JP6564811B2 - Passage presentation control device, passage presentation method, and passage presentation program - Google Patents

Description

  The present invention relates to a passage presentation control device, a passage presentation method, and a passage presentation program, and particularly suitable for a request when a passage is answered to a question and search related to a structured document described in a structured language. The present invention relates to a technique for controlling creation and presentation of passages.

  A large amount of text data is available on the Internet, and the importance of techniques for retrieving necessary knowledge from documents is increasing. In a document search for a specific document, a passage level search is required to extract and output a part of a document highly relevant to a search request input by a user. Here, “passage” refers to a partial document in an arbitrary range having a content strongly related to the content of the search request in the document, and includes at least one sentence (text).

  For a structured document described in a structured language such as HTML / XML (also referred to as a markup language), for example, a part of the tag structure in the structured document is automatically created as a passage and searched. There is a method presented as a result (see, for example, Non-Patent Document 1).

  In this method, by creating multiple tag structure parts that have a higher / lower relationship in a structured document as one passage, not only text that describes information but also text related to headings, headers, and conditions as search results. Can be presented.

Satoshi, Jun Miyazaki, Kenji Hatano, Goro Yamamoto, Takashi Taketomi, Hirokazu Kato "Application of XML Partial Document Retrieval Technology to Web Documents", DEIM Forum, 2014

  When searching in a structured document (hereinafter referred to as “document”), there is an expectation that knowledge related to the search request will be confirmed at the same time, and an interface that can be confirmed at the same time as viewing the document, such as highlighting text related to passages in the document. Is desirable.

  Here, for example, consider a case where a passage generated from a document is viewed on a browser.

  Since there is a range where the document can be displayed at once (hereinafter referred to as “displayable range”) on the browser, if there is a huge passage consisting of continuous text or a passage consisting of text scattered in the document, There may be a problem that the passage cannot be presented as a search result on the browser at a time.

  For huge passages composed of continuous text, the amount of information is excessive and is not appropriate for a search request. Therefore, in Non-Patent Document 1, a passage with a text size larger than a certain value is not created, or the text size Is taking measures to exclude passages above a certain level as output results. By the correspondence, presentation of passages including a text size larger than the displayable range is controlled.

  On the other hand, a passage that is composed of text scattered in a document and exceeds the displayable range may or may not be an appropriate passage for a search request.

  FIG. 4 is a diagram illustrating an example of the type of passage displayed on a document. FIG. 4A shows a passage including text 1 and text 2 included in the displayable range of the display area 8. In this case, since each text is included in the displayable range, the text 1 and the text 2 can be presented on the browser at a time.

  FIG. 4B shows a passage having continuous text 1 exceeding the displayable range. In this case, naturally, the text 1 cannot be presented on the browser at a time.

  FIG. 4C shows a passage having a plurality of text exceeding the displayable range when the text 1 and the text 2 are combined when the text alone is included in the displayable range.

  FIG. 4D shows a passage including a text 1 and a text 2 having a specific correspondence, for example, a header and content corresponding to a headline or a condition. In this case, each text alone is included in the displayable range, but when the text 1 and the text 2 are combined, an example exceeding the displayable range is shown.

  FIG. 4E is a composite example of the passage shown in FIG. 4C and the passage shown in FIG. 4D. Text 1 and text 3, and text 2 and text 4 correspond to headers and contents, respectively. Have a relationship. In addition, each text is included in the displayable range, but the entire text 1 to text 4 represents a passage having a plurality of texts exceeding the displayable range.

  The passages shown in FIGS. 4C, 4D, and 4E are examples of passages composed of scattered text.

  Here, the passages shown in FIG. 4C and FIG. 4E each include a plurality of texts corresponding to the contents, and there are cases where the amount of information to be presented at one time is excessive and not suitable for a search request. Many. Therefore, control such as not presenting a specific text or presenting a set of text divided is required.

  FIG. 5 is a diagram showing a detailed example of the passage shown in FIG.

  The passage shown in FIG. 5 describes a text (content included in a displayable range) in which content that can present appropriate information in response to a search request is described, and a header indicating a heading, a condition, and the like for the text. This is an example of a passage made up of text (text indicated by <h2> and <h3> tags). In this case, as shown in FIG. 5, for example, by performing control such as presenting only the text describing the content, it is possible to include appropriate content for the search request and present it on the browser at a time. It seems that a passage can be presented.

