included in the similar partial structure.

  If the element in which the search keyword is detected is not included, one or more elements in which the search keyword is detected are included. For example, a structured document having the next highest similarity is examined, and if an element in which a search keyword is detected is included, the element is added to the similar partial structure. In addition, when the element included in the structured document including the added element and the element included in the structured document having the highest similarity with the structured document are added, and a keyword is detected in the sum An element including a search keyword can be added by any method such as a method of adding the element.

  In some cases, after extracting a similar partial structure from a feature vector and determining a document corresponding to the similar partial structure, a similar partial structure can be further extracted for other documents not corresponding to the similar partial structure. FIG. 15 is an explanatory diagram illustrating another example of the generated feature vector. When a feature vector as shown in FIG. 15 is generated, it can be seen that the similarity between the document IDs = 1, 2, and 3 is high, so the sum of the elements of these structured documents is set as a similar partial structure. Since it can be determined that the remaining structured documents, document ID = 4, 5 are also similar to each other, the sum of the elements of document ID = 4, 5 can be extracted as another similar partial structure.

  Next, details of the partial document extraction processing in step S805 will be described. FIG. 16 is a flowchart showing the overall flow of partial document extraction processing.

  First, the document extraction unit 108 acquires a document that conforms to a similar partial structure (step S1601). For example, when a similar partial structure as shown in FIG. 14 is extracted, structured documents with document IDs = 1, 2, and 3 that are the basis of extraction are acquired.

  Next, the document extraction unit 108 traces the hierarchical structure of the obtained structured document using the element in which the search keyword is detected (hereinafter referred to as hit element) as a base point, and detects an element corresponding to the similar partial structure. Extracted as a partial document (step S1602).

  The partial document is extracted from an area close to the element in which the search keyword is detected. For example, when the element in which the search keyword is detected is a text element, an element close to its parent element, that is, its sibling element is confirmed. When the search keyword is detected in the element name, the structure below the element is examined first, and if necessary, the partial document is extracted by expanding to the parent, sibling, and ancestor elements. Note that all elements included in the similar partial structure may not be included in the partial document, but in this case, the elements not included are ignored and the same processing is performed.

  A specific example of partial document extraction will be described below. For example, in the case of the structured document of FIG. 4, the similar partial structure includes both text elements in which the search keyword is detected. First, sibling elements of the element in which the search keyword is detected are examined as elements that are based on these points. Next, the investigation range is expanded as necessary, such as a parent element, a structure below the parent element of the parent element, and an ancestor element. In this example, an element included in the similar partial structure is detected as a sibling element of the text element in which the search keyword is detected. The document extraction unit 108 extracts a part below the parent element of the detected element as a partial document that matches the similar partial structure.

  In the structured document of FIG. 5, the similar partial structure includes one text element in which the search keyword “Limited Time” is detected from the text data, and a text element in which the keyword “CD” is detected from the element name. One included. Elements included in the similar partial structure are detected as sibling elements of the text element in which the search keyword is detected from the text data. Further, since the parent element of the element is an element itself (“CD” element) in which another search keyword (“CD”) is detected in the element name, the document extraction unit 108 includes a sub-document below the “CD” element. Extract as

  In addition, when a search is performed using a plurality of search keywords, if all the structures in which the search keyword is detected are not included in the similar partial structure, for example, only one may be included. FIG. 17 is an explanatory diagram showing another example of the logical structure of a structured document in which such a case may occur.

  For example, in the case of a structured document having a logical structure as shown in FIG. 17, when a partial document is extracted from a similar partial structure, a partial document having an element ID = 5 or less is extracted. This is a result of extraction based on the “media” element, which is an element in which the search keyword “CD” included in the similar partial structure is detected.

  In the case of this structured document, another search keyword “Limited Time” is included in the “title” element with the element ID = 2, but when a structure including the element is extracted, “ All elements below the “doc” element will be included. Therefore, in such a case, not the partial document including all the search keywords but the partial document including the element with element ID = 5 that is a partial document including at least one search keyword is extracted.

  In the case of a structured document having a logical structure as shown in FIG. 17, the “media” element included in the similar partial structure is an element that matches the search keyword, and is detected multiple times in the same structured document. It is a possible element. In such a case, a process of extracting a partial document is performed using each detected element as a base point. That is, the “music” element with element ID = 10 extracted from the element with element ID = 13 as a base point is also extracted as a partial document.

