TWI774105B - Official document analysis method - Google Patents

Abstract

An official document analysis method is provided. The official document analysis method includes: receiving an electronic official document with a word content, which includes a plurality of content areas; building at least one feature point of each of the content areas, wherein those feature points form a topology structure; reading an official document database with a plurality of official document format information, which includes an area deployment information and a plurality of title information; deciding the format of official document with a relative position score, computed by analyzing the topology structure with the area deployment information; building a blank official document with a standard format of the electronic official document, such blank official document includes a plurality of blank content areas, which separately corresponding to at least one title information; dividing the word content and writing said word content into the corresponding blank content area bases on the title information.

Description

Analysis method of official documents

This case is about the analysis method of the document content of a public document.

Official documents are referred to as official documents. Official documents can be documents used to exchange information between units, or documents used to save information for inspection. Official documents have different standard formats depending on their purpose. Taking the notification letter and the meeting record as an example, the two official documents need to convey or record different contents, so the presentation of the text content such as the participants, recipients, and explanatory text is also different.

In the process of document management, how to efficiently classify documents and ensure the correct content is a labor-intensive and important task. Due to the diversification of the types of official documents, the process of digitizing paper official documents mostly relies on personnel to proofread the paper documents one by one and type them word by word. This method is not only inefficient, but also prone to human error.

In view of this, the applicant proposes a method for parsing public documents. According to some embodiments, a method for parsing an official document includes receiving an electronic official document, the electronic official document includes text content, and the text content includes a plurality of content blocks. At least one feature point of each content block is established, and the feature points constitute a topology structure. Read the document type database, the document type database includes a plurality of document format information, and each of the document format information includes block structure information and a plurality of title information. Based on the topology and block structure information, the relative position score is determined. The format of the electronic document is judged according to the relative position score. A blank official document corresponding to the electronic official document format is established. The blank official document includes a plurality of blank content blocks, and each blank content block corresponds to at least one title information. According to the title information of the official document format information corresponding to the electronic official document, the text content is divided and written into the blank content block corresponding to each title information.

According to some embodiments, in the step of dividing the text content and writing it into the blank content block corresponding to the text content, it further comprises determining the text content of the electronic official document line by line, when it is determined that the text content of the Xth line contains the first title information , when the text content of the X+n line contains the second title information, and there is no other title information between the Xth line and the X+n line, the Xth line to the (X+n-1) line The text content is written into the blank content block corresponding to the first title information.

According to some embodiments, the text content further includes keywords. The document format information also includes keyword information. The official document parsing method determines keyword scores based on keywords and keyword information. The official document parsing method judges the format of the electronic document according to the relative position score and the keyword score.

According to some embodiments, the block structure information includes block structure weights, the keyword information includes keyword weights, the document parsing method applies a machine learning algorithm, and the format of the electronic document is determined according to the block structure weights and the keyword weights.

According to some embodiments, the official document parsing method further includes determining the difference between the topological structure of the electronic official document and the block structure information according to the document format information corresponding to the format of the electronic official document, and issuing a warning when there is a discrepancy. According to the official document format information corresponding to the format of the electronic official document, determine the difference between the keywords of the electronic official document and the keyword information, and issue a warning when there is a difference.

According to some embodiments, the official document parsing method further includes removing keywords whose keyword scores are lower than a score threshold in the electronic official documents.

According to some embodiments, the official document parsing method further includes determining the keyword score according to the content block in which the keyword is located.

According to some embodiments, the method for parsing official documents further includes removing characters without grammatical meanings in the text message.

According to some embodiments, the electronic document is generated by scanning the original document using optical character recognition.

According to some embodiments, the method for parsing the official document further includes writing the text content of the electronic official document into the blank content block of the blank official document and outputting it as a file in a physical format file.

To sum up, according to some embodiments, the official document parsing method analyzes the distribution of the content blocks of the official document and the keywords contained in each content block, so as to determine the type of the official document, and provide a warning when the text content or the format of the official document is wrong or correction.

The analysis method of official documents is a method of analyzing official documents. In terms of users of official documents, official documents may be, but not limited to, documents used by government agencies or units or documents used by private agencies or units. In terms of the functions of official documents, official documents can be, but are not limited to, documents used to communicate information or documents used to record information. According to some embodiments, the electronic device executes the document parsing method. The electronic device may be, but is not limited to, a personal computer, a mobile phone, a tablet computer, a server, and other devices. For example, the mobile phone converts the document image file captured by the camera of the mobile phone to generate an electronic document 1, and then the electronic document 1 is analyzed by the document parsing method executed on the mobile phone. For example, the official document image file captured by the camera of the mobile phone is sent to the server, and after the server converts the official document image file into electronic document 1, the electronic document 1 is analyzed by the official document parsing method executed on the server.