  However, as in Non-Patent Document 1, in the method of controlling the presentation of passages using an index such as text size, documents such as those shown in FIGS. 4 (C), 4 (D), and 4 (E) are used. Since it is not possible to determine which type corresponds to a passage composed of text scattered therein, there is a problem that appropriate control for presentation of the passage cannot be performed.

  The present invention has been made in view of the above circumstances, and is capable of presenting a passage suitable for a request from a structured document having a hierarchical document structure in consideration of the displayable range. It is an object to provide a control device, a passage presentation method, and a passage presentation program.

  In order to achieve the above object, the passage presentation control apparatus according to the first invention is such that the overall size of the passage including the text described using the tag indicating the text structure is within the range of the display area in the display device. Determining whether or not it is included, and accepting two texts, a display enablement determining unit that determines whether or not a passage that is determined not to be included in the display area is composed of a plurality of texts, One of the two received texts is a header and the other is content, and the header / content relationship estimation unit and the displayability determination unit are not included in the display area, and For a passage determined to be composed of multiple texts, all combinations of two texts included in the passage are extracted, and the extracted text If the size of the set of text estimated as content is not included within the range of the display area using the estimation result of the header / content relation estimation unit for each combination of the list, the text estimated as content A passage filter unit that divides a set of text estimated as content based on a document structure tree obtained from the tag included in the passage into subsets until the size of the subset is included within the range of the display area When the display size determining unit determines that the entire size of the passage is included in the display area, the display unit displays the entire passage in the display region, and the display enable determination unit displays the display When it is determined that it is not included in the range of the region and is composed of a plurality of texts, the passage filter unit Preferentially a subset of al accepted text and a passage presenting unit to be displayed in the display area.

  In the passage presentation control apparatus according to the first invention, the header / content relation estimation unit converts each of the received two texts into a dependency tree representing a dependency of a clause included in the text, and the document Combining each dependency tree of the two texts to a structure tree, and based on the smallest subtree containing each dependency tree of the two texts in the document structure tree, one of the two texts is a header. Yes, it is estimated whether the other is content.

  In the passage presentation control apparatus according to the first aspect, the header / content relation estimation unit uses the tag included in the two texts, and one of the two texts is a header and the other is content. Estimate whether or not there is.

  In the passage presentation control apparatus according to the first aspect, when the passage presentation unit determines that the entire size of the passage is not included in the range of the display area, and the passage is composed of a plurality of texts. The text subset received from the passage filter unit is preferentially displayed in the display area, and the text estimated as a header corresponding to the text subset received from the passage filter unit is displayed. Displayed in the header display area of the device.

  The passage presentation method according to the second invention determines whether or not the entire size of the passage including the text described using the tag indicating the text structure is included in the range of the display area in the display device. Determining whether or not the passage determined to be not included in the range of the display area is composed of a plurality of texts; and not included in the range of the display area and composed of a plurality of texts Two texts included in the passage determined to be received, a step of estimating whether one of the two received texts is a header and the other is content, and a set of texts estimated as content If the size is not included within the display area, the size of the subset of text estimated as content is the range of the display area. Dividing the set of text estimated as content based on the document structure tree obtained from the tag included in the passage into a subset until the size of the passage is within the display area. If it is determined that it is included in the display area, the entire passage is displayed in the display area, and if it is determined that the passage is not included in the display area and is composed of a plurality of texts, it is divided. Displaying the displayed subset of text in the display area with priority.

  In the step of estimating whether one of the two texts received by the passage presentation method according to the second invention is a header and the other is a content, each of the two texts is replaced with a phrase included in the text. Converting to a dependency tree representing a dependency, combining the dependency tree of each of the two texts with the document structure tree, and including the dependency tree of each of the two texts in the document structure tree; , It is estimated whether one of the two texts is a header and the other is content.

  In the step of estimating whether one of the two texts accepted by the passage presentation method according to the second invention is a header and the other is a content, using the tag included in the two texts, It is estimated whether one of the two texts is a header and the other is content.

  A passage presentation program according to a third aspect causes a computer to function as each part of the passage presentation control device according to any one of claims 1 to 4.

