Natural language processing of formatted documents

Patent metadata

Field	Value
Publication number	US-10628525-B2
Application number	US-201715597212-A
Country	US
Kind code	B2
Filing date	May 17, 2017
Priority date	May 17, 2017
Publication date	Apr 21, 2020
Grant date	Apr 21, 2020

Abstract

Detecting and incorporating formatting characteristics within natural language processing analytics. Source documents are ingested and the markup formatting language is identified by the program. Once identified, the markup language is parsed and examined for formatting characteristics, embedded notes, comments and other metadata. The formatting characteristics of the plain text are extracted, along with the plain text, and converted into a common analysis structure (CAS), or CAS-equivalent structure, which annotates the natural language text together with its respective formatting characteristics. The CAS or CAS-equivalent structures are stored and sent to a natural language processing pipeline for further analysis via complex algorithms and rules. The natural language processing results data are curated to reflect meaningful analysis of the extracted CAS or CAS-equivalent structure.

First claim

The invention claimed is:

1. A computer-implemented method for processing text, the method comprising:
determining, by a computer, that a span of natural language text is associated with one or more formatting characteristics;
applying, by the computer, optical character recognition (OCR) to the span of natural language text associated with the one or more formatting characteristics, wherein
identifying, by the computer, that the span of the natural language text is bold type by comparing the pixel thickness of the characters of the span of the natural language text to an average pixel thickness of the natural language text;
identifying, by the computer, that the span of the natural language text is italics type by analyzing the angle of the pixels of the characters in the span of the natural language text;
identifying, by the computer, that the span of the natural language text is underlined by analyzing the number of pixels in a consistent horizontal line underneath the characters in the span of the natural language text;
identifying, by the computer, that the span of the natural language text is a subscript by recognizing that a numerical character is located slightly below the span of the natural language text;

calculating, by the computer, integer offsets denoting the span of natural language text modified by the identified formatting characteristics;
applying, by the computer, numerical integer offsets denoting a beginning and an end of the identified formatting characteristics;
denoting, by the computer, a page number as well as a beginning character and an end character of the span of natural language text that is modified;
converting, by the computer, the span of natural language text, denoted by the numerical integer offsets at the beginning and the end of the identified formatting characteristics, into a common analysis structure (CAS);
generating, by the computer, a data structure for storage in memory comprising at least one of the one or more formatting characteristics, and a corresponding span of the natural language text, wherein the data structure comprises the CAS;
appending, by the computer, the CAS to a CAS file;
transmitting, by the computer, the generated CAS comprising the at least one of the one or more formatting characteristics and the corresponding span of the natural language text to a natural language processing (NLP) pipeline to identify an intent of the corresponding span of the natural language text;
performing, by the computer, the one or more actions associated with the one or more formatting characteristics;
associating, by the computer, a level of importance with the corresponding span of the natural language text based on the at least one of the one or more formatting characteristics, wherein the level of importance for the at least one of the one or more formatting characteristics is pre-configured;
ranking, by the computer, the one or more actions associated with the identified one or more formatting characteristics, wherein bold type is ranked as more important than underline text; and
incorporating, by the computer, the generated CAS data structure into a machine learning model that learns from and makes predictions on natural language text data.

2. The method of claim 1 , wherein the NLP pipeline comprises a Question and Answer (QA) pipeline, and wherein the input text comprises a question, the QA pipeline analyzing the question based on the at least one of the one or more formatting characteristics and the corresponding span of the natural language text in the question.

3. The method of claim 1 , wherein the NLP pipeline comprises a relationship extraction pipeline, the relationship extraction pipeline incorporating the generated data structure into its detection and classification of semantic relationships within a natural language text analysis.

4. The method of claim 1 , wherein the NLP pipeline comprises a syntax tree parsing pipeline, the syntax tree parsing pipeline incorporating the generated data structure into its construction of parse trees for sentences in natural language text.

5. The method of claim 1 , wherein the NLP pipeline comprises a text mining pipeline, the text mining pipeline incorporating the generated data structure into its text mining analytics.

6. The method of claim 1 , wherein determining, by a computer, that a natural language text is associated with one or more formatting characteristics comprises:
determining, by the computer, that the natural language text is structured; and
identifying the one or more formatting characteristics by identifying formatting meta tags associated with the natural language text.

7. The method of claim 1 , further comprising:
associating, by the computer, the one or more formatting characteristics with one or more actions, the one or more actions comprising any one of:
categorizing as irrelevant the span of natural language text associated with a strikethrough formatting characteristic;
emphasizing the span of natural language text associated with an underline formatting characteristic;
emphasizing the span of natural language text associated with a bold type formatting characteristic;
emphasizing the span of natural language text associated with an italics formatting characteristic;
categorizing as a chemical formula the span of natural language text associated with a subscript formatting characteristic; and
categorizing as a mathematical formula the span of natural language text associated with a superscript formatting characteristic.

8. The method of claim 7 , further comprising:
ranking, by the computer, the action associated with the identified at least one formatting characteristic.

9. A computer program product for processing text, comprising a non-transitory tangible storage device having program code embodied therewith, the program code executable by a processor of a computer to perform a method, the method comprising:
determining, by a processor, that a span of natural language text is associated with one or more formatting characteristics;
applying, by the processor, optical character recognition (OCR) to the span of natural language text associated with the one or more formatting characteristics, wherein
identifying, by the processor, that the span of the natural language text is bold type by comparing the pixel thickness of the characters of the span of the natural language text to an average pixel thickness of the natural language text;
identifying, by the processor, that the span of the natural language text is italics type by analyzing the angle of the pixels of the characters in the span of the natural language text;
identifying, by the processor, that the span of the natural language text is underlined by analyzing the number of pixels in a consistent horizontal line underneath the characters in the span of the natural language text;
identifying, by the processor, that the span of the natural language text is a subscript by recognizing that a numerical character is located slightly below the span of the natural language text;

calculating, by the processor, integer offsets denoting the span of natural language text modified by the identified formatting characteristics;
applying, by the processor, numerical integer offsets denoting a beginning and an end of the identified formatting characteristics;
denoting, by the processor, a page number as well as a beginning character and an end character of the span of natural language text that is modified;
converting, by the processor, the span of natural language text, denoted by the numerical integer offsets at the beginning and the end of the identified formatting characteristics, into a common analysis structure (CAS);
generatin

Assignees

• IBM

Inventors

• Fink Patrick W
• Mcneil Kristin E
• Parker Philip E
• Werts David B

Classifications

• G06F40/211

  Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars · CPC title

• G06F40/109

  Font handling; Temporal or kinetic typography · CPC title

• G06K9/00442

  Physics · mapped topic

• G06N20/00 Primary

  Machine learning · CPC title

• G06F17/214

  Physics · mapped topic