  Returning to FIG. 16, the document extraction unit 108 determines whether or not the size of the extracted partial document is smaller than a predetermined threshold (threshold regarding the partial document size) (step S1603). If it is larger than the threshold (step S1603: NO), the partial document extraction process is terminated to perform a process of reducing the size of the partial document in the partial document integration process described later.

  If it is smaller than the threshold (step S1603: YES), the document extraction unit 108 determines whether or not a plurality of partial documents have been extracted (step S1604). This is because, as described above, a plurality of partial documents corresponding to the similar partial structure may be extracted, and therefore it is necessary to determine which partial document is to be adopted by the following processing.

  When a plurality of partial documents are extracted (step S1604: YES), the document extraction unit 108 calculates the number of elements that match the search keyword for each partial document (step S1605). Then, the partial document having the largest number of matching elements is selected (step S1606). If a plurality of partial documents are not extracted (step S1604: NO), the partial document extraction process is terminated.

  In the partial document selection process, the selection may be made in consideration of not only the number of elements but also the size of the partial document. That is, priorities are added to partial documents according to the following criteria, and the partial document with the highest priority is selected.

  The partial document with the highest priority is the partial document that includes the text elements that contain all the search keywords, includes the largest number of elements included in the similar partial structure, and has the lowest depth of the partial document hierarchy. is there. If there are multiple equivalent partial documents at this stage, all of them are extracted. If a partial document can be extracted at this stage, other partial document candidates for the similar partial structure of the structured document are not extracted.

  The partial document having the next highest priority is a partial document including the text element when a plurality of search keywords are matched with different text elements or element names and the text element is included in the similar partial structure.

  The next highest priority partial document has multiple search keywords that match with different text elements or element names, contains more elements that match similar substructures, and contains more matched elements It is a partial document.

  Further, if the partial document is not extracted up to this stage, the similar partial structure and the element including the search keyword are all extracted as a partial document.

  For all similar partial structures, after extracting a partial document from a structured document that conforms to a search that has a structure close to that similar partial structure, if there is a compatible document that does not extract one partial document, the relevant document A partial document including only elements that match the search keyword is extracted from the document.

  FIG. 18 is an explanatory diagram showing another example of the logical structure of the structured document. This figure shows an example of the logical structure of a structured document having a completely different logical structure from the structured documents having a similar logical structure described so far (FIGS. 4, 5, and 17).

  In the case of a structured document as shown in FIG. 18, since a partial document corresponding to a similar partial structure cannot be extracted, the document extraction unit 108 extracts a “content” element that is an element suitable for the search keyword as a partial document.

  Next, details of the partial document integration processing in step S806 will be described. FIG. 19 is a flowchart showing the overall flow of partial document integration processing.

  First, the second determination unit 109 determines, for each extracted partial document, whether or not there is a search keyword (hereinafter referred to as a deficient keyword) that is not included in the partial document among the search keywords (step S1901). . Such a situation may occur when a plurality of search keywords are specified.

  When the missing keyword exists (step S1901: YES), the adding unit 110 detects an element including the missing keyword (step S1902). Next, the adding unit 110 integrates the detected elements into the partial document (step S1903).

  For example, in the case of the structured document as shown in FIG. 17, the search keyword “Limited Time” is not included in the document below the element with element ID = 5 extracted as the partial document. In such a case, the “title” element that is an element including the search keyword “Limited Time” is combined with the extracted partial document. In the example of FIG. 17, partial documents below the element with element ID = 10 are also extracted. If there is a missing keyword also for this partial document, processing for adding an element including the missing keyword is performed.

  Further, when there are a plurality of elements including the missing keyword, all of them may be selected and added. When there are a large number of elements to be added, priority may be given using the proximity of the extracted partial document, the character string length, etc., and elements with higher priority may be added.

  If there is no missing keyword (step S1901: NO), or after elements are integrated (step S1903), the second determination unit 109 determines whether the partial document size is larger than a predetermined threshold (threshold for the partial document size). It is determined whether or not (step S1904).

  If larger than the threshold (step S1904: YES), the adding unit 110 reduces the size of the partial document (step S1905). As a technique for reducing the document size, a method of further deleting unnecessary elements from the extracted partial document is used.

