HK1095902B

HK1095902B - Document information mining tool

Info

Publication number: HK1095902B
Application number: HK07103281.5A
Authority: HK
Inventors: L. Hadley Brent; J. Eames Patrick; C. Chew Susan
Original assignee: The Boeing Company
Priority date: 2004-04-30
Filing date: 2005-04-29
Publication date: 2015-07-17

Description

The present invention relates to a method for extracting referential keys from a document.

The present invention also relates to a system for extracting referential keys from a document, wherein the system comprises a parser configured to parse a document to identify at least one key, the key being identified from at least one indicator.

Finally, the present invention relates to a computer-readable medium having stored thereon instructions for extracting referential keys from a document.

Such a method and such a system are known from US 6,424,982 .

FIELD OF THE INVENTION

This invention relates generally to information extraction and, more specifically, to processing documents to enhance access to information stored in the documents.

BACKGROUND OF THE INVENTION

People often need to access information that is recorded in documents. Such documents may range in length from a single page to many volumes. Certainly, the longer the documents are, the more difficult it is to access the specific information desired. As a result, long or complex documents include multiple tables and indices to facilitate hierarchical and key word searches for the information of interest. These tables and indices are time-consuming to create. Furthermore, although the tables and indices are helpful in locating information, using them is time-consuming since each page referred to in the table or index must be individually examined to find the entry of interest. The foregoing may be repeated numerous times before the substantive entry of interest is found.

The proliferation of computers has revolutionized how information is accessed. Accordingly, computer-readable documents may now be searched using various software routines for terms of interest in the document. In particular, hyper-text linking has enabled referencing and cross-referencing of key terms simply by using a pointing device to select a point of interest in a document.

Despite the great advantages that computers permit in accessing and retrieving information, processes for information retrieval still can be improved. For example, creating a hyper-link to a source document involves human intervention to identify the term with which the link will be created and associating it with the link to the related information. Further, to create hyper-linked information, all the documents or at least those documents containing the links need to be computer-readable documents. As a result, a non-computer-readable document may be scanned so that it is computer-viewable, but unless the document is computer-readable such as a text or graphics document, it is generally not possible to associate links with portions of the document.

Similarly, even though a referenced document need not be computer-readable to be accessed from a link, if the reference target is not computer-readable, then a person linking to the document may be required to manually navigate through the target document to find the information of interest. Certainly, the task becomes even more complicated if one desires information in both documents and needs to switch back and forth between the documents. In such cases, to avoid the difficulty of navigating back and forth, a user desiring portions of such documents will typically print the needed documents or parts of the documents. When users print such documents instead of accessing them on the computer clearly undermines one of the objectives of making such documents accessible by computer.

To avoid the complexities of moving back and forth between documents, one possible is to extract information from documents that is expected to be relevant. Unfortunately, removing only the content from the documents may present other problems. For example, some regulatory agencies require that extracted content be verified as accurately including the content of the original document before it can be used. This verification is a time-consuming and costly process. In addition, extraction of content also may obliterate inferential information a user might otherwise obtain from the document. Such inferential information might exist as an interrelationship between parts, or as an annotation regarding other parts that might be useful, and other similarly useful information. Extracting information expected to be relevant thus may obliterate other useful information.

US 6,424,982 , mentioned at the outset, discloses a parsing system and method in which the break characters in a document are used to rapidly parse the document and extract one or more key phrases from the document which characterize the document. The break characters in the document may include explicit break characters, such as punctuation, soft stop words and hard stop words. The determination of which phrases in the document are extracted, depends upon the type of break character appearing after the phrase in the document.

US 2002/0065857 discloses a system and a method for analyzing and clustering of documents for a search engine. By incorporating computational intelligence and statistical methods, the document information is analyzed and clustered using novel techniques for knowledge extraction. A comprehensive dictionary is built based on the key words identified by these techniques from the entire text of the document. A document is identified by the knowledge that is represented in its contents. Based on such knowledge extracted from all the documents, the documents are clustered into meaningful groups in a catalogue tree.