  As described above, according to the passage presentation control apparatus, the passage presentation method, and the passage presentation program of the present invention, it is suitable for a request from a structured document having a hierarchical document structure in consideration of a displayable range. It has the effect that a passage can be presented.

It is a figure which shows the structural example of a passage presentation control apparatus. It is a flowchart which shows an example of the flow of a passage presentation control routine. It is a flowchart which shows an example of the flow of a passage presentation control routine. It is a figure which shows the example of a classification on the display of a passage. It is a figure which shows the detailed example of a passage.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following description, the same reference numerals are given to the components or processes having the same functions throughout the drawings, and the repeated description is omitted as appropriate.

<Configuration of Passage Presentation Control Device According to Embodiment of the Present Invention>
FIG. 1 is a diagram showing a configuration example of a passage presentation control apparatus 100 according to an embodiment of the present invention.

  The passage presentation control apparatus 100 is composed of a computer including a CPU, a RAM, and a ROM that stores a passage presentation program for executing a passage presentation control routine described later, and functionally as shown in FIG. It is configured.

  That is, the passage presentation control apparatus 100 includes the input unit 10, the calculation unit 20, and the storage unit 30. The calculation unit 20 includes a display enablement determination unit 22, a passage filter unit 24, a header / content relationship estimation unit 26, and a passage presentation unit 28.

  The input unit 10 accepts, for example, passage candidates (hereinafter simply referred to as “passage”) obtained as a result of a query and search regarding a document, stores the received passage in the storage unit 30, and sends a reception notification to the calculation unit 20. Notice.

  When the calculation unit 20 receives a passage reception notification from the input unit 10, the display possibility determination unit 22 first displays, for example, the displayable range of the display area 8 on the browser in which the size of the received passage is displayed on the display device. It is determined whether it is included in. Whether or not the accepted size of the passage is included in the displayable range can be displayed in the display area 8 by counting the number of characters (text size) of the text included in the passage as shown in Non-Patent Document 1, for example. Judged by comparing with the correct text size.

  The display possibility determination unit 22 outputs the received passage to the passage presentation unit 28 when the size of the entire passage is included in the displayable range.

  On the other hand, when the size of the entire passage is not included in the displayable range and the passage is composed of a single continuous text that does not include a break such as a punctuation mark, the displayable determination unit 22 displays Information indicating that the passage cannot be output to the area 8, that is, output impossibility information is output to the passage presentation unit 28.

  In addition, when the size of the entire passage is not included in the displayable range and the passage is made up of a plurality of divided texts by a break such as a punctuation mark, the displayability determining unit 22 displays the received passage. Output to the passage filter unit 24.

  The passage filter unit 24 receives a passage composed of a plurality of texts from the display possibility determination unit 22, and based on the header / content relationship between the texts, the passage corresponding to the text corresponding to the header and the text corresponding to the content. A passage in which something is associated is generated, and the generated passage is output to the passage presentation unit 28.

  Specifically, the passage filter unit 24 selects all combinations of two texts included in the passage, and uses each estimation result of the header / content relationship estimation unit 26 described later to determine each combination of the selected texts. On the other hand, it is estimated whether one of the two texts is a header and the other is content.

  The passage filter unit 24 determines whether or not the size is included in the displayable range with respect to a set of text estimated as content (hereinafter referred to as “content text”) among the text included in the passage. . If the size of the set of content texts exceeds the displayable range, the passage filter unit 24 converts the set of content texts into subsets using the upper tags based on the document structure tree represented using the tag structure. To divide. At this time, the passage filter unit 24 sequentially subdivides the division unit of the set of content text from the upper tag to the lower tag until the size of the divided subset of the content text is included in the displayable range. Repeat the process of dividing the text set into subsets.

  The passage filter unit 24 includes a subset of the content text divided until the size is included in the displayable range, and a text estimated as a header corresponding to each content text in the subset (hereinafter, “header text”). )) Is regarded as one passage and is output to the passage presentation unit 28.

  When the header-content relationship estimation unit 26 receives two texts including a tag from the passage filter unit 24, one of the two received texts is a header, and the other is content indicating the contents of the header. And the estimation result is output to the passage filter unit 24.