  For example, let us consider a case where the elements below the “new music” element are extracted as partial documents for the structured document as shown in FIG. If the partial document has a size equal to or larger than a predetermined threshold, the size is reduced by omitting elements that are not included in the similar partial structure. For example, the partial document size is reduced by deleting the “price” element and the “label” element. In addition, when the text data size of the text element is large, it is configured to reduce the size by extracting a partial character string of a certain size containing many search keywords from the text data and omitting the others. May be.

  Below, the output method of the partial document extracted as mentioned above is demonstrated. FIG. 20 is an explanatory diagram showing an example of the output format of the partial document output by the output unit 111. In this example, partial documents including “artist”, “media”, “title”, and “date” elements are displayed in a tabular format. Further, conforming documents as shown in FIG. 18 that do not conform to the similar partial structure are displayed in an enumerated form below the table.

  Further, in the example of displaying the result in a table format, elements other than the elements included in the similar partial structure among the elements included in the partial document are displayed in the other column (EX column). The display format is not limited to the enumerated type and the table format, and any other display format such as a definition type or an XML format can be used.

  As described above, the structured document search apparatus according to the first embodiment extracts and extracts similar elements between a plurality of structured documents based on the statistical information calculated for the elements of the searched structured document. The partial document corresponding to the selected element can be acquired and output in a list format. Therefore, it is possible to appropriately acquire and output a search result even for a structured document with an undefined structure such as XML. As a result, the cost for confirming the search result by the user can be reduced.

  Traditionally, as part of a proposal for a large-volume document management method, technologies that automatically extract similar documents from a large number of documents, clustering methods that automatically classify documents according to the contents of multiple documents, etc. Various techniques have been proposed. However, these technologies classify documents semantically based on the frequency of words, the frequency of their co-occurrence, document attributes, and user judgment. The extraction of the optimum partial document in consideration cannot be realized.

  On the other hand, according to the present embodiment, even if it is a search result for a structured document whose structure cannot be specified, an appropriate partial document can be extracted, classified and output according to the characteristics of the document structure. .

  Furthermore, even if the user does not have knowledge about the logical structure of the structured document and does not execute a search specifying a condition related to the structure, a search result utilizing the document structure can be acquired.

(Second Embodiment)
The structured document search apparatus according to the second embodiment registers a structured document to be searched in advance in the own apparatus and executes a search process on the registered structured document. Further, at that time, by executing in advance a process that can be executed before the search process, the load of the search process is reduced and the search speed is increased.

  FIG. 21 is a block diagram illustrating a configuration of a structured document search apparatus 2100 according to the second embodiment. As shown in the figure, the structured document search apparatus 2100 includes a statistical information storage unit 122, a structured document storage unit 2121, a reception unit 101, a search unit 2102, a first determination unit 2103, and a determination unit 104. A change unit 2105, a calculation unit 106, a generation unit 113, a structure extraction unit 107, a document extraction unit 108, a second determination unit 109, an addition unit 110, an output unit 111, and a registration unit 2112. It is equipped with.

  The second embodiment is different from the first embodiment in that a structured document storage unit 2121 and a registration unit 2112 are added. The functions of the search unit 2102, the first determination unit 2103, and the change unit 2105 are different from those of the first embodiment. Other configurations and functions are the same as those in FIG. 1 which is a block diagram showing the configuration of the structured document search apparatus 100 according to the first embodiment, and therefore, the same reference numerals are given and description thereof is omitted here. .

  Similar to the structured document storage unit 221, the structured document storage unit 2121 stores a structured document to be searched.

  The search unit 2102 is different from the search unit 102 of the first embodiment in that the structured document storage unit 2121 searches for a structured document.

  The first determination unit 2103 stores in the structured document storage unit 2121 or for each element included in the stored structured document, whether or not there is an element name having a synonym relationship between the structured documents. Is to judge.

  The change unit 2105 performs element name normalization processing for unifying similar element names on the element names determined by the first determination unit 2103 to have a synonym relationship, and stores the changed element names in the structured document storage. This is stored in the unit 2121.

  The registration unit 2112 acquires a structured document from the structured document storage unit 221 of the document management apparatus 200 and registers it in the structured document storage unit 2121 in the structured document search apparatus 2100 before the search.

