JP7041103B2 - Structured document creation device and its method - Google Patents

Description

The present invention relates to a structured document creating apparatus for creating a structured document from an unstructured document and a method thereof.

There is a technology to extract information from knowledge and documents such as manuals. There is also a technology that processes information using machine learning based on the extracted information and generates answers to questions. Information needs to be structured (tagged) for effective machine learning.

However, the document format such as a product manual is often an unstructured document (untagged PDF). Therefore, in order to perform information processing using machine learning, pre-processing for structuring an unstructured document is required.

Focusing on the file format, there is optical character recognition (OCR) as a technique for creating a structured document from an unstructured document. However, OCR can only reproduce the layout and cannot acquire section information representing the structure of the document such as chapters and sections.

A structured document creating apparatus for creating a structured document is disclosed in, for example, Patent Document 1.

Japanese Unexamined Patent Publication No. 2010-186325

However, in the technology disclosed in Patent Document 1, the document to be structured is limited to the business report. Documents that do not conform to the business report format cannot be structured. That is, the conventional technique has a problem of low versatility.

The present invention has been made in view of this problem, and an object of the present invention is to provide a highly versatile structured document creation device and a method thereof that can handle all kinds of documents as long as they are classified unstructured documents. And. The classified unstructured documents are, for example, manuals, instruction manuals, specifications, and the like. Unstructured documents that are not classified are, for example, essays.

The structured document creation device of the present invention is a structured document creation device that creates a structured document described so as to conform to a predetermined type from a classified unstructured document, and is the unstructured document. For each page, a document structure acquisition unit that acquires a text frame element, a table frame element, and an image frame element, and the positions of the text frame element, the table frame element, and the image frame element on the page and their positions thereof. A management list generation unit that generates a management list that is listed in association with the contents, a text frame element arrangement unit that refers to the management list and arranges the text frame elements in the writing order for each page, and the arranged text. The text information structuring unit that generates the first structured document in which the frame element is divided based on the rule set in the division definition property, and the table frame element in the first structured document with reference to the management list. The rule includes a rule for identifying a font size, a character position, and a heading .

Further, the structured document creation method of the present invention is a structured document creation method performed by the above-mentioned structured document creation apparatus, and a text frame element, a table frame element, and an image are used for each page of the unstructured document. A management to generate a management list in which the document structure acquisition step for acquiring a frame element and the position of each of the text frame element, the table frame element, and the image frame element on the page and their contents are associated with each other and listed. Based on the rule set in the division definition property, the list generation step, the text frame element array step that refers to the management list and arranges the text frame elements for each page in the writing order, and the arranged text frame elements. A text information structuring step to generate a divided first structured document , and an HTML file to generate an HTML file in which the table frame element and the image frame element are inserted into the first structured document with reference to the management list. The rules include a rule for identifying font size, character position, and heading .

INDUSTRIAL APPLICABILITY According to the present invention, it is possible to provide a highly versatile structured document creation device and a method thereof that can handle all kinds of documents as long as they are classified unstructured documents.

It is a figure which shows the functional structure example of the structured document making apparatus which concerns on 1st Embodiment of this invention. It is a flowchart which shows the processing procedure of the structured document making apparatus shown in FIG. It is a figure which shows typically one page of the divided unstructured document. It is a figure which shows the example of a part of the text frame element acquired by the document structure acquisition part shown in FIG. It is a figure which shows the example of the text management list generated by the management list generation part shown in FIG. It is a figure which shows an example of the image management list generated by the management list generation part shown in FIG. It is a figure which shows the example of the table management list generated by the management list generation part shown in FIG. It is a figure which shows an example of the division rule set in the division definition property shown in FIG. It is a figure which shows the example of the 1st structured document generated by the text information structuring part shown in FIG. It is a figure which shows the functional structure example of the structured document making apparatus which concerns on 2nd Embodiment of this invention. It is a flowchart which shows the processing procedure of the structured document making apparatus shown in FIG. It is a figure which shows typically the process performed by the division part shown in FIG.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. The same reference numerals are given to the same objects in a plurality of drawings, and the description is not repeated.