  As an estimation method for estimating the header-content relationship, for example, a supervised learning method described in Non-Patent Document 2 can be used.

[Non-Patent Document 2]: Richard Socher, Eric H. Huang, Jeffrey Pennington, Andrew Y. Ng, Christopher D. Manning. "Dynamic Pooling and Unfolding Recursive Autoencoders for Paraphrase Detection" Advances in Neural Information Processing Systems, pp.801- 809, 2011.

  Specifically, the header / content relationship estimation unit 26 constructs text by performing dependency analysis using sequence labeling described in Non-Patent Document 3, for each of the two texts. A dependency tree representing the dependency relationship of a clause is acquired for each text.

[Non-Patent Document 3] Kenji Imamura, Genichiro Kikui and Norihito Yasuda. "Japanese Dependency Parsing Using Sequential Labeling for Semi-spoken Language" In Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics Companion Volume Proceedings of the Demo and Poster Sessions , pp.225-228, 2007.

  Next, the header / content relation estimation unit 26 combines dependency trees corresponding to the respective texts with the document structure tree representing the entire passage, and represents the respective texts from the document structure tree representing the entire passage. Generate a minimum subtree that contains a receiving tree.

  Then, the header / content relationship estimation unit 26 uses the classifier obtained by supervised learning for the generated minimum subtree, and one of the two texts is the header and the other is the content of the header. It is estimated whether or not the content is represented.

  Another method for estimating the header / content relationship is to use a tag included in two texts.

  Specifically, when the text is an HTML document, for example, the header / content relationship estimation unit 26 includes a <hn> (n is an integer of 1 or more) tag in which one text represents a header, and the other text is content. When a <p> tag representing a paragraph, which is an example of a tag used in the description of, is included, it can be estimated that one text is a header and the other text is content.

  The passage presentation unit 28 performs control related to the presentation of the passage according to the type of received data.

  When the passage presentation unit 28 receives the entire passage from the displayability determination unit 22, the passage presentation unit 28 presents the passage so that the entire passage is displayed in the display area 8. Here, presentation of the passage means setting of attributes (display attributes) relating to the display of the passage, such as the range of the passage displayed in the display area 8 and the display position of the passage in the display area 8.

  In addition, when the passage presenting unit 28 receives the non-outputable information from the displayability determining unit 22, a message indicating that the question about the document and the passage for the search cannot be displayed is displayed in the display area 8. Set display attributes.

  Further, when the passage presenting unit 28 receives a passage from the passage filter unit 24, the passage presenting unit 28 presents the passage according to the type of the received passage.

  Specifically, when the size of the entire text included in the received passage is included in the displayable range, the passage presentation unit 28 displays the passage so that all the text included in the passage is displayed in the display area 8. Make a presentation.

  When the header text included in the accepted passage is included and the entire text is not included in the display area 8, the passage presentation unit 28 presents the passage so that the content text is preferentially displayed in the display area 8. I do.

  In this case, the passage presentation unit 28 prepares an interface (header display area) for displaying header text that cannot be displayed in the display area 8, so that the entire received passage is hierarchically divided into the display area 8 and the header display area. Passage may be presented so that it can be presented.

<Operation of the passage presentation control apparatus according to the embodiment of the present invention>
Next, the operation of the passage presentation control apparatus 100 according to the embodiment of the present invention will be described. When the CPU of the passage presentation control apparatus 100 reads the passage presentation program stored in the ROM and accepts a passage representing a question about a document and a search response, the passage presentation program stipulated in the passage presentation program is shown in FIGS. The passage presentation control routine shown is executed. It is assumed that the passage presentation control apparatus 100 generates a document structure tree for the received passage in advance using a known technique and stores it in the storage unit 30 before executing the passage presentation control routine.

  First, in step S <b> 10, the size of text included in the passage received by the input unit 10 is counted to determine whether or not the entire passage is included in the displayable range of the display area 8. Note that the size of text that can be displayed in the displayable range may be stored in advance in the storage unit 30, for example. When the size of the entire passage is included in the displayable range, the process proceeds to step S130.

  In this case, since the size of the entire passage is included in the displayable range, the passage received by the input unit 10 is displayed in the display area 8 as it is, and the passage presentation control routine is terminated.