  As described above, in the second embodiment, the structured document can be stored in the own apparatus and the normalization process can be executed in advance. Therefore, it is not necessary to execute the normalization process in the search process. High speed can be realized.

  Next, a structured document registration process performed by the structured document search apparatus 2100 according to the second embodiment configured as described above will be described. FIG. 22 is a flowchart showing an overall flow of the structured document registration process in the second embodiment.

  First, the registration unit 2112 acquires a structured document from the document management apparatus 200 (step S2201). Next, the first determination unit 2103 determines whether there is an element having a synonym relationship between the acquired structured documents or between the acquired structured document and the already stored structured document. Is determined (step S2202).

  If there is an element having a synonym relationship (step S2202: YES), the changing unit 2105 executes an element name normalization process (step S2203). The normalization process is the same process as step S903 in the first embodiment.

  When there is no element having a synonym relationship (step S2202: NO), or after executing the normalization process, the registration unit 2112 stores the structured document in the structured document storage unit 2121 (step S2204).

  Next, the registration unit 2112 updates the statistical information of the structured document (step S2205). Specifically, the registration unit 2112 calculates the frequency of each element included in the stored structured document, and updates the frequency information in the statistical information storage unit 122.

  As described above, in the second embodiment, normalization processing is executed when a structured document is registered. Note that the process that can be executed at the time of registration is not limited to the normalization process, and can also include a feature vector generation process.

  Next, a structured document search process performed by the structured document search apparatus 2100 according to the second embodiment configured as described above will be described. The overall flow of the structured document search process in the second embodiment is the same as that in FIG. 8 of the first embodiment. However, the details of the structure statistical information calculation processing in step S803 are different.

  Details of the structural statistical information calculation process in the second embodiment will be described below. FIG. 23 is a flowchart illustrating an overall flow of the structural statistical information calculation process according to the second embodiment.

  This figure is a process in which the normalization process in steps S902 and S903 is deleted from FIG. 9 showing the overall flow of the structural statistical information calculation process in the first embodiment. That is, in the second embodiment, since normalization processing is executed at the time of registration, there is no need to execute normalization processing at the time of search.

  Steps S2301 and S2302 to S2305, which are the other processes, are the same as steps S901 and S904 to S907, respectively, and thus description thereof is omitted.

  As described above, in the structured document search apparatus according to the second embodiment, the search process is executed on the structured document registered in the self apparatus in advance, and the process that can be executed before the search process is executed in advance. As a result, the search processing load can be reduced and the search speed can be increased.

(Third embodiment)
A large number of structured documents often have a structure format determined when they are created. For example, a format of a structure in which an element “member” exists as a child element of the root element “project” and at least one element “member name” exists as a child element can be determined.

  In this case, a structured document having two “member name” elements may be created depending on a project, or a structured document having five “member name” elements may be created. As described above, since the degree of freedom of the format of the structure is high, even a structure created for a similar purpose may differ from a structure created in another place. For example, even if a document is used for the same purpose at a city hall, the format may be different if the city in which the document was created is different.

  On the other hand, the similar partial structure extraction process in the first embodiment is executed for each structured document suitable for the search. However, when the structured document to be searched is a plurality of structured document sets according to the respective structural formats, the similar partial structure investigation process is performed between the structural formats of the respective structured documents conforming to the search. For example, the processing cost can be reduced. This is because if the structure format common to each structured document is set as the verification target, the verification target is limited as compared with the case of examining each structured document.

  The structured document search apparatus according to the third embodiment simplifies the similar partial structure extraction process with reference to the structure information regarding the structure format of the structured document registered in advance.

  FIG. 24 is a block diagram illustrating a configuration of a structured document search apparatus 2400 according to the third embodiment. As shown in the figure, the structured document search device 2400 includes a statistical information storage unit 122, a structure information storage unit 2413, a reception unit 101, a search unit 102, a first determination unit 103, and a determination unit 104. A change unit 105, a calculation unit 106, a generation unit 113, a structure extraction unit 2407, a document extraction unit 108, a second determination unit 109, an addition unit 110, and an output unit 111.

  In the third embodiment, the addition of the structure information storage unit 2413 and the function of the structure extraction unit 2407 are different from those of the first embodiment. Other configurations and functions are the same as those in FIG. 1 which is a block diagram showing the configuration of the structured document search apparatus 100 according to the first embodiment, and therefore, the same reference numerals are given and description thereof is omitted here. .

