Aspect-based sentiment analysis and report generation using machine learning methods

• US 10,198,432 B2
• Filed: 08/16/2016
• Issued: 02/05/2019
• Est. Priority Date: 07/28/2016
• Status: Active Grant

• Current Assignee: ABBYY Development LLC
• Original Assignee: ABBYY Production LLC (ABBYY Software)
• Inventors: Mikhaylov, Maksim Borisovich, Pasechnikov, Konstantin Alekseevich
• Primary Examiner(s): Spooner, Lamont M
• Application Number:
  US15/237,954
• Publication Number:
  US 20180032507A1
• Time in Patent Office: 903 Days
• Field of Search: 704 1, 704 9, 704 10
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/56 Natural language generation

Abstract

Systems and methods for aspect-based sentiment analysis using machine learning methods. An example method comprises: receiving, by a computer system, a custom dictionary comprising a list of lexemes referencing at least one of: a target entity or an aspect associated with the target entity; performing, using the custom dictionary, a syntactico-semantic analysis of at least part of a natural language text to produce a plurality of syntactico-semantic structures representing the part of the natural language text; evaluating, using one or more text characteristics produced by the syntactico-semantic analysis, a classifier function to determine polarities associated with one or more aspect terms; and generating a report comprising the aspect terms and polarities of aspects referenced by the aspect terms.

Claims

1. A method, comprising;

receiving, by a computer system, a custom dictionary comprising a list of lexemes referencing at least one of;

a target entity or an aspect associated with the target entity;

performing, using the custom dictionary, a syntactico-semantic analysis of at least part of a natural language text to produce a plurality of syntactico-semantic structures representing the part of the natural language text;

interpreting the plurality of syntactico-semantic structures to detect, within the part of the natural language text, an aspect term representing an aspect associated with a target entity;

identifying, in the plurality of syntactico-semantic structures, a highest constituent having a kernel comprised by the aspect term;

evaluating a classifier function to determine a polarity associated with the aspect term, wherein a domain of the classifier function comprises one or more attributes of a context of the highest constituent; and

generating a report comprising the aspect term and the polarity of the aspect term.

2. The method of claim 1, wherein the natural language text represents a plurality of consumer reviews of the target entity.

3. The method of claim 1, wherein the target entity is represented by at least one of:

a consumer product or a service.

4. The method of claim 1, wherein the polarity associated with the aspect term is represented by one of:

a negative polarity, a neutral polarity, or a positive polarity.

5. The method of claim 1, wherein interpreting the syntactico-semantic structures is performed by applying a set of production rules.

6. The method of claim 5, further comprising:

receiving a user input modifying the identified aspects term.

7. The method of claim 1, further comprising:

receiving a user input modifying the identified polarity of the aspects term.

8. The method of claim 1, wherein the classifier function is represented by one of:

a linear classifier, a linear tree classifier, a random forest classifier, a conditional random field (CRF) classifier, a latent Dirichlet allocation (LDA) classifier, a support vector machine (SVM) classifiers, or a neural network-based classifier.

9. The method of claim 1, further comprising:

determining, using a training data set, at least one parameter of the classifier function, wherein the training data set comprises a training natural language text comprising a plurality of aspect terms.

10. The method of claim 1, wherein each syntactico-semantic structure of the plurality of syntactico-semantic structures is represented by a graph comprising a plurality of nodes corresponding to a plurality of syntactico-semantic classes and a plurality of edges corresponding to a plurality of syntactico-semantic relationships.

11. The method of claim 1, further comprising:

producing the custom dictionary represented by a semantic hierarchy comprising definitions of a plurality of semantic classes.

12. The method of claim 1, wherein the aspect term represents one of:

a function of the target entity or a component of the target entity.

13. A system, comprising:

a memory; and

a processor, coupled to the memory, the processor configured to;

receive a custom dictionary comprising a list of lexemes referencing at least one of;

a target entity or an aspect associated with the target entity;

perform, using the custom dictionary, a syntactico-semantic analysis of at least part of a natural language text to produce a plurality of syntactico-semantic structures representing the part of the natural language text;

interpret the plurality of syntactico-semantic structures to detect, within the part of the natural language text, an aspect term representing an aspect associated with a target entity;

identifying, in the plurality of syntactico-semantic structures, a highest constituent having a kernel comprised by the aspect term;

evaluate a classifier function to determine a polarity associated with the aspect term, wherein a domain of the classifier function comprises one or more attributes of a context of the highest constituent; and

generate a report comprising the aspect term and the polarity of the aspect term.

14. The system of claim 13, wherein the polarity associated with the aspect term is represented by one of:

a negative polarity, a neutral polarity, or a positive polarity.

15. The system of claim 13, wherein interpreting the syntactico-semantic structures is performed by applying a set of production rules.

16. The system of claim 15, wherein the processor is further configured to:

receive a user input modifying the identified aspects term.

17. The system of claim 13, wherein the processor is further configured to:

receive a user input modifying the identified polarity of the aspects term.

18. The system of claim 13, wherein the processor is further configured to:

determine, using a training data set, at least one parameter of the classifier function, wherein the training data set comprises a training natural language text comprising a plurality of aspect terms.

19. The system of claim 13, wherein the aspect term represents one of:

a function of the target entity or a component of the target entity.

20. A computer-readable non-transitory storage medium comprising executable instructions that, when executed by a computer system, cause the computer system to:

receive a custom dictionary comprising a list of lexemes referencing at least one of;

a target entity or an aspect associated with the target entity;

perform, using the custom dictionary, a syntactico-semantic analysis of at least part of a natural language text to produce a plurality of syntactico-semantic structures representing the part of the natural language text;

interpret the plurality of syntactico-semantic structures to detect, within the part of the natural language text, an aspect term representing an aspect associated with a target entity;

identify, in the plurality of syntactico-semantic structures, a highest constituent having a kernel comprised by the aspect term;

evaluate a classifier function determine a polarity associated with the aspect term, wherein a domain of the classifier function comprises one or more attributes of a context of the highest constituent; and

generate a report comprising the aspect term and the polarity of the aspect term.