Applying a genetic algorithm to compositional semantics sentiment analysis to improve performance and accelerate domain adaptation

• US 9,373,075 B2
• Filed: 12/12/2013
• Issued: 06/21/2016
• Est. Priority Date: 12/12/2013
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Bufe, John P. III, Winkler, Timothy P.
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Sitiriche, Luis
• Application Number:
  US14/104,811
• Publication Number:
  US 20150170051A1
• Time in Patent Office: 922 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition

Abstract

A mechanism is provided in a data processing system for applying a genetic algorithm to semantic sentiment analysis. The mechanism provides a sentiment analysis model to a sentiment analysis algorithm. The mechanism trains the sentiment analysis model using a genetic algorithm based on a training corpus of documents with corresponding desired sentiment analysis values for a given domain to form a trained sentiment analysis model. The mechanism performs the sentiment analysis algorithm on an input document using the trained sentiment analysis model to form a domain-specific sentiment analysis result. The mechanism outputs the domain-specific sentiment analysis result.

Claims

1. A method, in a data processing system, for applying a genetic algorithm to semantic sentiment analysis, the method comprising:

providing, by the data processing system, a sentiment analysis model to a sentiment analysis algorithm;

training, by the data processing system, the sentiment analysis model using a genetic algorithm based on a training corpus of documents with corresponding desired sentiment analysis values for a given domain to form a trained sentiment analysis model, wherein training the sentiment analysis model using the genetic algorithm comprises for each given training document in the training corpus;

performing the sentiment analysis algorithm on the given training document to form a sentiment analysis result;

making modifications to the sentiment analysis model to form a plurality of individual models;

identifying a best fitness individual model from the plurality of individual models using the genetic algorithm, wherein the best fitness individual model minimizes a distance from desired sentiment analysis values corresponding to the given training document, wherein fitness of each individual model is based on an absolute value of a distance from a generated sentiment value to a desired sentiment value weighted by an amount of change from an initial value in the sentiment analysis model; and

storing the best fitness individual model as the trained sentiment analysis model;

performing, by the data processing system, the sentiment analysis algorithm on an input document using the trained sentiment analysis model to form a domain-specific sentiment analysis result;

outputting, by the data processing system, the domain-specific sentiment analysis result;

providing the domain-specific sentiment analysis result to a question answering system; and

performing analysis of an input question or a candidate answer in the question answering system using the domain-specific sentiment analysis result.

2. The method of claim 1, wherein making modifications to the sentiment analysis model using the genetic algorithm comprises modifying rules or token values in the sentiment analysis model.

3. The method of claim 2, wherein a token value comprises a numerical value or a multiplier.

4. The method of claim 1, wherein the input document comprises a sample token, phrase, sentence, paragraph, or document.

5. The method of claim 1, wherein the training corpus of documents comprises sample tokens, phrases, sentences, paragraphs, or documents for which desired sentiment values are known.

6. A computer program product comprising a non-transitory computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:

provide a sentiment analysis model to a sentiment analysis algorithm;

train the sentiment analysis model using a genetic algorithm based on a training corpus of documents with corresponding desired sentiment analysis values for a given domain to form a trained sentiment analysis model;

wherein training the sentiment analysis model comprises for each given training document in the training corpus;

performing the sentiment analysis algorithm on the given training document to form a sentiment analysis result;

making modifications to the sentiment analysis model to form individual models;

identifying a best fitness individual model from the plurality of individual models using the genetic algorithm, wherein the best fitness individual model minimizes a distance from desired sentiment analysis values corresponding to the given training document, wherein fitness of each individual model is based on an absolute value of distance from a generated sentiment value to a desired sentiment value weighted by an amount of change from an initial value in the sentiment analysis model; and

storing the best fitness individual model as the trained sentiment analysis model;

perform the sentiment analysis algorithm on an input document using the trained sentiment analysis model to form a domain-specific sentiment analysis result; and

output the domain-specific sentiment analysis result;

provide the domain-specific sentiment analysis result to a question answering system; and

perform analysis of an input question or a candidate answer in the question answering system using the domain-specific sentiment analysis result.

7. The computer program product of claim 6, wherein making modifications to the sentiment analysis model using the genetic algorithm comprises modifying rules or token values in the sentiment analysis model.

8. The computer program product of claim 7, wherein a token value comprises a numerical value or a multiplier.

9. The computer program product of claim 6, wherein the training corpus of documents comprises sample tokens, phrases, sentences, paragraphs, or documents for which desired sentiment values are known.

10. The computer program product of claim 6, wherein the input document comprises a sample token, phrase, sentence, paragraph, or document.

11. An apparatus comprising:

a processor; and

a memory coupled to the processor, wherein the memory comprises instructions which, when executed by the processor, cause the processor to;

provide a sentiment analysis model to a sentiment analysis algorithm;

train the sentiment analysis model using a genetic algorithm based on a training corpus of documents with corresponding desired sentiment analysis values for a given domain to form a trained sentiment analysis model;

wherein training the sentiment analysis model comprises for each given training document in the training corpus;

performing the sentiment analysis algorithm on the give training document to form a sentiment analysis result;

making modifications to the sentiment analysis model to form individual models;

indentifying a best fitness individual model from the plurality of individual models using the genetic algorithm, wherein the best fitness individual model minimizes a distance from desired sentiment analysis values corresponding to the given training document, wherein fitness of each individual model is based on an absolute value of a distance from a generated sentiment value to a desired sentiment value weighted by an amount of change from an initial value in the sentiment analysis model; and

storing the best fitness individual model as the trained sentiment analysis model;

perform the sentiment analysis algorithm on an input document using the trained sentiment analysis model to form a domain-specific sentiment analysis result;

output the domain-specific sentiment analysis result;

provide the domain-specific sentiment analysis result to a question answering system; and

perform analysis of an input question or a candidate answer in the question answering system using the domain-specific sentiment analysis result.

12. The apparatus of claim 11, wherein making modifications to the sentiment analysis model using the genetic algorithm comprises modifying rules or token values in the sentiment analysis model.

13. The apparatus of claim 12, wherein a token value comprises a numerical value or a multiplier.

14. The apparatus of claim 11, wherein the training corpus of documents comprises sample tokens, phrases, sentences, paragraphs, or documents for which desired sentiment values are known.

15. The apparatus of claim 11, wherein the input document comprises a sample token, phrase, sentence, paragraph, or document.