  On the other hand, if the determination process in step S10 is negative, that is, if it is determined that the size of the entire passage is not included in the displayable range, the process proceeds to step S20.

  In step S20, it is determined whether or not the accepted passage is composed of one continuous text. When the passage is composed of one continuous text, the process proceeds to step S140.

  In this case, since the passage is composed of a single continuous text, the passage becomes meaningless if the passage is divided. Therefore, since the passage cannot be divided any more, and the size of the entire passage cannot be displayed within the displayable range, a message indicating that the passage cannot be output is displayed in step S140. It is created and displayed in the display area 8, and the passage presentation control routine is terminated.

  On the other hand, if the determination process in step S20 is negative, that is, if the received passage is composed of a plurality of divided texts, the process proceeds to step S30.

  In step S30, one unselected combination is selected from all the combinations related to the two texts included in the received passage.

  In step S40, using the method described in the header / content relationship estimation unit 26, it is estimated whether one of the two texts selected in step S30 is a header and the other is content.

  In step S50, it is determined whether or not all combinations relating to two texts have been selected for the received passage. If the determination is negative, that is, if there is a combination of unselected texts, the process proceeds to step S30, and a combination of two unselected texts is selected. By repeatedly executing the processing of steps S30 to S50, it is estimated whether one is a header and the other is content with respect to all combinations of two texts included in the received passage. become.

  When all the combinations related to the two texts are selected, the determination process in step S50 is affirmative, and the process proceeds to step S60.

  In step S60, it is determined whether or not the text associated with the content in step S40, that is, the size of the entire set of content texts is included in the displayable range. When the entire set of content texts is included in the displayable range, the process proceeds to step S90 described later.

  On the other hand, when the determination process in step S60 is negative, that is, when the size of the entire set of content texts is not included in the displayable range, the process proceeds to step S70.

  In step S70, based on the document structure tree representing the received passage stored in advance in the storage unit 30, the upper tag is referred to and the set of content texts is divided into subsets.

  In step S80, it is determined whether the size of each subset of the content text divided in step S70 is included in the displayable range. If the subset of the content text includes a subset having a size not included in the displayable range, the process proceeds to step S70. In step S70, the subset of the content text is further subdivided into a plurality of parts using a tag positioned higher in the subtree of the document structure tree corresponding to the subset of the content text not included in the displayable range. Divide into sets.

  That is, by repeatedly executing steps S70 and S80, a plurality of content text subsets each having a size included within the displayable range are generated.

  In step S90, one of the content text subsets divided up to the size included in the displayable range in step S70 and a set of header texts corresponding to each content text in the subset are generated as edit passages.

  In step S100, it is determined whether or not the size of the edit passage generated in step S90 is included in the displayable range. When the size of the edit passage is included in the displayable range, the process proceeds to step S110, and the edit passage is displayed in the display area 8 in step S110. Accordingly, it is possible to present an appropriate passage in which the content text and the header text corresponding to the content text are not interrupted in the middle of the text.

  On the other hand, if the determination in step S100 is negative, that is, if the size of the edit passage is not included in the displayable range, the process proceeds to step S120.

  In many cases, content text is more appropriate as a question about a document and an answer to a search than the header text. Therefore, in step S120, the content text included in the edit passage is preferentially included in the display area 8. The passage is displayed in the display area 8.

  That is, even if the passage text cannot be displayed within the displayable range when the header text is included, an appropriate passage that does not interrupt the content text in the middle of the text can be displayed in the display area 8.

  Here, as an example, an example has been described in which a passage generated from one of the divided content text subsets is displayed in the display area 8, but each of the content text subsets divided in step S70 corresponds to each. The edited passages may be generated, and steps S100 to S120 may be executed for each edited passage based on a user instruction so that each of the edited passages is sequentially displayed in the display area 8.

  For example, when there are a plurality of edit passages, a message with a link indicating that the next answer exists is displayed in the display area 8, and when the user selects a message with a link using a mouse or the like, an undisplayed edit passage is displayed. May be displayed in the display area 8.

  As described above, according to the passage presentation control apparatus 100 according to the embodiment of the present invention, a passage suitable for a question and a search response is generated from a document having a hierarchical document structure and displayed in the display area 8. The passage is generated and displayed in consideration of the displayable range. Therefore, it is possible to present a passage suitable for the user's request in a form that the user can easily grasp from a document having a hierarchical document structure.