[First Embodiment]
FIG. 1 is a diagram showing a functional configuration example of the structured document creating device according to the first embodiment of the present invention. The structured document creation device 100 shown in FIG. 1 is a device that creates a structured document described so as to conform to a predetermined type from a classified unstructured document.

The structured document creation device 100 includes a PDF generation unit 10, a document structure acquisition unit 20, a management list generation unit 30, a text frame element arrangement unit 40, a text information structuring unit 50, an HTML organization unit 60, an operation unit 70, and a display. A unit 80 is provided. The structured document creation device 100 can be realized by, for example, a computer including a ROM, a RAM, a CPU, and the like. When each functional component is realized by a computer, the processing content of the function that each functional component should have is described by a program. This also applies to other embodiments described later.

In FIG. 1, the notation of the communication unit that communicates with other devices via a network (not shown) is omitted. The description of the communication unit (not shown), the operation unit 70, and the display unit 80, which are general functional components, will be omitted.

The PDF generation unit 10 is composed of, for example, application software for creating a document, and generates a document created by a user in a PDF (Portable Document Format) file structure. The PDF file structure consists of a page tree and an outline tree. The page tree contains page-by-page content and thumbnails (reduced samples). The outline tree contains bookmark information and the like indicating chapters and sections of the document. Hereinafter, the page is referred to as a page.

FIG. 2 is a flowchart showing a processing procedure of the structured document creating apparatus 100. From here, the operation of the structured document creation device 100 will be described with reference to FIGS. 1 and 2.

The document structure acquisition unit 20 acquires a text frame element, a table frame element, and an image frame element for each page of the divided unstructured document (step S1). A classified unstructured document is a document whose structure such as chapters and sections is clear.

FIG. 3 is a diagram schematically showing one page of the divided unstructured document targeted by the present embodiment. As shown in FIG. 3, one page P of an unstructured document includes a title T, a subtitle ST, a text C, a figure Z, and a table H. It should be noted that there may be a page P composed of only the text C (not including FIG. Z and the like).

In FIG. 3, the text frame elements are the title T, the subtitle ST, and the text C. The table frame element is table H. The image frame is shown in FIG. Z. Each is referred to as a frame element because each is described in a predetermined range (coordinates) on page P.

FIG. 4 shows a part of the text frame element acquired by the document structure acquisition unit 20. Text frame elements are represented in XML (eXtensible Markup Language) format.

Each of the lines shown in FIG. 4 represents the font that constitutes the document. The 1st and 2nd lines are the font "ko", the 3rd and 4th lines are the same "no", and the 5th and 6th lines are the same "ta". The 7th and subsequent lines are omitted.

In FIG. 4, ahp: l represents the left position of the font, ahp: r represents the same right position, ahp: t represents the same top position, and ahp: b represents the coordinates on the page P at the same bottom position.

The management list generation unit 30 generates a management list in which the positions (coordinates) of the text frame element, the table frame element, and the image frame element on each page P and their contents are associated and listed (step S2). The text frame element acquires a character string by using the appearance order of the font represented by XML. Also, the position of the character string is acquired from the left position ahp: l to the lower position ahp: b.

The positions of the text frame elements of "This time" shown in Fig. 4 are the left position ahp: l = 62.32, the right position ahp: r = 124.12, the upper position ahp: t = 739.70, and the lower position ahp: b = 749.70. be. The management list generation unit 30 lists the positions of the text frame elements and their contents in association with each other.

FIG. 5 is a diagram showing an example of a text management list 31 in which text frame elements are listed. The text management list 31 includes, for example, a character (character string), a character string length, a page, a left position, an upper position, a right position, a lower position, an operation number, and a stroke order item. The operation No. and the stroke order are information indicating the order of appearance of characters.

The management list generation unit 30 generates the image management list 32 and the table management list 33 by associating the positions of the figures and tables in the page P with their respective positions and their contents in the same manner as the text frame elements.

FIG. 6 is a diagram showing an example of an image management list 32 in which image frame elements are listed. The image management list 32 includes items of an image, a page, a left position, an upper position, a right position, and a lower position. The image consists of an image file name and an image file path.

