Input entity identification from natural language text information

• US 9,817,814 B2
• Filed: 02/23/2016
• Issued: 11/14/2017
• Est. Priority Date: 12/31/2015
• Status: Active Grant

• Current Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Original Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Inventors: Misra, Janardan, Dubash, Neville, Podder, Sanjay, Rawat, Divya
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US15/050,908
• Publication Number:
  US 20170192958A1
• Time in Patent Office: 630 Days
• Field of Search: 707736, 707769, 707917, 707955, 707999002-999004, 715210, 715256, 715230, 715255, 715708, 704 8, 704 9, 704 4, 704 10, 704 6, 704257
• CPC Class Codes: G06F 16/313 Selection or weighting of terms for indexing
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3331 Query processing
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A device may include one or more processors. The device may receive text to be processed to identify input entities included in the text. The device may identify text sections of the text. The device may generate a list of terms included in the text sections of the text. The device may perform one or more feature extraction techniques, on the terms included in the text sections, to identify the input entities included in the text. The device may generate information that identifies the input entities included in the text, based on performing the one or more feature extraction techniques. The device may provide the information that identifies the input entities included in the text.

Claims

1. A device, comprising:

one or more processors to;

receive text to be processed to identify input entities included in the text;

identify text sections of the text;

generate a list of terms included in the text sections of the text;

perform a feature extraction technique to determine whether each term, in the list of the terms, is an object of an action,

the feature extraction technique causing a subject-predicate-object relationship to be extracted from one of the text sections;

perform one or more other feature extraction techniques, on the terms included in the text sections, to identify the input entities included in the text,

the one or more other feature extraction techniques including at least one of;

a technique to determine tag patterns for the terms based on tags associated with the terms,

a technique to determine whether the terms are capitalized,

a technique to determine a headword in each term than includes multiple words,

a technique to determine a number of constituent words in each term,

a technique to determine semantic similarities of input type actions acting on the terms,

a technique to determine semantic similarities of non-input type actions acting on the terms,

 

the non-input type actions acting on the terms being action terms that are not in the list of terms,

a technique to determine a distance of each term from an action appearing in a same text section associated with each term, or

a technique to identify surrounding context for each term and associated tags with the surrounding context;

generate information that identifies the input entities included in the text, based on performing the feature extraction technique and the one or more other feature extraction techniques; and

provide the information that identifies the input entities included in the text.

2. The device of claim 1, where the one or more processors, when performing the one or more other feature extraction techniques, are to:

generate a term co-occurrence matrix based on the terms and the text sections;

apply information theoretic weighting to adjust values in the term co-occurrence matrix;

generate an approximation matrix of the term co-occurrence matrix with the adjusted values;

truncate the approximation matrix; and

determine a latent semantic similarity between two terms by calculating a cosine of an angle between term vectors, associated with the two terms, in the truncated approximation matrix.

3. The device of claim 1, where the one or more processors, when generating the information that identifies the input entities included in the text, are to:

generate a feature data structure that includes;

the list of the terms included in the text sections,

information identifying features associated with the terms and generated by the feature extraction technique and the one or more other feature extraction techniques,

information identifying entity labels associated with the terms and generated based on the features, and

information identifying confidence levels associated with the entity labels and generated based on the features.

4. The device of claim 3, where each of the entity labels includes one of:

a label indicating that a corresponding term is not an entity,

a label indicating that a corresponding term is not an input entity, or

a label indicating that a corresponding term is an input entity.

5. The device of claim 3, where the one or more processors are further to:

provide a query to a user requesting entity labels for a portion of the terms included in the feature data structure;

receive, based on the query and from the user, the entity labels for the portion of the terms included in the data structure; and

update the feature data structure based on the received entity labels for the portion of the terms.

6. The device of claim 1, where the one or more processors are further to:

receive information associated with processing the text;

process the text based on the information associated with processing the text;

receive user identified entities and user identified input entities; and

identify terms, included in the text sections of the text, as potential entities or potential input entities based on the user identified entities and the user identified input entities.

7. The device of claim 1, where the one or more processors are further to:

provide a query to a user requesting entity identification for a portion of the terms;

receive, based on the query and from the user, the entity identification for the portion of the terms;

update the information that identifies the input entities included in the text, based on the entity identification for the portion of the terms; and

provide the updated information that identifies the input entities included in the text.

8. A non-transitory computer-readable medium storing instructions, the instructions comprising:

one or more instructions that, when executed by one or more processors, cause the one or more processors to;

receive text to be processed to identify input entities included in the text;

identify text sections of the text;

generate a list of terms included in the text sections of the text;

perform a feature extraction technique to determine whether each term, in the list of the terms, is an object of an action,

the feature extraction technique causing a subject-predicate-object relationship to be extracted from one of the text sections;

perform one or more other feature extraction techniques, on the terms included in the text sections, to identify the input entities included in the text;

generate information that identifies the input entities included in the text, based on performing the feature extraction technique and the one or more other feature extraction techniques; and

provide the information that identifies the input entities included in the text.