  In addition, the passage presentation control apparatus 100 uses the estimation result to estimate whether one of the two texts is a header and the other is content, so that the passages are associated with each other in the passage. Passage can be presented according to the type of the.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made without departing from the gist of the present invention.

  For example, the passage presentation control apparatus 100 may be realized by a plurality of computers connected by communication lines.

  Moreover, although the embodiment in which the passage presentation program is installed in the ROM in advance has been described, the program can be provided by being stored in a computer-readable recording medium and provided via a communication line. Is also possible.

DESCRIPTION OF SYMBOLS 8 Display area 10 Input part 20 Calculation part 22 Display possibility determination part 24 Passage filter part 26 Header content relation estimation part 28 Passage presentation part 30 Storage part 100 Passage presentation control apparatus

Claims (8)

It is determined whether or not the entire size of the passage including the text described using the tag indicating the text structure is included in the display area of the display device, and is not included in the display area. A displayability determination unit that determines whether or not the passage determined to be composed of a plurality of texts;
A header-content relationship estimation unit that accepts two texts, and estimates whether one of the two received texts is a header and the other is content;
For the passage that is not included in the display area and is determined to be composed of a plurality of texts by the display enablement determination unit, all combinations of two texts included in the passage are extracted, Using the estimation result of the header-content relationship estimation unit for each combination of extracted text, if the size of the set of text estimated as content is not included in the range of the display area, it is estimated as content. A passage filter that divides a text set estimated as content based on a document structure tree obtained from the tag included in the passage into subsets until the size of the text subset included in the display area is included. And
When the displayable determination unit determines that the entire size of the passage is included in the range of the display area, the entire passage is displayed in the display area, and the displayability determination unit displays the display area. A passage presenting unit that preferentially displays a subset of the text received from the passage filter unit in the display area when it is determined that it is not included in the range and is composed of a plurality of texts;
Passage presentation control device. The header / content relationship estimating unit converts each of the received two texts into a dependency tree representing a dependency of a clause included in the text, and converts each of the two texts into the document structure tree. Estimate whether one of the two texts is a header and the other is content based on a minimum subtree that includes a dependency tree for each of the two texts in the document structure tree The passage presentation control apparatus according to claim 1. The header-content relation estimation unit estimates whether one of the two texts is a header and the other is content by using the tag included in the two texts. Item 3. The passage presentation control apparatus according to Item 2. The passage presentation unit, when it is determined that the entire size of the passage is not included in the range of the display area and the passage is composed of a plurality of texts, a subset of the text received from the passage filter unit The text presumed to be a header corresponding to a subset of the text received from the passage filter unit is displayed in the header display area in the display device. The passage presentation control apparatus according to claim 3. It is determined whether or not the entire size of the passage including the text described using the tag indicating the text structure is included in the display area of the display device, and is not included in the display area. Determining whether the passage determined to be composed of a plurality of texts;
Two texts included in a passage that is determined not to be included in the display area and are composed of a plurality of texts are received, and one of the two received texts is a header and the other is content. Estimating whether or not,
If the size of the set of text estimated to be content is not included in the display area, it is included in the passage until the size of the subset of text estimated to be content is included in the display area. Dividing a set of text estimated as content based on a document structure tree obtained from the tag into subsets;
When it is determined that the entire size of the passage is included in the range of the display area, the entire passage is displayed in the display area, is not included in the display area, and includes a plurality of texts. If it is determined that it is composed, the step of displaying the subset of the divided text with priority in the display area;
Passage presentation method including In the step of estimating whether one of the two received texts is a header and the other is a content, each of the two texts is converted into a dependency tree representing a dependency of a clause included in the text. , By connecting each dependency tree of the two texts to the document structure tree, and using a minimum subtree including each dependency tree of the two texts in the document structure tree. The passage presentation method according to claim 5, wherein one is a header and the other is content. In the step of estimating whether one of the two received texts is a header and the other is content, one of the two texts is a header using the tag included in the two texts The passage presentation method according to claim 5, wherein the other is content.   The passage presentation program for functioning a computer as each part of the passage presentation control apparatus of any one of Claims 1-4.