FIG. 7 is a diagram showing an example of a table management list 33 in which table frame elements are listed. The table management list 33 includes items of a table ID, a page, a left position, an upper position, a right position, and a lower position.

By referring to each of the text management list 31, the image management list 32, and the table management list 33, all the contents described on the page P can be reproduced.

The text frame element arrangement unit 40 refers to the text management list 31 and rearranges the text frame elements page by page (step S3). For example, as shown in FIG. 3, the text frame elements are arranged in the order of title T, subtitle ST, and body C.

The text information structuring unit 50 arranges the text frame elements arranged by the text frame element arrangement unit 40 with any one of the font size, the character position, the regular expression representing a predetermined rule, and the bookmark information set in the division definition property 51. Generate a first structured document structured based on the rule of (step S4). The division definition property 51 is information set by the user in advance, and defines rules such as arranging text frame information based on bookmark information and separating title T and subtitle ST according to the font size. A specific example of a regular expression will be described later.

FIG. 8 is a diagram showing an example of a rule set in the division definition property 51. As shown in FIG. 8, a rule is set such that the smaller the x-coordinate and the y-coordinate on the page P, the higher the section, and the larger the font size, the higher the section. A section is a delimited part of a document and represents each chapter> section> section.

Adding the character position (10, 5) ⇒ <h1> means adding the heading tag <h1> to the character string described in the coordinates (10, 5). The coordinates (10, 5) are an example. The heading tag <h1> is attached to the chapter title, for example.

When the font size is 14p⇒ <h2>, it means that the heading tag <h2> is added to the character string of the font size of 14p. Starting with (number) ⇒ Adding <h3> means adding the heading tag <h3> when the beginning of the line is a number. Heading tags are available up to <h6>. The size of the character string represented by the heading tag is related to <h1 >> <h2 >> <h3> ....

FIG. 9 is a diagram showing an example of a first structured document generated by the text information structuring unit 50. The element between the start tag <h1> and the end tag </ h1> of the heading tag is "Section 3 About the Optical Collaboration Reception Center". For example, the rule that "character position is (10, 5) ⇒ <h1> is given" is applied to this element, and it becomes the top heading of the page P.

In the second line shown in FIG. 9, "font size is 14p⇒ <h2> is given" set in the division definition property 51 is applied, and "1. What is the optical collaboration reception center" is started with the start tag <h2>. It indicates that it was structured by sandwiching it with the end tag </ h2>.

The third line starts with (number) ⇒ The rule of <h3> is applied, and "(1) Overview" is structured by sandwiching the start tag <h3> and the end tag </ h3>. Is shown. The 4th to 7th lines are structured so that the body C of the text frame element is sandwiched between tags <p>.

In this way, the text information structuring unit 50 structures the text frame elements arranged with reference to the management list based on the rules set in the division definition property. Therefore, it is possible to structure a document based on a rule even if it is a document of any kind.

The rule shows an example based on the character position and the font size, but other possibilities are possible. A regular expression representing a predetermined rule, or a rule based on bookmark information contained in an unstructured document may be set in the division definition property.

A regular expression that represents a given rule is, for example, a fixed rule that identifies a heading. For example, regular expressions such as section level 1: (I-IX) and section level 2: (1-9) can be considered.

Section level 1: (I-IX) is a regular expression that structures Roman numerals in parentheses as section level 1 (eg, "chapter"), such as (I), (II), ..., (X). Is. Further, section level 2: (1-9) structures arabia numbers sandwiched between parentheses such as (1), (2), ..., (10) as section level 2 (for example, "section"). It is a regular expression.

Further, the text frame elements arranged by referring to the management list may be structured based on the bookmark information included in the outline tree of the PDF file structure of the unstructured document.

The HTML-forming unit 60 generates an HTML file in which FIG. Z and Table H are inserted into the first structured document generated by the text information structuring unit 50 (step S5). FIG. Z is inserted into the first structured document with reference to the image management list 32 (FIG. 6). Table H is inserted into the first structured document with reference to the table management list 33 (FIG. 7).