  The structure information storage unit 2413 stores structure information related to the logical structure extracted from the structured document. FIG. 25 is an explanatory diagram showing an example of the data structure of the structure information stored in the structure information storage unit 2413. As shown in the drawing, the structure information stores an element name ID, an element name, a child element, and a sibling element in association with each other. The figure shows an example in which the structure information extracted from the structured document shown in FIG. 2 is stored. The structure information storage unit 2413 stores such structure information for each structure having a different format.

  Next, a structured document search process performed by the structured document search apparatus 2400 according to the third embodiment configured as described above will be described. The overall flow of the structured document search process in the third embodiment is the same as that in FIG. 8 of the first embodiment. However, the details of the similar partial structure extraction processing in step S804 are different.

  Details of the similar partial structure extraction process in the third embodiment will be described below. FIG. 26 is a flowchart illustrating an overall flow of the similar partial structure extraction process according to the third embodiment.

  First, the structure extraction unit 2407 obtains a structured document including the search keyword matching elements (step S2601). Next, the structure extraction unit 2407 acquires structure information corresponding to the acquired structured document from the structure information storage unit 2413 (step S2602).

  Next, the production | generation part 113 produces | generates the feature vector of an element with high appearance frequency for every acquired structural information (step S2603). The generation of the feature vector is the same as that in step S1202 of FIG.

  Next, the structure extraction unit 2407 calculates the similarity between the structure information from the feature vector (step S2604). The method for calculating the similarity is the same as that in step S1203 in FIG. Next, the structure extraction unit 2407 extracts the sum of elements included in the structure information having a high degree of similarity as a similar partial structure (step S2605).

  If extraction of similar partial structures between certain pieces of structure information is executed in this way, extraction processing of similar partial structures can be omitted when structured documents as other search results have the same structure information. . This makes it possible to increase the speed of search processing.

  Even if the structure information is obtained in advance and cannot be stored in the structure information storage unit 2413, the document can be stored if the structured document is registered in the own apparatus as in the second embodiment. When registering, the structure information can be analyzed to extract the format and stored in the structure information storage unit 2413. In the format extraction process, for example, when the element name sets included in the structure are common and the parent-child relationship between the element name and the element name is the same, a method such as the same format can be applied. If comprised in this way, even if it is a case where structure information cannot be obtained beforehand, the effect similar to the above can be acquired.

  As described above, the structured document search apparatus according to the third embodiment refers to the structure information related to the format of the structure of the structured document registered in advance, and extracts similar partial structures between the structure information. As compared with the case where the similar partial structure extraction process is executed between all structured documents, the process can be simplified, and the search process can be speeded up as a result.

  FIG. 27 is an explanatory diagram of a hardware configuration of the structured document search device according to the first to third embodiments.

  The structured document search device according to the first to third embodiments is connected to a control device such as a CPU (Central Processing Unit) 51, a storage device such as a ROM (Read Only Memory) 52 and a RAM 53, and a network. A communication I / F 54, an external storage device such as an HDD or a CD (Compact Disc) drive device, a display device such as a display device, an input device such as a keyboard or a mouse, and a bus 61 that connects each unit. It has a hardware configuration using a normal computer.

  The structured document search program executed by the structured document search apparatus according to the first to third embodiments is an installable format or executable file, a CD-ROM (Compact Disk Read Only Memory), The program is recorded on a computer-readable recording medium such as a flexible disk (FD), a CD-R (Compact Disk Recordable), and a DVD (Digital Versatile Disk).

  In addition, the structured document search program executed by the structured document search apparatus according to the first to third embodiments is stored on a computer connected to a network such as the Internet and downloaded via the network. It may be configured to provide. The structured document search program executed by the structured document search device according to the first to third embodiments may be provided or distributed via a network such as the Internet.

  Further, the structured document search program according to the first to third embodiments may be provided by being incorporated in advance in a ROM or the like.

  The structured document search program executed by the structured document search apparatus according to the first to third embodiments includes the above-described units (reception unit, search unit, first determination unit, determination unit, change unit, calculation unit). , A structure extraction unit, a document extraction unit, a second determination unit, an addition unit, an output unit, and a registration unit), and the actual hardware includes a CPU 51 (processor) from the storage medium as a structured document. By reading and executing the search program, the above-described units are loaded on the main storage device, and the above-described units are generated on the main storage device.