"Preparations for Semantics-Based XML Mining" by Jung-Won Lee, et al., Data Mining, 2001. ICDM 2001, Proceedings IEEE International Conference on San Jose, CA, USA 29 Nov.-2 Dec. 2001, Los Alamitos, CA, USA, IEEE Comput. Soc, US, 29 November 2001, pages 345-352, XP010583296, ISBN: 0-7695-1119-8, proposes a new methodology for preparing XML documents for quantitative determination of similarity between XML documents by taking account of XML semantics. An accurate quantitative determination of similarity between XML documents is said to provide an important basis for a variety of applications of XML documents mining and processing. The disclosed methodology is supposed to bring a significant improvement over a traditional vector-space model.

"Adaptive Automatic Classification on the Web" by Jenkins C. et al., Database And Expert Systems Applications, 2000. Proceedings. 11th International Workshop on 4-8 September 2000, Piscataway, NJ, USA, IEEE, 4 September 2000, pages 504-511, XP010515542, ISBN: 0-7695-0680-1, addresses the automatic classification of documents on the web. Experimental software tools are introduced which enable the automatic generation of vocabularies, representing any given classification scheme using pre-classified documents. An internal representation of the required classification hierarchy is automatically generated and subsequent documents are filtered through this hierarchy and automatically classified. An environment is created in which different collections of documents can be classified according to different, extensible classification schemes. Evolving, changing vocabularies and hierarchies can be dynamically generated and modified over time.

"Automatic RDF metadata generation for resource discovery" by Jenkins C. et al., Elsevier Science Publishers B.V., Amsterdam, NL, Volume 31, No. 11-16, pages 1305-1320, 17 May 1999 addresses automatic metadata generation of HTML documents according to Dewey Decimal Classification through an automatic classifier which can be used to extract context sensitive metadata which is then represented using research description framework data. The document further describes the process of automatic classification, wherein an appropriate metadata element set is identified comprising those elements that can be extracted during classification.

It is therefore an object of the present invention to provide a system that improves the recognition, the storing and the retrieving of data that is present in a document.

A solution according to the present invention is given by the method as mentioned in claim 1 and, further, the method of claim 18.

A further solution according to the present invention is given by the system as mentioned in claim 10.

Yet another solution according to the present invention is given by the computer-readable medium as mentioned in claim 9.

SUMMARY OF THE INVENTION

The various embodiments of the present invention are useful in mining documents to identify keys that facilitate the creation of references within the documents and that may also form a basis for links between documents. The various embodiments exploit the fact that documents of certain types adhere to formatting standards that may include titles, section headings, and other parts of the document that are useful for navigating two and within the document. The various embodiments may also utilize pattern formats that may be present in documents of various types, such as character patterns that may include a number of characters having expected separators so that keys may be identified when that pattern appears on the page. One these parts of the document representing potentially useful navigational keys have been identified, the keys are stored in a structured format associated with the original document to facilitate linking the keys with related documents. In one non-limiting example, embodiments of the present invention are used to mine a .pdf type document to find textual and graphical keys in the document, using optical character recognition as needed to extract text from graphically-presented keys, and to store the keys in an extensible markup language (XML) document to facilitate creating links to the keys in the original .pdf document.

More particularly, embodiments of the present invention provide methods, computer-readable media, and systems for extracting referential keys from a document. A document is parsed to identify at least one key, the key being identified from at least one contextual indication. The key is classified according to a key type, the key type being identified from the contextual indication. The key is extracted and then stored in a location in a structured shell with the location corresponding to the key type. As a result, the key can be found by a search seeking one of the key and the key-type allowing a searcher to identify the document from which the key was extracted.

BRIEF DESCRIPTION OF THE DRAWINGS

The preferred and alternative embodiments of the present invention are described in detail below with reference to the following drawings.