As described above, the structured document creation device 100 according to the present embodiment is a structured document creation device that creates a structured document described so as to conform to a predetermined type from a classified unstructured document. Therefore, for each page of the unstructured document, the document structure acquisition unit 20 for acquiring the text frame element, the table frame element, and the image frame element, and each page of the text frame element, the table frame element, and the image frame element. A management list generation unit 30 that generates a management list in which the positions in the above are associated with each other and their contents are listed, a text frame element arrangement unit 40 that refers to the management list and arranges text frame elements page by page, and an arranged text. The frame element is provided with a text information structuring unit 50 that generates a first structured document in which the frame element is structured based on the rule set in the division definition property 51. This makes it possible to provide a highly versatile structured document creation device and a method thereof that can handle all kinds of documents as long as they are classified unstructured documents.

[Second Embodiment]
FIG. 10 is a diagram showing a functional configuration example of the structured document creation device according to the second embodiment of the present invention. The structured document creating device 200 shown in FIG. 10 differs from the structured document creating device 100 (FIG. 1) in that it includes a dividing unit 210 and an HTML-forming unit 260. FIG. 11 is a flowchart showing a processing procedure of the structured document creating apparatus 200.

The division unit 210 vectorizes each of a plurality of sentences included in the text constituting the first structured document generated by the text information structuring unit 50 to obtain the similarity before and after, and determines the similarity and the division definition property 51. A second structured document in which the above text is divided is generated by comparing with the threshold value set in (step S6).

A plurality of sentences contained in the text are vectorized using, for example, a well-known BoW (bag-of-words). BoW morphologically analyzes a sentence, assigns a unique numerical value to the divided words, and converts all the words into one hot vector. One hot vector is a vector in which the words existing in the above text are set to 1 and the words other than the above are set to 0. BoW is well known and further description is omitted.

The division unit 210 obtains the similarity before and after the vectorized sentence. Then, the division unit 210 compares the threshold value set in the division definition property 51 with the degree of similarity to generate a second structured document in which the text constituting the first structured document is divided. The threshold for dividing the text is, for example, 0.2. In this case, if the similarity is smaller than 0.2, it is determined that the relevance between the preceding and following sentences is low, and the text is divided and structured before and after the sentence having the similarity of 0.2.

FIG. 12 is a diagram schematically showing how the text is divided. As shown in FIG. 12, the text constituting the first structured document is composed of, for example, five sentences, and the similarity between one sentence a and the next sentence b is 0.88. The similarity between one sentence b and one sentence c is 0.75. The similarity between one sentence c and one sentence d is 0.11. The similarity between one sentence d and one sentence e is 0.97.

In this case, the division unit 210 generates a second structured document divided and structured between the sentences c and d. The second structured document in which the text is divided into irrelevant sentences in this way can clarify the meaning of the divided group of texts.

As described above, the structured document creation device 200 according to the present embodiment vectorizes each of the sentences constituting the divided text constituting the first structured document to obtain the similarity before and after, and obtains the similarity before and after. And the threshold value set in the division definition property 51, and if the similarity is smaller than the threshold value, the division unit 210 that generates the second structured document in which the text is divided, the image management list 32, and the table management list 33. The second structured document is provided with an HTML conversion unit 260 for generating an HTML file in which an image and a table are inserted. This allows the text to be split into appropriate lengths.

(Modification example)
Not only the first structured document may be divided, but also the texts constituting the first structured document may be combined. The threshold value when combining texts is, for example, 0.8. In this case, sentences with a similarity greater than 0.8 are combined and structured as representing the same paragraph.

Each of the sentences constituting the divided text constituting the first structured document is vectorized to obtain the similarity before and after, and the similarity is compared with the first threshold value set in the division definition property to obtain the similarity. If is less than the first threshold, generate a second structured document that divides the text, or vectorize each of the bottom line sentence of the front text and the first sentence of the back text before and after. The similarity is obtained, the similarity is compared with the second threshold set in the division definition property, and if the similarity is larger than the second threshold, the second structured document in which the texts before and after the division are combined is combined. It is provided with a division / combination unit 211 (not shown) for generating the image and an HTML unit for generating an HTML file in which an image and a table are inserted in a second structured document with reference to an image management list and a table management list. This allows the text to be split into appropriate lengths.

