Topical sentiments in electronically stored communications

• US 7,523,085 B2
• Filed: 09/30/2005
• Issued: 04/21/2009
• Est. Priority Date: 09/30/2004
• Status: Active Grant

• Current Assignee: Buzzmetrics, Ltd. (Nielsen Holdings plc)
• Original Assignee: Buzzmetrics, Ltd. (Nielsen Holdings plc)
• Inventors: Hurst, Matthew F., Nigam, Kamal P.
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US11/245,542
• Publication Number:
  US 20060069589A1
• Time in Patent Office: 1,299 Days
• Field of Search: 706/55
• US Class Current: 706/55
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

The present application presents methods for performing topical sentiment analysis on electronically stored communications employing fusion of polarity and topicality. The present application also provides methods for utilizing shallow NLP techniques to determine the polarity of an expression. The present application also provides a method for tuning a domain-specific polarity lexicon for use in the polarity determination. The present application also provides methods for computing a numeric metric of the aggregate opinion about some topic expressed in a set of expressions.

Claims

1. A computer implemented method for obtaining topical sentiments from a plurality of electronically stored communications comprising:

determining with the assistance of a computer whether each communication in a plurality of communications is topical to a first topic;

for each communication determined to be topical to the first topic, separating with the assistance of a computer the communication into least one expression;

for each of the at least one expression, determining with the assistance of a computer if the expression is topical to a second topic;

for each expression that is determined to be topical to the second topic, determining with the assistance of a computer a polarity of the expression; and

reporting the polarity of each expression.

2. The method of claim 1, wherein the polarity is taken from at least one of the following sentiments:

positive, negative, and neutral.

3. The method of claim 1, wherein determining the polarity of the expression includes:

establishing a topical domain being explored;

generating a polarity lexicon of at least one sentimental word or sentimental phrase associated with the topical domain;

utilizing with the assistance of a computer the polarity lexicon against at least one word or phrase found in the expression; and

assigning the at least one word or phrase in the expression a polarity associated with at least one matching sentimental word or sentimental phrase in the polarity lexicon.

4. The method of claim 1, wherein determining the polarity of the expression further comprises analyzing the expression with at least one of syntactic or semantic rules.

5. The method of claim 1, wherein determining with the assistance of a computer whether each communication in a plurality of communications is topical to a first topic includes processing each communication with a text classifier.

6. The method of claim 5, wherein the text classifier utilizes at least one of:

a Winnow algorithm, a Support Vector Machine algorithm, a k-Nearest Neighbor algorithm and a rules-based classifier.

7. The method of claim 1, wherein, for each expression, determining with the assistance of a computer if the expression is topical to a second topic includes processing each expression with a text classifier.

8. The method of claim 7, wherein the text classifier utilizes at least one of:

a Winnow algorithm, a Support Vector Machine algorithm, a k-Nearest Neighbor algorithm and a rules-based classifier.

9. The method of claim 1, further comprising calculating with the assistance of a computer an aggregate metric from the plurality of expressions which estimates the frequency of at least one of positive or negative polar expressions.

10. The method of claim 9, wherein calculating an aggregate metric from the plurality of expressions comprises generating of statistically-valid confidence bounds on the aggregate metric.

11. The method of claim 10, wherein calculating an aggregate metric from the plurality of expressions includes:

for each of the plurality of expressions, estimating an opinion based upon at least one of a presence, absence or strength of polarity associated with the first topic; and

aggregating the opinions for the plurality of expressions into an overall opinion.

12. The method of claim 11, wherein aggregating the opinions for the plurality of expressions includes normalizing a ratio of at least one of empirical or estimated frequency of positive and negative polarity associated with the first topic.

13. The method of claim 10 wherein calculating an aggregate metric from the plurality of expressions includes utilizing Bayesian statistics to derive estimates for positive and negative frequencies of polar expressions.

14. The method of claim 1, wherein the first topic is a general topic and the second topic is a specific topic associated with the general topic.

15. The method of claim 14, wherein the general topic is at least one of a product or service and the specific topic is a feature of the at least one of the product or service.

16. The method of claim 14, wherein the general topic is a commercial brand and the specific topic is a feature of the commercial brand.

17. The method of claim 1, wherein the first topic and the second topic are the same topic.

18. A method for calculating, from a plurality of electronically stored expressions, an aggregate metric which estimates a frequency of at least one of positive or negative polar expressions contained in the expressions, comprising:

determining with the assistance of a computer whether each communication in a plurality of communications is topical to a first topic;

for each communication determined to be topical to the first topic, separating the communication into a plurality of electronically stored expressions;

determining with the assistance of a computer an opinion contained in each of the plurality of electronically stored expressions based upon at least one of a presence, absence and strength of polarity associated with the first topic;

calculating an aggregate metric from the determined opinions; and

reporting the aggregate metric.

19. The method of claim 18, wherein calculating the aggregate metric from the determined opinions includes generating statistically-valid confidence bounds on the aggregate metric.

20. The method of claim 18, wherein calculating the aggregate metric from the determined opinions includes normalizing a ratio of at least one of empirical or estimated frequency of positive and negative polarity associated with the first topic.