FIG. 1 is a schematic diagram of a notification letter according to some embodiments. Referring to FIG. 1 , the electronic document 1 includes text content 11 . For example, "Address: No. 0, OO Road, OO District, 33004OO City" or "Description: 1. OOOOOO...". The text content 11 can be divided into a plurality of content blocks 12 , so that the official document generates text clusters of several blocks. According to some embodiments, the content blocks 12 are distinguished from other content blocks 12 using wireframes. According to some embodiments, the content block 12 is distinguished from other content blocks 12 with a fixed margin. According to some embodiments, content blocks 12 utilize header information 14 to distinguish them from other content blocks 12 .

According to some embodiments, the official document parsing method utilizes image recognition to analyze the relative position distribution relationship of the content blocks 12 , so as to determine the type of the official document. FIG. 2 is a schematic diagram of performing image recognition according to a notification letter according to some embodiments. Please refer to FIG. 2 , the dotted area shown in FIG. 2 is the content block 12 of the notification letter in FIG. 1 . In this embodiment, a total of seven content blocks 12 are included. The official document parsing method establishes at least one feature point 121 of each of the content blocks 12 , and the feature points 121 form a topology structure 122 . The feature point 121 may be a feature used to identify the content block 12 in the image recognition algorithm. According to some embodiments, the feature point 121 is the geometric center point of the content block 12 . According to some embodiments, the feature points 121 are points on the edge of the content block 12 . According to some embodiments, the feature points 121 are corner points of the content block 12 . According to some embodiments, the feature points 121 are points on the textual content 11 contained in the content block 12 . For a single content block 12, the feature point 121 is not limited to one point or multiple points. According to some embodiments, the distance between the feature points 121 is extremely small, so that the feature points 121 are close to a contour line. The official document parsing method identifies the relative positional relationship of the feature points 121 , that is, the topology 122 formed by the feature points 121 . According to some embodiments, a topology structure of a plurality of vectors is formed between the feature points 121 , and the vectors form a feature matrix; the document parsing method identifies the feature matrix to determine the type of the document. 3 is a schematic diagram of a meeting minutes sheet according to some embodiments. Please refer to FIG. 3 , which is different from the embodiment of the notification letter in FIG. 1 , the embodiment of the meeting record sheet in FIG. 3 includes four content blocks 12 . FIG. 4 is a schematic diagram of performing image recognition according to a conference record sheet according to some embodiments. Please refer to FIG. 2 and FIG. 4 together. The image recognition algorithm utilizes the difference in the relative position distribution of the feature points 121 of the content blocks 12 of the two document types, and distinguishes the two as notification letters and meeting minutes.

The official document parsing method reads a document type database, the document type database includes a plurality of document format information, and each of the document format information includes a block structure information and a plurality of title information 14 . The official document format information includes information on various types of official documents, such as but not limited to notification letters, research records, meeting records, sign-in forms and other information related to the format or content of various official documents. The block structure information may include the format information of the official document or the information used to assist in identifying the format of the official document. According to some embodiments, the block structure information stores the topology 122 of the feature points 121 of the template document. According to some embodiments, the block structure information stores blank content blocks of the template document. For example, the block structure information stores the seven blank content blocks 12 of the notification letter shown in FIG. 2 , and the blank content blocks 12 correspond to the text content of the content block 12 shown in FIG. 1 respectively. 11. The title information 14 may be information used to distinguish the text content 11 and determine the content block 12 corresponding to the text content 11 . 5 is a schematic diagram of header information according to some embodiments. Taking the meeting minutes sheet in FIG. 5 as an example, the title information 14 includes words such as "record sheet", "meeting name", "attendees", and "meeting minutes". According to some embodiments, title information 14 may be used to differentiate textual content 11 . For example, the text content 11 between "meeting name" and "attendees" in Fig. 5 belongs to the same content block 12; the text content 11 between "attendees" and "meeting minutes" belongs to another content block Block 12. According to some embodiments, the title information 14 can be used to determine the content block 12 corresponding to the text content 11 . For example, in FIG. 5 , the “record sheet” corresponds to the topmost content block 12 ; the “meeting minutes” corresponds to the bottommost content block 12 .