FIGURE 1 is a block representation of an embodiment of the present invention.
FIGURE 2 is a document to be mined according to an embodiment of the present invention.
FIGURE 3 is a flowchart of a routine for mining a document according to an embodiment of the present invention.
FIGURES 4-7 are examples of portions of documents mined to extract keys into a structured format.
FIGURE 8 is a structured format collecting a collection of keys mined from a document.
FIGURE 9 is a system according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE INVENTION

The present invention relates generally to a system and method for linking related documents and, more specifically, to identifying, extracting, and collecting keys used in linking the related documents. Many specific details of certain embodiments of the invention are set forth in the following description and in FIGURES 1-9 to provide a thorough understanding of such embodiments. One skilled in the art, however, will understand that the present invention may have additional embodiments, or that the present invention may be practiced without several of the details described in the following description.

Using prior knowledge of the formats employed by authors in creating documents, for a given document format (e.g., .pdf files) and document type (e.g., aircraft maintenance manuals), the document information mining tool reviews the document for elements an author of the document has included that can be used to navigate to and within the document. For example, the document mining tool identifies titles, headers, footers, page numbers, off-page references, annotations to graphics, and other authored elements that a human reader of the document might use to navigate for additional information. The document mining tool identifies these navigational attributes automatically by dissecting the document. The information mined is then stored using a suitable structured format, such as an XML document, allowing the mined information to be used as keys or tags to support linking to other documents. The document mining tool is further suitably equipped with an optical character reading (OCR) capability to analyze text that appears in graphic form in order to mine it from the document.

The document information mining tool thus provides a low-level, literal extraction process by parsing a document, identifying sections of the document that can be used as navigational points to navigate to, from, and throughout the document, and storing the mined information in a useful structure.

FIGURE 1 is a block diagram of a generalized system 100 for mining a source document 110 according to an embodiment of the invention. The document 110 is subjected to examination according to mining rules 120 directed to identify and extract various navigational keys, as will be described in greater detail below. The extracted keys are subsequently stored in a structured format 130. The structured format 130 may include an XML document, although it will be appreciated that the extracted keys may also be stored in a substantive content-based document. In one particular example, the keys are stored in a structured format used to facilitate navigation to and within the source document. In still another particular example,

The mined information is used to navigate through the existing documents to facilitate access to the complete, underlying documents.

Referring still to FIGURE 1, in another particular example, keys may be created by a user and transferred to the mining rules 120. The source document 110 may then be parsed to locate at least one portion of the document 110 that includes a contextual indication corresponding to the key created by the user. The user may then review the information located within the at least one portion identified during the parsing.

FIGURE 2 shows a sample document 200, which is a page selected from an aircraft maintenance manual (AMM). The document 200 is formatted as a .pdf document and presented on a .pdf reader 210. It will be appreciated that the document 200 could have been generated from a computer-readable document or scanned from a printed document. For purposes of this description, a computer-readable document is one in which characters and figures are generated for display on a computer screen from character and figure representations. In contrast, non-computer-readable documents are documents received in image form because they were scanned into or otherwise stored in a computing system in an image format stored such as rasterized data. The sample document 200 represents a document that will be mined to extract the keys in the document, which may include titles, task designations, and other information, that might be useful in navigating to, from, and within the document 200.

FIGURE 3 is a generalized flowchart of a method300 for mining a document such as the sample document 200 (FIGURE 2) according to an embodiment of the invention. The routine 300 begins at a block 302. At a block 304, the document to be mined is received. As previously described, the document may be a computer-readable document or non-computer-readable image-type document. At a block 306, a next area of a document is parsed. The exemplary routine 300 parses a document area by area. However, alternatively, the document could be parsed looking for particular features of the document as will be further described below.

At a decision block 310 it is determined if a placement indicator is identified. A placement indicator may indicate presence of a title, a header, footer, a page number, or another key having relevance as indicated by the position of the key on a page. Accordingly, the manner in which the key is detected may be based upon generalized knowledge of the formatting methods used in a document, or upon tribal knowledge of the formatting methods, or other specific document formatting rules. Generalized knowledge, for example, includes knowledge that indicates that a series of numerals centered or in a corner of a header or footer of a series of pages represent page numbers. General knowledge also may include, for example, that a term appearing at a top of a page generally includes a title. By contrast, tribal knowledge includes knowledge known to a group of persons familiar with a specific type of document. For example, in certain industries or organizations, it may be common practice to place a subject identifier in a header or footer of each page, even though the inclusion of that key may not be indicated by a specific rule. Finally, specific document formatting rules are rules literally dictated by document conventions. For example, it may be an established practice in an industry or organization to position system identifiers on a page of a document in a generally similar location. For example, the format used in legal briefs submitted to courts include titles, page numbers, and other elements placed in generally similar locations according to rules specified by the courts.

