System and method for automatically summarizing fine-grained opinions in digital text

• US 9,092,514 B2
• Filed: 09/27/2012
• Issued: 07/28/2015
• Est. Priority Date: 10/29/2007
• Status: Active Grant

• Current Assignee: Cornell University
• Original Assignee: Cornell University
• Inventors: Cardie, Claire, Breck, Eric, Choi, Yejin, Stoyanov, Vaselin
• Primary Examiner(s): Filipczyk, Marc
• Application Number:
  US13/629,004
• Publication Number:
  US 20130024183A1
• Time in Patent Office: 1,034 Days
• Field of Search: 706/46, 706/56, 707/737, 707/755, 707/758, 707/999.006, 707/999.101, 715/764
• US Class Current: 1/1
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06F 16/38 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/93 Document management systems

Abstract

A method and system for automatically summarizing fine-grained opinions in digital text are disclosed. Accordingly, a digital text is analyzed for the purpose of extracting all opinion expressions found in the text. Next, the extracted opinion expressions (referred to herein as opinion frames) are analyzed to generate opinion summaries. In forming an opinion summary, those opinion frames sharing in common an opinion source and/or opinion topic may be combined, such that an overall opinion summary indicates an aggregate opinion held by the common source toward the common topic.

Claims

1. A computer-implemented method, using a hardware processor, for generating an opinion summary from a digital text, said method comprising:

using at least one of syntactic analysis, semantic analysis and discourse analysis to assist in automatically extracting a plurality of opinion expressions from the digital text using the processor, each opinion expression identified by a grammatical element and associated with an opinion source to which the opinion expression is to be attributed, an opinion topic, an opinion polarity and an opinion strength, wherein for each opinion expression the opinion source is at least one person or entity to which such opinion expression is to be attributed, and wherein the digital text includes opinions expressions from a plurality of opinion sources;

identifying those opinion expressions that share a common opinion source and a common opinion topic using the processor; and

combining the opinion polarities and opinion strengths of those opinion expressions that share a common opinion source to form an overall opinion summary with one aggregate opinion frame for each common opinion source-topic pairing.

2. The computer-implemented method of claim 1, wherein identifying those opinion expressions that share a common opinion source includes identifying those opinion expressions that have a common real-world opinion source expressed in the digital text with different language.

3. The computer-implemented method of claim 1, wherein identifying those opinion expressions that share a common opinion topic includes identifying those opinion expressions that have a common real-world opinion topic expressed in the digital text with different language.

4. The computer-implemented method of claim 1, further comprising:

for those opinion expressions that share a common opinion source and common opinion topic, generating an overall opinion summary having opinion polarities and opinion strengths based on averaging those opinion polarities of the opinion expressions that share the common opinion source and common opinion topic, wherein the averaging takes into consideration the opinion strength associated with each opinion polarity.

5. The computer-implemented method of claim 1, further comprising:

for those opinion expressions that share a common opinion source and a common opinion topic, generating an overall opinion summary having an opinion polarity and opinion strength based on the opinion polarity and opinion strength of the opinion expression identified as having a most extreme opinion polarity and opinion strength of the opinion expressions that share a common opinion source and common opinion topic.

6. The computer-implemented method of claim 1, further comprising:

for those opinion expressions that share a common opinion source and a common opinion topic, generating an overall opinion summary which indicates whether there is a conflict between any two opinion polarities of any two opinion expressions that share a common opinion source and common opinion topic.

7. The computer-implemented method of claim 1, wherein each opinion expression is associated with at least one of an author and a publisher of the digital text in which the opinion expression is located.

8. The computer-implemented method of claim 1, wherein the grammatical element is at least one of a word, a plurality of words, and a phrase.

9. The computer-implemented method of claim 1, further comprising:

for each overall opinion summary associated with a unique opinion source, determining a number of at least one of positive opinion expressions and negative opinion expressions associated with the unique opinion source.

10. The computer-implemented method of claim 1, further comprising:

for each overall opinion summary associated with a unique opinion source, determining a percentage of at least one of positive opinion expressions and negative opinion expressions associated with the unique opinion source.

11. A computer-implemented method, using a hardware processor, for generating an opinion summary from a digital text, said method comprising:

using at least one of syntactic analysis, semantic analysis and discourse analysis to assist in automatically extracting a plurality of opinion expressions from the digital text using the processor, each opinion expression identified by a grammatical element and associated with an opinion source to which the opinion expression is to be attributed, an opinion topic, an opinion polarity and an opinion strength, wherein for each opinion expression the opinion source is at least one person or entity to which such opinion expression is to be attributed, and wherein the digital text includes opinions expressions from a plurality of opinion sources;

organizing the plurality of opinion expressions into data fields, including an opinion source field, using the processor;

identifying those opinion expressions that share a common opinion topic using the processor; and

generating an opinion summary with one aggregate opinion frame for each unique opinion topic, wherein the opinion summary is based at least in part on combining the opinion polarities and opinion strengths of those opinion expressions that share the common opinion topic.

12. The computer-implemented method of claim 11, wherein identifying those opinion expressions that share a common opinion topic includes identifying those opinion expressions that have a common real-world opinion topic expressed in the digital text with different language.

13. The computer-implemented method of claim 11, further comprising:

for those opinion expressions that share a common opinion topic, generating an overall opinion summary having opinion polarities and opinion strengths based on averaging those opinion polarities of the opinion expressions that share the common opinion topic, wherein the averaging takes into consideration the opinion strength associated with each opinion polarity.

14. The computer-implemented method of claim 11, further comprising:

for those opinion expressions that share a common opinion topic, generating an overall opinion summary having an opinion polarity and opinion strength based on the opinion polarity and opinion strength of the opinion expression identified as having a most extreme opinion polarity and opinion strength of the opinion expressions that share a common opinion topic.