According to some embodiments, the official document analysis method analyzes the relative positional relationship of the feature points 121 of the content block 12 of the electronic official document 1, and compares the relative positional relationship with the block structure information of various official document templates in the official document category database, Determine the relative position score of the similarity between the electronic document 1 and various document templates. The relative position score is the basis for judging the format of the electronic document 1 . For example, the official document parsing method receives the electronic document 1 in the embodiment of FIG. 3, analyzes the image and compares it with the official document in the notification letter, and obtains a relative position score of 40 points according to the preset scoring mechanism; After the template document of the document is compared, a relative position score of 90 points is obtained according to the preset scoring mechanism, so the electronic document 1 is classified as a meeting record sheet.

The official document parsing method confirms which official document format information the electronic official document 1 corresponds to in the official document category database, and establishes a blank official document in the format, and the blank official document includes a plurality of blank content blocks. The text content 11 includes a plurality of title information 14 , and each title information 14 may correspond to a specific content block 12 . According to the title information 14 , the official document parsing method divides and writes the text content 11 into the blank content block corresponding to the title information 14 . According to some embodiments, each document format information includes an algorithm specific to that document format. After confirming the format of the official document, the official document parsing method determines how to divide the text content 11 and how to write the divided text content 11 into each content block 12 according to the algorithm of the format. According to some embodiments, the algorithm divides the text content 11 of a certain number of characters after the title information 14 into the same content block 12 . According to some embodiments, the algorithm divides the entire line of text content 11 where the title information 14 is located into the same content block 12 . According to some embodiments, the algorithm divides the text content 11 between any two title information 14 into the same content block 12; When the text content 11 of the X line contains the first title information 14, the text content 11 of the Xth line and the text content 11 of the following line are written into the same content block 12 until it is determined that the text content 11 of the X+nth line contains the text content 11 of the Xth line. After the second title information 14, stop writing the text content 11 of the X+nth line and the text content 11 of the following lines into the same content block 12. The X and n are integer values.

For example, the official document in FIG. 5 is a paper official document, and the official document parsing method determines that the official document in FIG. 5 is a meeting minutes sheet and generates a blank document in the meeting minutes sheet format. The official document parsing method uses optical character recognition to scan a paper official document to obtain the text content 11 . Since the text content 11 generated by scanning is in a mixed state, it needs to be divided into different content blocks 12 . The official document analysis method reads the title information 14 of the official document format information of the meeting record sheet, and obtains the title information 14 as "record sheet", "meeting name", "attendees", and "meeting minutes". The algorithm will include the entire line of text 11 containing the "record sheet" into the first blank content block from top to bottom in the blank document; the algorithm will include the entire line of text 11 of the "meeting name" and the content of The text content 11 above the line of "people" is written into the second blank content block from top to bottom in the blank document; the algorithm will include the entire line of text content 11 for "attendees" and the text content containing "meeting minutes". The text content 11 above the line is written into the third blank content block from top to bottom in the blank document; the algorithm will include the entire line of text content 11 of "meeting minutes" and the next line containing "meeting minutes". The text content 11 below is written into the fourth blank content block from top to bottom in the blank official document.

According to some embodiments, the textual content 11 includes keywords 111 . The keyword 111 can be, but is not limited to, words with special meanings, words used to indicate the purpose of official documents, or words used to indicate the content of the document, such as "original", "letter", "subject", "telephone", "date of publication" "Wait. The keyword 111 may be, but is not limited to, vendor-defined, user-defined, or algorithm-generated. According to some embodiments, the document format information may include keyword information 13 . Depending on the format of the official document, it can correspond to different keyword information 13 . FIG. 6 is a schematic diagram of keyword information of a notification letter according to some embodiments. Referring to FIG. 6 , for example, the keyword information 13 of the notification letter includes “letter”, “subject”, “description”, “original”, “copy”, “address” and “telephone” of the keywords 111 , etc. . The keyword information 13 of the meeting record sheet may include keywords 111 "record sheet", "meeting name", "host" and "meeting record" (not shown). When the keyword 111 "address" appears in any electronic official document 1, the official document parsing method is based on the keyword information 13 of various official document format information and determines that the keyword 111 corresponds to the keyword score of the notification letter according to the preset scoring mechanism 60 points, and the keyword score corresponding to the meeting minutes is 40 points; when another keyword 111 "Subject" appears in the electronic document 1, the document parsing method is based on the keyword information 13 of various document format information and According to the preset scoring mechanism, it is determined that the keyword 111 corresponds to the keyword score of the notification letter as 80 points, and the keyword score corresponding to the meeting minutes is 30 points. According to some embodiments, the document parsing method determines the format of the electronic document 1 according to the keyword score and the relative position score.