Still referring to FIGURE 3, once a placement indicator is identified at the decision block 310, the routine 300 proceeds to a decision block 312 where it is determined if the placement is determinative of what type of key has been identified. One area of a document could include a placement-indicated key, a format-indicated key, and a font-indicated key, such as, for example, if footer includes both a page number and title or other identifier. The position of a numeral alone, such as in a part of a header reserved for a particular identifier may alone be determinative. On the other hand, other aspects of the key may have to be considered. If it is determined at the decision block 312 that placement is determinative, the routine 300 proceeds to a block 314 where the key is extracted and stored in a structured format. If the document is a computer-readable document, characters representing the key are copied. On the other hand, if the document is a non-computer-readable image-type document as previously described, an optical character recognition (OCR) routine is invoked to scan the identified key to extract the characters from the key.

If a placement indicator is not identified or placement alone is not determinative, the routine proceeds to a decision block 316 where it is determined if a format indicator is identified. A format-indicated key may include a string of digits identifiable because of their format. For example, a number including three numbers followed by a separator such as a dash, followed by three more numbers, another dash, and four more numbers might be expected to be a phone number. A number including three numbers followed by a dash, followed by two more numbers followed by another dash, followed by four more numbers might be expected to be a Social Security number. In a further example, a string of characters followed by an "@" symbol followed by another string of characters including at least one period might be expected to be an e-mail address. Accordingly, a type of key can be determined based upon the format.

Once a format indicator is identified at the decision block 316, the routine 300 proceeds to a decision block 318 where it is determined if the format is determinative of what type of key has been identified by itself or in combination with the placement indicator. If it is determined at the decision block 318 that format alone or with the placement is determinative, the routine 300 proceeds to a block 314 where the key is extracted and stored in a structured format. If the document is a computer-readable document, characters representing the key are copied. On the other hand, if the document is a non-computer-readable image-type document as previously described, an optical character recognition (OCR) routine is invoked to scan the identified key to extract the characters from the key.

If placement or format indicators are not identified, or they are not determinative alone, the routine proceeds to a decision block 320 where it is determined if a font indicator is identified. A font-indicated key may include a key having a type signified by its font or typeface. For example, according to commonly accepted document conventions, information that is underlined, or that appears in bold type, or is presented in an enlarged font, or any combination of the foregoing may represent a title, a part name, a document identifier, or similar information.

Once a font indicator is identified at the decision block 320, the routine 300 proceeds to a decision block 322 where it is determined if the font is determinative of what type of key has been identified by itself or in combination with other indicators. If it is determined at the decision block 322 that format alone or in combination with other indicators is determinative, the routine 300 proceeds to a block 314 where the key is extracted and stored in a structured format. If the document is a computer-readable document, characters representing the key are copied. On the other hand, if the document is a non-computer-readable image-type document as previously described, an optical character recognition (OCR) routine is invoked to scan the identified key to extract the characters from the key.

If indicators are not identified or are not determinative, the routine proceeds to a decision block 326 where it is determined if all areas of the document have been parsed. If not, the routine loops to the block 306 where the next area of the document will be parsed. On the other hand, if all areas of the document have been parsed, the routine 300 ends at a block 328.

Although FIGURE 3 shows a particular sequential ordering of the steps 310, 316 and 320, it will be appreciated that the order of the steps 310, 316 and 320 may be altered to suit a particular application. Moreover, the steps 310, 316 and 320 may also be executed in parallel, so that keys based on placement, format and font may be identified simultaneously.

FIGURES 4-7 illustrate examples of how keys of interest are mined from a document. The examples 400, 500, 600, and 700 from FIGURES 4, 5, 6, and 7, respectively show examples of keys being extracted from a page of an aircraft maintenance manual (AMM). The keys are extracted and then stored in a structured document that, in these examples, are XML documents.