  As described above, the structured document search device, the structured document search method, and the structured document search program according to the present invention include a search device that searches for a structured document whose structure can be freely defined, such as XML, Suitable for search method and search program.

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 a structured document. It is explanatory drawing which shows an example of a structured document. It is explanatory drawing which showed the logical structure of the structured document. It is explanatory drawing which showed the logical structure of the structured document. It is explanatory drawing which shows an example of the data structure of the statistical information memorize | stored in a statistical information storage part. It is explanatory drawing which showed an example of the search result by a search part. 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 flow of the whole structure statistical information calculation process in 1st Embodiment. It is explanatory drawing which shows an example of the result of having performed the adaptation element name change process. It is explanatory drawing which shows an example of the calculated statistical information. It is a flowchart which shows the whole flow of the similar partial structure extraction process in 1st Embodiment. It is explanatory drawing which shows an example of the produced | generated feature vector. It is explanatory drawing which shows an example of the extracted similar partial structure. It is explanatory drawing which shows another example of the produced | generated feature vector. It is a flowchart which shows the whole flow of a partial document extraction process. It is explanatory drawing which showed another example of the logical structure of a structured document. It is explanatory drawing which showed another example of the logical structure of a structured document. It is a flowchart which shows the whole flow of a partial document integration process. It is explanatory drawing which showed an example of the output format of the partial document output. It is a block diagram which shows the structure of the structured document search apparatus concerning 2nd Embodiment. It is a flowchart which shows the whole flow of the structured document registration process in 2nd Embodiment. It is a flowchart which shows the flow of the whole structure statistical information calculation process in 2nd Embodiment. It is a block diagram which shows the structure of the structured document search apparatus concerning 3rd Embodiment. It is explanatory drawing which shows an example of the data structure of the structure information memorize | stored in the structure information storage part. It is a flowchart which shows the whole flow of the similar partial structure extraction process in 3rd Embodiment. It is explanatory drawing which shows the hardware constitutions of a structured document search apparatus.

Explanation of symbols

51 CPU
52 ROM
53 RAM
54 Communication I / F
61 Bus 100 Structured Document Search Device 101 Reception Unit 102 Search Unit 103 First Determination Unit 104 Determination Unit 105 Change Unit 106 Calculation Unit 107 Structure Extraction Unit 108 Document Extraction Unit 109 Second Determination Unit 110 Addition Unit 111 Output Unit 113 Generation Unit 122 statistical information storage unit 200 document management device 221 structured document storage unit 300 network 2100 structured document search device 2102 search unit 2103 first determination unit 2105 change unit 2112 registration unit 2121 structured document storage unit 2400 structured document search device 2407 Structure extraction unit 2413 Structure information storage unit

Claims (14)