According to some embodiments, the block structure information of each document format information includes the block structure weight, and the keyword information 13 of each document format information includes the keyword weight. According to some embodiments, the block structure weight is a weight value obtained by inputting the feature points 121 of the content block 12 of the template document as training data into the image recognition machine learning algorithm to learn the type of the template document. According to some embodiments, the keyword weight is a weight value obtained by inputting the keyword 111 of the text content 11 of the template document as training data into a text recognition machine learning algorithm to learn the type of the template document. According to some embodiments, the official document parsing method takes the content block 12 of the electronic document 1 as input and outputs the relative position score according to the trained machine learning algorithm; the official document parsing method uses the trained machine learning algorithm to The keyword 111 of the official document 1 is used as an input to output a keyword score, and the format of the electronic official document 1 is determined according to the keyword score and the relative position score. Depending on the applicable machine learning algorithm, according to some embodiments, block structure weights and keyword weights may be modified during the determination process.

According to some embodiments, after confirming the format of the electronic document 1, the official document parsing method determines the difference between the topological structure 122 of the electronic document 1 and the block structure information according to the block structure information or the keyword information 13 of the document format information, and When there is a difference, a warning is issued, and the difference between the keyword 111 of the electronic document 1 and the keyword information 13 is determined, and a warning is issued when there is a difference. FIG. 7 is a schematic diagram of a notification letter of content omission according to some embodiments. Please refer to the example of FIG. 7 . The official document parsing method determines that the electronic document 1 is relatively high according to the relative positional relationship of the content blocks 12 of the electronic document 1 in FIG. 7 . Close to the format of the notification letter, however the electronic document 1 contains the missing field 15. Compared with the template official document shown in Figure 1, the missing field 15 should be filled with the original or copy information. The topological structure 122 formed by the feature points 121 of the content block 12 of the electronic official document 1 is different from the block structure information, so the official document parsing method raises a warning that the content of the official document is wrong. FIG. 8 is a schematic diagram of a notification letter of incorrect content according to some embodiments. Please refer to the example of FIG. 8 . The official document parsing method determines the electronic document 1 according to the relative positional relationship of the content blocks 12 of the electronic document 1 in FIG. 8 . It is close to the format of the notification letter, but the document title "Meeting Minutes" of the electronic document 1 contains the wrong keyword 111 "Minutes". According to the keyword information 13 of the notification letter, the content block 12 should contain "letter" of the keyword 111 . The keyword 111 of the text content 11 of the electronic official document 1 is different from the keyword information 13, so the official document parsing method raises a warning that the content of the official document is wrong.

According to some embodiments, the document parsing method removes keywords 111 in the electronic document 1 whose keyword scores are lower than the score threshold. Scoring thresholds can be, but are not limited to, vendor-defined, user-defined, or algorithm-generated. For example, the official document parsing method includes a score threshold of 50 points. When the “subject” of the keyword 111 appears in the electronic document 1, the official document parsing method determines that the keyword 111 corresponds to the notification letter according to the keyword information 13 The keyword score is 80 points, and the keyword score corresponding to the meeting minutes is 30 points. If the final official document parsing method determines that the format of the electronic official document 1 is a meeting minutes sheet according to the relative position score of the electronic document document 1, because the keyword score of the keyword 111 corresponding to the “subject” of the meeting minutes sheet is lower than the scoring threshold , so remove it. According to some embodiments, the document parsing method determines the keyword score according to the content block 12 in which the keyword 111 is located. FIG. 9 is a schematic diagram of different content blocks at the same keyword position according to some embodiments. Please refer to FIG. 9. According to some embodiments, the machine learning algorithm learns the “note sheet” of the keyword 111 according to the template document of the meeting note sheet. ” mainly appears in the first content block 12 of the electronic document 1 from top to bottom. Therefore, when the "note sheet" of the keyword 111 of the electronic document is located in the first content block 12 from top to bottom, the keyword 111 corresponding to the keyword score of the meeting note sheet is 80 points. When the "note sheet" of the keyword 111 of the electronic document is located in the fourth content block 12 from the top to the bottom, the keyword 111 corresponds to the keyword score of the meeting note sheet with a score of 30 points. If the official document parsing method includes a score threshold of 50 points, the "record sheet" of the keyword 111 located in the fourth content block 12 from the top to the bottom will be removed. FIG. 10 is a schematic diagram of a meeting record sheet with incorrect content according to some embodiments. Please refer to FIG. 10. According to some embodiments, the “meeting time” of the keyword 111 is located in the second content block 12 from top to bottom. , the keyword 111 corresponds to a keyword score of 90 points in the meeting record sheet. When the "meeting time" of the keyword 111 of the electronic document is located in the third content block 12 from the top to the bottom, the keyword 111 corresponds to the keyword score of the meeting record sheet as 10 points. If the official document parsing method includes a score threshold of 50 points, the "meeting time" of the keyword 111 located in the third content block 12 from the top to the bottom will be removed. According to some embodiments, the official document parsing method removes the characters 112 that have no grammatical meaning in the text message. The ungrammatical characters 112 may be, but are not limited to, vendor-defined, user-defined, or algorithm-generated. For example, in the embodiment of FIG. 10 , the last line of the fourth content block 12 from top to bottom contains “&%$**/” of characters 112 with or without grammatical meaning. remove it.