9. The non-transitory computer-readable medium of claim 8, where the one or more other feature extraction techniques include at least one of:

a technique to determine tag patterns for the terms based on tags associated with the terms;

a technique to determine whether the terms are capitalized;

a technique to determine a headword in each term than includes multiple words;

a technique to determine a number of constituent words in each term;

a technique to determine semantic similarities of input type actions acting on the terms;

a technique to determine semantic similarities of non-input type actions acting on the terms;

a technique to determine a distance of each term from an action appearing in a same text section associated with each term;

or

a technique to identify surrounding context for each term and associate tags with the surrounding context.

10. The non-transitory computer-readable medium of claim 8, where the one or more instructions, that cause the one or more processors to generate the information that identifies the input entities included in the text, cause the one or more processors to:

generate a feature data structure that includes;

the list of the terms included in the text sections,

information identifying features associated with the terms and generated by the feature extraction technique and the one or more other feature extraction techniques,

information identifying entity labels associated with the terms and generated based on the features, and

information identifying confidence levels associated with the entity labels and generated based on the features.

11. The non-transitory computer-readable medium of claim 10, further comprising:

one or more instructions, that when executed by the one or more processors, cause the one or more processors to;

provide a query to a user requesting entity labels for a portion of the terms included in the feature data structure;

receive, based on the query and from the user, the entity labels for the portion of the terms included in the feature data structure; and

update the feature data structure based on the received entity labels for the portion of the terms.

12. The non-transitory computer-readable medium of claim 10, where each of the entity labels includes one of:

a label indicating that a corresponding term is not an entity,

a label indicating that a corresponding term is not an input entity, or

a label indicating that a corresponding term is an input entity.

13. The non-transitory computer-readable medium of claim 8, further comprising:

one or more instructions, that when executed by the one or more processors, cause the one or more processors to;

receive information associated with processing the text;

process the text based on the information associated with processing the text;

receive user identified entities and user identified input entities; and

identify terms, included in the text sections of the text, as potential entities or potential input entities based on the user identified entities and the user identified input entities.

14. The non-transitory computer-readable medium of claim 8, further comprising:

one or more instructions, that when executed by the one or more processors, cause the one or more processors to;

provide a query to a user requesting entity identification for a portion of the terms;

receive, based on the query and from the user, the entity identification for the portion of the terms;

update the information that identifies the input entities included in the text, based on the entity identification for the portion of the terms; and

provide the updated information that identifies the input entities included in the text.

15. A method, comprising:

receiving, by a device, text to be processed to identify input entities included in the text;

identifying, by the device, text sections of the text;

generating, by the device, a list of terms included in the text sections of the text;

performing, by the device, a feature extraction technique to determine whether each term, in the list of the terms, is an object of an action,

the feature extraction technique causing a subject-predicate-object relationship to be extracted from one of the text sections;

performing, by the device, one or more other feature extraction techniques, on the terms included in the text sections, to identify the input entities included in the text,

the one or more other feature extraction techniques including at least one of;

determining tag patterns for the terms based on tags associated with the terms,

determining whether the terms are capitalized,

determining a headword in each term than includes multiple words,

determining a number of constituent words in each term,

determining semantic similarities of input type actions acting on the terms,

determining semantic similarities of non-input type actions acting on the terms,

the non-input type actions acting on the terms being action terms that are not in the list of terms,

determining a distance of each term from an action appearing in a same text section associated with each term, or

identifying surrounding context for each term and associating tags with the surrounding context;

generating, by the device, information that identifies the input entities included in the text, based on performing the feature extraction technique and the one or more other feature extraction techniques; and

providing, by the device, the information that identifies the input entities included in the text.

16. The method of claim 15, where generating the information that identifies the input entities included in the text comprises:

generating a feature data structure that includes;

the list of the terms included in the text sections,

information identifying features associated with the terms and generated by the feature extraction technique and the one or more other feature extraction techniques,

information identifying entity labels associated with the terms and generated based on the features, and

information identifying confidence levels associated with the entity labels and generated based on the features.

17. The method of claim 16, where each of the entity labels includes one of:

a label indicating that a corresponding term is not an entity,

a label indicating that a corresponding term is not an input entity, or

a label indicating that a corresponding term is an input entity.

18. The method of claim 16, further comprising:

providing a query to a user requesting entity labels for a portion of the terms included in the feature data structure;

receiving, based on the query and from the user, the entity labels for the portion of the terms included in the data structure; and

updating the feature data structure based on the received entity labels for the portion of the terms.

19. The method of claim 15, further comprising:

receiving information associated with processing the text;

processing the text based on the information associated with processing the text;

receiving user identified entities and user identified input entities; and

identifying terms, included in the text sections of the text, as potential entities or potential input entities based on the user identified entities and the user identified input entities.

20. The method of claim 15, further comprising:

providing a query to a user requesting entity identification for a portion of the terms;

receiving, based on the query and from the user, the entity identification for the portion of the terms;

updating the information that identifies the input entities included in the text, based on the entity identification for the portion of the terms; and

providing the updated information that identifies the input entities included in the text.