15. The computer-implemented method of claim 11, further comprising:

for those opinion expressions that share a common opinion topic, generating an overall opinion summary which indicates whether there is a conflict between any two opinion polarities of any two opinion expressions that share a common opinion topic.

16. The computer-implemented method of claim 11, further comprising:

for each overall opinion summary associated with a unique topic, determining at least one of a number of and a percentage of at least one of positive opinion expressions and negative opinion expressions directed toward the unique topic.

17. A computer-implemented method, using a hardware processor, for generating an opinion summary from a digital text, said method comprising:

using at least one of syntactic analysis, semantic analysis and discourse analysis to assist in automatically extracting a plurality of opinion expressions from the digital text using the processor, each opinion expression identified by a grammatical element and associated with an opinion source to which the opinion expression is to be attributed, an opinion topic, an opinion polarity and an opinion strength, wherein for each opinion expression the opinion source is at least one person or entity to which such opinion expression is to be attributed, and wherein the digital text includes opinions expressions from a plurality of opinion sources;

identifying those opinion expressions that share a common opinion source using the processor; and

combining the opinion polarities and opinion strengths of those opinion expressions that share a common opinion source to form an overall opinion summary with one aggregate opinion frame for each common opinion source.

18. The computer-implemented method of claim 17, wherein identifying those opinion expressions that share a common opinion source includes identifying those opinion expressions that have a common real-world opinion source expressed in the digital text with different language.

19. The computer-implemented method of claim 17, further comprising:

for those opinion expressions that share a common opinion source, generating an overall opinion summary having opinion polarities and opinion strengths based on averaging those opinion polarities of the opinion expressions that share the common opinion source, wherein the averaging takes into consideration the opinion strength associated with each opinion polarity.

20. The computer-implemented method of claim 17, further comprising:

for those opinion expressions that share a common opinion source, generating an overall opinion summary having an opinion polarity and opinion strength based on the opinion polarity and opinion strength of the opinion expression identified as having a most extreme opinion polarity and opinion strength of the opinion expressions that share a common opinion source.

21. The computer-implemented method of claim 17, further comprising:

for those opinion expressions that share a common opinion source, generating an overall opinion summary which indicates whether there is a conflict between any two opinion polarities of any two opinion expressions that share a common opinion source.

22. The computer-implemented method of claim 17, further comprising:

for each overall opinion summary associated with a unique opinion source, determining at least one of a number of and a percentage of at least one of positive opinion expressions and negative opinion expressions associated with the unique opinion source.

23. A computer-implemented method, using a hardware processor, for identifying and grouping opinion expressions, said method comprising:

using at least one of syntactic analysis, semantic analysis and discourse analysis to assist in automatically extracting a plurality of opinion expressions from a digital text using the processor, each opinion expression identified by a grammatical element and associated with an opinion source to which the opinion expression is to be attributed, an opinion topic, an opinion polarity and an opinion strength, wherein for each opinion expression the opinion source is at least one person or entity to which such opinion expression is to be attributed, and wherein the digital text includes opinions expressions from a plurality of opinion sources;

identifying and grouping those opinion expressions that share a common opinion source and a common opinion topic using the processor; and

presenting the grouped opinion expressions on a display, wherein the grouped opinion expressions are based at least in part on combining the opinion polarities and opinion strengths of those opinion expressions that share a common opinion source to form an overall opinion summary with one aggregate opinion frame for each common opinion source-topic pairing.

24. The computer-implemented method of claim 23, wherein identifying and grouping those opinion expressions that share at least one of a common opinion source and a common opinion topic includes identifying those opinion expressions that have at least one of a common real-world opinion source and a common real-world opinion topic expressed in the digital text with different language.

25. The computer-implemented method of claim 23, wherein identifying and grouping those opinion expressions that share at least one of a common opinion source and a common opinion topic includes identifying those opinion expressions that have a common real-world opinion topic expressed in the digital text with different language.

26. The computer-implemented method of claim 23, further comprising:

for those opinion expressions that share a common opinion source and a common opinion topic, generating an overall opinion summary having opinion polarities and opinion strengths based on averaging those opinion polarities of the opinion expressions that share the common opinion source and common opinion topic, wherein the averaging takes into consideration the opinion strength associated with each opinion polarity.

27. The computer-implemented method of claim 23, further comprising:

for those opinion expressions that share a common opinion source and a common opinion topic, generating an overall opinion summary having an opinion polarity and opinion strength based on the opinion polarity and opinion strength of the opinion expression identified as having a highest opinion polarity and opinion strength of the opinion expressions that share a common opinion source and a common opinion topic.

28. The computer-implemented method of claim 23, further comprising:

for those opinion expressions that share a common opinion source and a common opinion topic, generating an overall opinion summary which indicates whether there is a conflict between any two opinion polarities of any two opinion expressions that share a common opinion source and opinion topic.

29. A non-transitory computer-readable storage medium storing a set of instructions for an execution via a processing circuit, wherein said execution implementing a method for generating an opinion summary from a digital text, the method comprising:

using at least one of syntactic analysis, semantic analysis and discourse analysis to assist in automatically extracting a plurality of opinion expressions from the digital text, each opinion expression identified by a grammatical element and associated with an opinion source to which the opinion expression is to attributed, an opinion topic, an opinion polarity and an opinion strength, wherein for each opinion expression the opinion source is at least one person or entity to which such opinion expression is to be attributed, and wherein the digital text includes opinion expressions from a plurality of opinion sources; and

identifying and grouping those opinion expressions that share a common opinion source and a common opinion topic; and

combining the opinion polarities and opinion strengths of those opinion expressions that share a common opinion source to form an overall opinion summary with one aggregate opinion frame for each common opinion source-topic pairing.