(Evaluation results)
For the purpose of evaluating the degree of improvement in the work efficiency of the structured document by the structured document creating apparatus 100 according to the present embodiment, a comparison was made with the case where the structured document was manually created. We used two documents to be structured, the E company manual consisting of 327 pages and 29 files and the F company manual consisting of 649 pages and 4 files.

As shown in Table 1, the work time required to create a structured document could be reduced to about 1/6.

Next, Table 2 shows the results of evaluating the influence of the text division by the structured document creation device 200 according to the present embodiment on the search accuracy.

As shown in Table 2, according to the present embodiment, it is possible to speed up the work of searching and answering. When the structured document created by the structured document creating apparatus 200 according to the present implementation is used, for example, in the customer consultation room, the effect of shortening the response time to the customer and improving the customer satisfaction can be obtained.

As described above, the structured document creation device 100 of the present embodiment provides a highly versatile structured document creation device and a method thereof that can handle all types of documents as long as they are classified unstructured documents. can do. Further, according to the structured document creation device 200 of the present embodiment, the meaning of a structured group of texts can be clarified.

Although the above embodiment has been described with an example including the PDF generation unit 10, the PDF generation unit 10 may not be provided. The classified unstructured document may be directly input to the structured document creation devices 100 and 200. As described above, the present invention is not limited to the above-described embodiment, and many modifications can be made within the scope of the gist thereof.

10: PDF generation unit 20: Document structure acquisition unit 30: Management list generation unit 40: Text frame element arrangement unit 50: Text information structuring unit 51: Division definition property 60, 260: HTML conversion unit 210: Division unit 211: Division・ Joint

Claims (4)

A structured document creation device that creates a structured document described to fit a predetermined type from a classified unstructured document.
A document structure acquisition unit that acquires text frame elements, table frame elements, and image frame elements for each page of the unstructured document.
A management list generation unit that generates a management list that lists the positions of the text frame element, the table frame element, and the image frame element on the page in association with each other and their contents.
A text frame element array section that arranges the text frame elements in the writing order for each page with reference to the management list.
A text information structuring unit that generates a first structured document in which the arranged text frame elements are divided based on the rules set in the division definition property.
It is provided with an HTML conversion unit that generates an HTML file in which the table frame element and the image frame element are inserted into the first structured document with reference to the management list.
The rules include rules for identifying font size, character position, and heading.
The text information structuring unit adds a heading tag and a paragraph tag at a level corresponding to the character position and the font size to the text frame element.
A structured document creation device characterized by this. Each of the sentences constituting the divided text constituting the first structured document is vectorized to obtain the similarity before and after, and the similarity is compared with the threshold value set in the division definition property to obtain the similarity. Is smaller than the threshold value, the text is divided into a second structured document, which is provided with a division unit.
The HTML processing unit according to claim 1, wherein the HTML conversion unit generates an HTML file in which the table frame element and the image frame element are inserted also for the second structured document with reference to the management list. Structured document creation device. It is a structured document creation method performed by a structured document creation device.
A document structure acquisition step for acquiring text frame elements, table frame elements, and image frame elements for each page of a partitioned unstructured document.
A management list generation step for generating a management list in which the positions of the text frame element, the table frame element, and the image frame element on the page and their contents are associated with each other and listed.
A text frame element array step that arranges the text frame elements in writing order for each page with reference to the management list.
A text information structuring step that generates a first structured document in which the arranged text frame elements are divided according to a rule set in the division definition property.
With reference to the management list, the first structured document is subjected to the HTMLization step of generating the HTML file in which the table frame element and the image frame element are inserted, and the rule determines the font size, character position, and heading. Including rules to identify
The text information structuring step adds a heading tag and a paragraph tag at a level corresponding to the character position and the font size to the text frame element.
A structured document creation method characterized by this. Each of the sentences constituting the divided text constituting the first structured document is vectorized to obtain the similarity before and after, and the similarity is compared with the threshold value set in the division definition property to obtain the similarity. If is less than the threshold, the split step to generate a second structured document that splits the text, and
And
The structured document creation according to claim 3, wherein the HTMLization step generates an HTML file in which the table frame element and the image frame element are inserted also for the second structured document with reference to the management list. Method.

Images