Example 400 (FIGURE 4) shows a key being extracted from a document segment 402. In particular, the key includes a section title, "INTEGRATED DRIVE GENERATOR (IDG) - REMOVAL/INSTALLATION." The key is identified by rules 404 recognizing a combination of indicators using general, tribal, and/or other knowledge. In terms of placement, the potential key begins at a left side of a page and is a first line of a page body. In terms of font, the potential key is all UPPERCASE. The potential key is thus identified as a section title and identified as a key of interest. The key is extracted and stored in a structured format 406, in this case an XML document, where the key is designated as a title so that if can be located for navigation purposes.

Example 500 (FIGURE 5) shows a second key being extracted from a document segment 502. The key includes a task identifier, an expected and sought-after key in an AMM. The task identifier is determined in this case by format indicators. The task identifier includes a string of two digits, two digits, two digits, three digits, and three digits all separated by dashes. Accordingly, a rule 504 identifies this format as a task identifier, which is thus identified as a key. The key is extracted and stored in an XML document 506 where the key is designated as a task number so that it can be located for navigation purposes.

Example 600 (FIGURE 6) shows a third key being extracted from a document segment 602. The key includes a chapter number (24-11-11), which may be another expected and sought-after key in an AMM. The chapter number is determined in this case by placement and format indicators. In the present example. the chapter number is always on a right side of a page, and is always in the form of two digits, two digits, and two digits separated by dashes. The present example also includes a page block (401) and a revision date (Jun 10/01) that follow the formatted chapter number. Again, a rule 604 identifies the placement and format indicators in the document segment 602 to identify the key. The key is then extracted and stored in an XML document 606 where the key is designated as a chapter number so that it can be located for navigation purposes. It will be appreciated that upon finding a formatted key such as the chapter number, the chapter number maybe further subdivided and stored in its constituent parts, including chapter (24), section (11), subject (11), and page block (401).

Example 700 (FIGURE 7) shows a fourth key being extracted from a document segment 702. The key includes a particular substantive section that, in this case, is "Effectivity" that is used to designate the applicability of a selected portion of the AMM to a particular aircraft. The substantive section is determined in this case by placement and format indicators. In the present example, the substantive section of interest is always on a left side of a page (a placement indicator) and always follows the term "Effectivity." As in the previous example, a rule 704 identifies the potential key as the substantive description "Effectivity" and identifies it as a key of interest. The key is extracted and stored in an XML document 706 where the key is designated as "Effectivity" so that it can be located for navigation purposes.

FIGURE 8 shows an XML document 800 that collects the extracted keys. The XML document 800 shows a page number 802, a chapter number 804 subdivided into its constituent parts, a title 806, a substantive section 808 (Effectivity), and a task identifier 810. The keys 802, 804, 806, 808, and 810 all are in a computer-readable format and are associated with the original document from which the keys 802, 804, 806, 808, and 810 wereextracted. Thus, the XML document 800 can be searched or otherwise accessed to find the keys 802, 804, 806, 808, and 810 of interest. The keys 802, 804, 806, 808, and 810 are in turn linked to the document 800 from which they were drawn. Thus, the mining tool facilitates access to the document of interest and the portion of interest.

FIGURE 9 is a block diagram of an exemplary system 900 according to an embodiment of the present invention. The system 900 is presented with a computer-readable or non-computer-readable document 910. The system 900 uses a parser 920 to scan the document for potential keys as previously described. The system also employs an identifier 930 that is programmed to use placement indicators, format indicators, font indicators, and other indicators to determine whether a potential key found by the parser is a key of interest. If it is a key of interest, a classifier 930 receives the key, classifies it according to indicators manifested by the key of interest, and stores the key in a structured format 950, such as in XML data. As previously described in connection with FIGURE 8, the XML data may then be used to navigate to, from, and within the document 910.

While preferred and alternate embodiments of the invention have been illustrated and described, many changes can be made without departing from the spirit and scope of the invention. Accordingly, the scope of the invention is not limited by the disclosure of the preferred and alternate embodiments. Instead, the invention should be determined entirely by reference to the claims that follow.

