System and method for extracting information from unstructured text

• US 10,002,129 B1
• Filed: 03/30/2017
• Issued: 06/19/2018
• Est. Priority Date: 02/15/2017
• Status: Active Grant

• Current Assignee: Wipro Limited
• Original Assignee: Wipro Limited
• Inventors: D'Souza, Shaun Cyprian
• Primary Examiner(s): Abebe, Daniel
• Application Number:
  US15/474,194
• Time in Patent Office: 446 Days
• Field of Search: 704 9
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking

Abstract

This disclosure relates generally to natural language processing, and more particularly to a system and method for extracting subject-verb-object (SVO) chunked text from an unstructured text. In one embodiment, a method is provided for extracting SVO chunked text from an unstructured text. The method comprises identifying a plurality of part of speech (PoS) tokens in the unstructured text, and determining a plurality of SVO chunked text directly from the plurality of PoS tokens using a machine learning chunker model. The machine learning chunker model is trained on a subject-verb-object (SVO) annotated training data.

Claims

1. A method for extracting subject-verb-object (SVO) chunked text from unstructured text, the method comprising:

identifying, by a SVO chunked text computing device, a plurality of part of speech (PoS) tokens in an unstructured text; and

determining, by the SVO chunked text computing device, a SVO chunked text directly from the plurality of PoS tokens using a machine learning chunker model, wherein the machine learning chunker model is trained on an SVO annotated training data, wherein the SVO annotated training data comprises a plurality of tokens, a plurality of corresponding PoS tags, and a plurality of corresponding SVO tags, the plurality of corresponding SVO tags comprises one or more of a subject tag, a verb tag, an object tag, or an object-subject tag, and the plurality of corresponding SVO tags is in beginninginside-other (BIO) format, and wherein the SVO annotated training data is generated based on a plurality of corresponding span information for the plurality of tokens by for each of a plurality of PoS tokens in each of a plurality of sets of syntactically related PoS tokens in a sentence, detecting a span information for a PoS token and tagging the PoS token as a subject, a verb, an object, or an object-subject based on the span information and a pervious tagging of the PoS token.

2. The method of claim 1, wherein identifying the plurality of PoS tokens comprises:

extracting a plurality of tokens from the input text, wherein each of the plurality of tokens comprises a word or a phrase; and

determining a PoS tag for each of the plurality of tokens.

3. The method of claim 1, wherein each of the plurality of SVO chunked text is a set of semantically related PoS tokens and comprises a verb phrase and at least two of a subject phrase, an object phrase, or an object-subject phrase and the object-subject phrase corresponds to an overlapping contiguous chunks that is an object phrase in an initial part of a sentence and a subject phrase in the subsequent part of the sentence.

4. The method of claim 1, wherein the machine learning chunker model is trained on one or more of:

a non-overlapping SVO annotated training data comprising one set of subject, verb, and object in each of the sentences;

or

an overlapping SVO annotated training data comprising one or more sets of subject, verb, object, and object-subject in each of the sentences.

5. The method of claim 1, wherein the machine learning chunker model determines the plurality of SVO chunked text directly from the plurality of PoS tokens without a set of heuristics or a set of rules.

6. A subject-verb-object (SVO) chunked computing device, comprising;

at least one processor; and

memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising;

identify a plurality of part of speech (PoS) tokens in an unstructured text; and

determine a SVO chunked text directly from the plurality of PoS tokens using a machine learning chunker model, wherein the machine learning chunker model is trained on an SVO annotated training data, wherein the SVO annotated training data comprises a plurality of tokens, a plurality of corresponding PoS tags, and a plurality of corresponding SVO tags, the plurality of corresponding SVO tags comprises one or more of a subject tag, a verb tag, an object tag, or an object-subject tag, and the plurality of corresponding SVO tags is in beginninginside-other (BIO) format, and wherein the SVO annotated training data is generated based on a plurality of corresponding span information for the plurality of tokens by for each of a plurality of PoS tokens in each of a plurality of sets of syntactically related PoS tokens in a sentence, detecting a span information for a PoS token and tagging the PoS token as a subject, a verb, an object, or an object-subject based on the span information and a pervious tagging of the PoS token.

7. The SVO chunked computing device of claim 6, wherein the instructions, when executed by the at least one processor, further cause the at least one processor to:

extract a plurality of tokens from the input text, wherein each of the plurality of tokens comprises a word or a phrase; and

determine a PoS tag for each of the plurality of tokens.

8. The SVO chunked computing device of claim 6, wherein each of the plurality of SVO chunked text is a set of semantically related PoS tokens and comprises a verb phrase and at least two of a subject phrase, an object phrase, or an object-subject phrase and the objectsubject phrase corresponds to an overlapping contiguous chunks that is an object phrase in an initial part of a sentence and a subject phrase in the subsequent part of the sentence.

9. The SVO chunked computing device of claim 6, wherein the machine learning chunker model is trained on one or more of:

a non-overlapping SVO annotated training data comprising one set of subject, verb, and object in each of the sentences;

or

an overlapping SVO annotated training data comprising one or more sets of subject, verb, object, and object-subject in each of the sentences.

10. The SVO chunked computing device of claim 6, wherein the machine learning chunker model determines the plurality of SVO chunked text directly from the plurality of PoS tokens without a set of heuristics or a set of rules.

11. A non-transitory computer-readable medium having stored thereon instructions for extracting subject-verb-object (SVO) chunked text from unstructured text comprising executable code which, when executed by one or more processors, causes the one or more processors to:

identify a plurality of part of speech (PoS) tokens in the unstructured text; and

determine a plurality of SVO chunked text directly from the plurality of PoS tokens using a machine learning chunker model, wherein the machine learning chunker model is trained on a subject-verb-object (SVO) annotated training data, wherein the SVO annotated training data comprises a plurality of tokens, a plurality of corresponding PoS tags, and a plurality of corresponding SVO tags, the plurality of corresponding SVO tags comprises one or more of a subject tag, a verb tag, an object tag, or an object-subject tag, and the plurality of corresponding SVO tags is in beginninginside-other (BIO) format, and wherein the SVO annotated training data is generated based on a plurality of corresponding span information for the plurality of tokens by for each of a plurality of PoS tokens in each of a plurality of sets of syntactically related PoS tokens in a sentence, detecting a span information for a PoS token and tagging the PoS token as a subject, a verb, an object, or an object-subject based on the span information and a pervious tagging of the PoS token.

12. The non-transitory computer-readable medium of claim 11, wherein the executable code, when executed by the one or more processor, further causes the one or more processor to:

extract a plurality of tokens from the input text, wherein each of the plurality of tokens comprises a word or a phrase; and

determine a PoS tag for each of the plurality of tokens.

13. The non-transitory computer-readable medium of claim 11, wherein each of the plurality of SVO chunked text is a set of semantically related PoS tokens and comprises a verb phrase and at least two of a subject phrase, an object phrase, or an object-subject phrase and the object-subject phrase corresponds to an overlapping contiguous chunks that is an object phrase in an initial part of a sentence and a subject phrase in the subsequent part of the sentence.

14. The non-transitory computer-readable medium of claim 11, wherein the machine learning chunker model is trained on one or more of:

a non-overlapping SVO annotated training data comprising one set of subject, verb, and object in each of the sentences;

or

an overlapping SVO annotated training data comprising one or more sets of subject, verb, object, and object-subject in each of the sentences.

15. The non-transitory computer-readable medium of claim 11, wherein the machine learning chunker model determines the plurality of SVO chunked text directly from the plurality of PoS tokens without a set of heuristics or a set of rules.