According to some embodiments, the document parsing method utilizes optical character recognition to scan the original document to generate the electronic document 1 . The original document can be, but is not limited to, a paper document, a Portable Document Format (PDF) file or a file in various image file storage formats.

According to some embodiments, after the document parsing method writes the text content 11 of the electronic document 1 into the blank content block of the blank document, the completed official document is obtained, and the completed official document is output as a document entity format (Document Instance, DI). )file.

1: Electronic document 11: Text content 111: Keywords 112: Characters with no grammatical meaning 12: Content Blocks 121: Feature Points 122: Topology 13:Keyword Information 14: Title information 15: Missing field

[FIG. 1] is a schematic diagram of a notification letter according to some embodiments; [FIG. 2] is a schematic diagram of performing image recognition according to a notification letter according to some embodiments; [FIG. 3] is a schematic diagram of a meeting minutes sheet according to some embodiments; [FIG. 4] is a schematic diagram of performing image recognition according to a meeting record sheet according to some embodiments; [FIG. 5] is a schematic diagram of header information according to some embodiments; [Fig. 6] is a schematic diagram of keyword information of notification letters according to some embodiments; [FIG. 7] is a schematic diagram of a notification letter of content omission according to some embodiments; [FIG. 8] is a schematic diagram of a notification letter of incorrect content according to some embodiments; [FIG. 9] is a schematic diagram of different content blocks at the same key position according to some embodiments; and [FIG. 10] is a schematic diagram of a meeting record sheet with incorrect content according to some embodiments.

1: Electronic document

11: Text content

111: Keywords

12: Content Blocks

Claims (10)

A method for parsing official documents, including: receiving an electronic official document, the electronic official document includes a text content, and the text content includes a plurality of content blocks; establishing at least one feature point of each of the content blocks, and the feature points constitute a topology structure; reading a document type database, the document type database includes a plurality of document format information, and each of the document format information includes a block structure information and a plurality of title information; determining a relative position score according to the topology structure and the block structure information; Determine the format of the electronic official document according to the relative position score; creating a blank document corresponding to the electronic document format, the blank document includes a plurality of blank content blocks, each of the blank content blocks corresponding to at least one of the title information; and According to the title information of the document format information corresponding to the electronic document, the text content is divided and written into the blank content blocks corresponding to the title information. The method for parsing official documents as described in claim 1, in the step of dividing and writing the text content into the blank content block corresponding to the text content, further comprising: Determine the text content of the electronic official document line by line, when it is determined that the text content of the Xth line contains a first title information, the text content of the X+nth line contains a second title information, and the Xth line and the Xth When there is no other title information between the n lines, the text content between the Xth line and the (X+n-1)th line is written into the blank content block corresponding to the first title information. The method for parsing public documents as described in claim 1, wherein: The text content further includes a keyword; The official document format information further includes a keyword information; The official document parsing method determines a keyword score according to the keyword and the keyword information; and The official document parsing method determines the format of the electronic official document according to the relative position score and the keyword score. The official document parsing method as claimed in claim 3, wherein the block structure information includes a block structure weight, the keyword information includes a keyword weight, the official document parsing method applies a machine learning algorithm, and is based on The block structure weight and the keyword weight determine the format of the electronic official document. The method for parsing an official document according to claim 3, further comprising determining the difference between the topological structure of the electronic document and the block structure information according to the document format information corresponding to the format of the electronic document, and issuing a warning when there is a discrepancy ; and determine the difference between the keyword of the electronic official document and the keyword information, and issue a warning when there is a difference. The method for parsing official documents as described in claim 3, further comprising removing keywords in the electronic official document whose keyword scores are lower than a score threshold. The official document parsing method according to claim 6, further comprising determining the keyword score according to the content block where the keyword is located. The method for parsing official documents as described in claim 1, further comprising removing a character without grammatical meaning in the text content. The method for parsing an official document as claimed in claim 1, wherein the electronic official document is generated by scanning an original official document using optical character recognition. The method for parsing an official document according to claim 1, further comprising writing the text content of the electronic official document into the blank content block of the blank official document and outputting it as a file in a physical format file.

Images