Claims

Method for extracting referential keys from a document (910), the method comprising:
parsing (306) the document (910) to identify at least one key, the key being identified from at least one indicator;

classifying the key according to a key type, the key type being identified from the at least one indicator;

extracting the key; and

storing (314) the key in a location in a structured shell (950), the location corresponding to the key type such that the key can be found by a search seeking at least one of the key and the key-type allowing a searcher to identify the document (910) from which the key was extracted;

characterized in that

the document (910) is a .pdf document;

the step of extracting the key comprises using optical character recognition to extract the key when the document (910) includes non-computer-readable content; and

the indicator includes at least one of a placement indicator, a format indicator, and a font indicator.
Method according to claim 1, wherein the key type includes at least one of a title, a header, a footer, a document type, a document identifier, a subject identifier, a section identifier, a chapter identifier and a page number.
Method according to claim 1, wherein the indicator is derived from knowledge including at least one of a general knowledge, a tribal knowledge, and a specific document formatting rule.
Method according to claim 1, wherein the placement indicator includes a page position.
Method according to claim 1, wherein the format indicator includes a character pattern including at least one of a pattern of digits and a pattern of separators.
Method according to claim 1, wherein the font indicator includes at least one of a typeface, a boldface, an underlined text portion, and a font size.
Method according to claim 1, wherein extracting the key includes copying the key.
Method according to claim 1, wherein the structured shell (950) includes an extensible markup language (XML) document.
Computer-readable medium having stored thereon instructions for extracting referential keys from a document (910), the computer-readable medium comprising computer program portions adapted to perform the method according to any of claims 1 to 8.
System (900) for extracting referential keys from a document (910), the system (900) comprising:
a parser (920) configured to parse the document (910) to identify at least one key, the key being identified from at least one indicator;

an identifier (930) configured to identify the key according to a key type, the key type being identified from the at least one indicator; and

a classifier (940) configured to extract the key and store the key in a location in a structured shell (950), the location corresponding to the key type such that the key can be found by a search seeking one of the key and the key-type allowing a searcher to identify the document (910) from which the key was extracted,

characterized in that

the document (910) is a .pdf document;

the system further comprises an optical character recognizer configured to use optical character recognition to extract the key when the document (910) includes non-computer-readable content; and

the indicator includes at least one of a placement indicator, a format indicator, and a font indicator.
System according to claim 10, wherein the key type includes at least one of a title, a header, a footer, a document type, a document identifier, a subject identifier, a section identifier, a chapter identifier and a page number.
System according to claim 10, wherein the indicator is derived from knowledge including at least one of a general knowledge, a tribal knowledge, and a specific document formatting rule.
System according to claim 10, wherein the placement indicator includes a page position.
System according to claim 10, wherein the format indicator includes a character pattern including at least one of a pattern of digits and a pattern of separators.
System according to claim 10, wherein the font indicator includes at least one of a typeface, a boldface, an underlined text portion, and a font size.
System according to claim 10, wherein extracting the key includes copying the key.
System according to claim 10, wherein the structured shell (950) includes an extensible markup language (XML) document.
A method of information searching in a document (910), comprising:
creating a reference key corresponding to at least one contextual indicator present in the document (910);

parsing (306) successive portions of the document (910);

identifying at least one portion of the document (910) that includes the reference key; and

reviewing the information included in the at least one portion,

characterized in that

the document (910) is a .pdf document;

the step of identifying the at least one portion comprises using optical character recognition to extract the key when the document (910) includes non-computer-readable content; and

the indicator comprises at least one of a placement indicator, a format indicator, and a font indicator.
The method of claim 18, wherein creating a reference key further comprises creating the key according to a predetermined key type.
The method of claim 18, wherein the key type comprises at least one of a title, a header, a footer, a document type, a document identifier, a subject identifier, a section identifier, a chapter identifier and a page number.
The method of claim 18, wherein the format indicator comprises a character pattern that includes at least one of a pattern of digits and a pattern of separators.
The method of claim 18, wherein the font indicator comprises at least one of a typeface, a boldface, an underlined text portion and a font size.