A structured document search device connected via a network to a document management device for storing a structured document having a hierarchical logical structure,
A receiving means for receiving an input of a search keyword;
Search means for searching the document management device for a plurality of structured documents including the received search keyword;
Calculating means for calculating statistical information regarding the appearance of a structural element that is a unit of the logical structure included in the plurality of structured documents searched;
Generating means for generating feature vectors of the plurality of structured documents searched based on the calculated statistical information;
Based on the generated feature vector, the structural elements that characterize the plurality of searched structured documents are selected, and the structural elements that are similar between the searched structured documents are selected from the selected structural elements. A structure extracting means for extracting a similar partial structure,
Document extracting means for extracting a partial document in the structured document corresponding to the extracted similar partial structure;
An output means for outputting the extracted partial document for each structural element;
A structured document retrieval apparatus characterized by comprising: The structure extraction unit calculates a similarity between the structural elements selected in each of the plurality of searched structured documents, and the calculated similarity is determined in advance from the plurality of searched structured documents. Obtaining the structured document that is greater than the first threshold value, and extracting the structural element included in the obtained structured document as the similar partial structure;
The structured document search apparatus according to claim 1, wherein: The structure extraction means further extracts, as the similar partial structure, the structure element including the search keyword from among the structure elements included in the searched structured document;
The structured document search apparatus according to claim 1, wherein: The calculation means calculates, as the statistical information, a frequency at which the structural element included in the searched structured document appears in the structured document stored in the document management device,
The structure extraction means selects the structure element having the frequency greater than a predetermined second threshold as the structure element characterizing the structured document;
The structured document search apparatus according to claim 1, wherein: First determination means for determining whether the element names of the structural elements are synonymous with each other among the plurality of structured documents searched;
Change means for changing one element name that is synonymous to the other element name, and
The calculation means calculates the statistical information by adding the frequency of the structural element corresponding to the element name of the change source to the frequency of the structural element corresponding to the element name of the change destination,
The structured document search apparatus according to claim 4, wherein: For each of the search keywords, among the structural elements included in the structured document that has been searched, the element of the structural element having the highest degree of matching that represents the degree to which the actual information of the structural element matches the search keyword A determination means for determining a representative element name which is a name;
Among the structural elements included in the structured document that has been searched, the element name of the structural element that includes the search keyword in the element name is changed to the representative element name, and the search is performed as actual information of the changed structural element. And a changing means for associating the element name including the keyword,
The calculation means calculates the statistical information by adding the frequency of the structural element whose element name is changed to the representative element name to the frequency of the structural element corresponding to the representative element name;
The structured document search apparatus according to claim 4, wherein: Second determination means for determining whether or not the search keyword that is not included in the extracted partial document exists;
And an additional means for extracting the partial document including the search keyword from the structured document and adding it to the partial document when there is the search keyword that is not included in the extracted partial document. thing,
The structured document search apparatus according to claim 1, wherein: The document extracting means extracts the partial document including the largest number of the structural elements included in the similar partial structure;
The structured document search apparatus according to claim 1, wherein: The document extracting means extracts the partial document having the smallest number of hierarchies of the logical structure;
The structured document search apparatus according to claim 1, wherein: Further comprising structured document storage means for storing the structured document;
The search means searches the structured document storage means for the structured document including the received search keyword;
The structured document search apparatus according to claim 1, wherein: First determination means for determining whether element names of the structural elements of the structured document stored in the structured document storage means are synonymous with each other among the plurality of structured documents;
Change means for changing one element name having a synonym relation to the other element name and saving the changed first structured document in the structured document storage means;
The structured document search device according to claim 10. Further comprising: structure information storage means for storing the logical structure extracted from the plurality of structured documents;
The structure extraction unit acquires the logical structure corresponding to the searched structured document from the structure information storage unit, and characterizes the acquired logical structure based on the generated feature vector And extracting the similar partial structure from the selected structural element,
The structured document search apparatus according to claim 1, wherein: A structured document search method in a structured document search apparatus connected via a network to a document management apparatus that stores a structured document having a hierarchical logical structure,
An accepting step for accepting an input of a search keyword by accepting means;
A search step of searching the document management device for a plurality of the structured documents including the received search keyword by a search means;
A calculation step of calculating statistical information on the appearance of a structural element that is a unit of the logical structure included in the plurality of structured documents searched by the calculation unit;
A generation step of generating feature vectors of the plurality of structured documents searched based on the statistical information calculated by the structure extraction unit;
Based on the generated feature vector, the structural elements that characterize the plurality of searched structured documents are selected, and the structural elements that are similar between the searched structured documents are selected from the selected structural elements. A structure extraction step for extracting a similar partial structure,
A document extracting step of extracting a partial document in the structured document corresponding to the extracted similar partial structure by a document extracting means;
An output step of outputting the extracted partial document for each structural element by an output means;
A structured document search method characterized by comprising: A structured document search program in a structured document search apparatus connected via a network to a document management apparatus that stores a structured document having a hierarchical logical structure,
Acceptance procedure to accept search keywords,
A search procedure for searching the document management device for a plurality of structured documents including the received search keyword;
A calculation procedure for calculating statistical information relating to the appearance of a structural element that is a unit of the logical structure included in the plurality of structured documents searched;
A generating procedure for generating feature vectors of the plurality of structured documents searched based on the calculated statistical information;
Based on the generated feature vector, the structural elements that characterize the plurality of searched structured documents are selected, and the structural elements that are similar between the searched structured documents are selected from the selected structural elements. A structure extraction procedure for extracting similar partial structures,
A document extraction procedure for extracting a partial document in the structured document corresponding to the extracted similar partial structure;
An output procedure for outputting the extracted partial document for each structural element;
A structured document search program that causes a computer to execute.

Images