21. The method of claim 20 wherein calculating the aggregate metric from the determined opinions further includes utilizing Bayesian statistics to derive estimates for positive and negative frequencies of polar expressions.

22. The method of claim 18, wherein calculating the aggregate metric from the determined opinions includes utilizing Bayesian statistics to derive estimates for positive and negative frequencies of polar expressions.

23. The method of claim 18, wherein determining the opinion for each of the plurality of electronically stored expressions further comprises:

determining with the assistance of a computer that the expression is topical to the first topic; and

determining with the assistance of a computer a polarity of the expression.

24. The method of claim 23, wherein determining with the assistance of a computer that the expression is topical to the first topic includes utilizing a text classifier.

25. The method of claim 24, wherein determining with the assistance of a computer a polarity of the expression includes:

establishing a topical domain being explored;

generating a polarity lexicon of at least one sentimental word or sentimental phrase associated with the topical domain;

utilizing with the assistance of a computer the polarity lexicon against at least one word or phrase found in the expression; and

assigning the at least one word or phrase in the expression a polarity associated with at least one of a matching sentimental word or sentimental phrase in the polarity lexicon.

26. The method of claim 24, wherein determining with the assistance of a computer the polarity of the expression includes analyzing with the assistance of a computer the expression with at least one of syntactic or semantic rules.

27. The method of claim 23, wherein determining with the assistance of a computer the polarity of the expression includes:

establishing a topical domain being explored;

generating a polarity lexicon of at least one sentimental word or sentimental phrase associated with the topical domain;

utilizing with the assistance of a computer the polarity lexicon against at least one word or phrase found in the expression; and

assigning the at least one word or phrase in the expression a polarity associated with at least one of a matching word or a matching phrase in the polarity lexicon.

28. The method of claim 23, wherein determining with the assistance of a computer the polarity of the expression includes analyzing with the assistance of a computer the expression with at least one of syntactic or semantic rules.

29. A computer implemented method for finding at least one polar expression in a plurality of electronically stored communications, comprising:

determining with the assistance of a computer whether each communication in a plurality of communications is topical to a first predefined topic;

for each communication determined to be topical to the first topic, separating the communication into at least one expression;

for each expression, determining with the assistance of a computer if the expression is topical to a second topic;

providing a polarity lexicon of at least one of sentimental words or sentimental phrases associated with a topical domain;

for each expression that is determined to be topical to the second topic, utilizing with the assistance of a computer the polarity lexicon against the at least one word or phrase found in the expression; and

assigning with the assistance of a computer a polarity associated with at least one matching sentimental word or sentimental phrase in the polarity lexicon to the at least one word or phrase in the expression.

30. The method of claim 29, wherein assigning with the assistance of a computer the at least one word or phrase in the expression a polarity associated with the at least one of the matching sentimental word or matching sentimental phrase in the polarity lexicon, includes:

separating the expression into at least one word chunk or phrase chunk;

tagging the at least one word chunk or phrase chunk with part-of-speech tags;

applying the polarity lexicon against the at least one word chunk or phrase chunk to tag at least one of the at least one word chunk or phrase chunk with a polarity tag; and

applying syntactic and semantic rules against the tagged at least one word chunks or phrase chunk to evaluate the polarity of the expression.

31. The method of claim 30, wherein applying syntactic and semantic rules against the at least one word chunk or phrase chunk to evaluate the polarity of the expression includes identifying at least one word chunk or phrase chunk in the expression that toggles the polarity of the at least one word chunk or phrase chunk tagged with the polarity tag.

32. The method of claim 30, wherein applying syntactic and semantic rules against the at least one of word chunk or phrase chunk to evaluate the polarity of the expression includes performing with the assistance of a computer grammatical analysis on the expression.

33. The method of claim 30, further comprising generating with the assistance of a computer the polarity lexicon from a domain-general lexicon that is tuned to a topical domain being explored.

34. The method of claim 33, further comprising tuning with the assistance of a computer implemented graphical user interface the polarity lexicon by providing a user with candidates for at least one of addition, subtraction or exclusion to the polarity lexicon and for at least one of adding, subtracting or excluding each candidate from the polarity lexicon according to input received by the graphical user interface.

35. A computer implemented method for obtaining topical sentiments from a body of electronically stored communications comprising:

isolating with the assistance of a computer a first subset of the communications topical to a particular topic;

isolating with the assistance of a computer a second subset of communications from the first subset of communications where the second subset of communications includes polar segments;

for each communication in the second subset of communications, separating the communication into at least one expression;

for each of the at least one expression, determining with the assistance of a computer if the expression is topical to a second topic; and

reporting each expression that is topical to the second topic.

36. The method of claim 35, wherein the polar segments are at least one of positive or negative.

37. The method of claim 35, wherein the polar segments located in the first subset of communications are positive and the method further includes isolating a third subset of communications from the first subset where the third subset of communications include negative segments located in the first set of communications.

38. The method of claim 35 further comprising associating the polar segments with the particular topic.

39. The method of claim 35 further comprising displaying the polar segments on a computer display.

40. The method of claim 39, further comprising displaying a communication from which the polar segment was extracted.

41. The method of claim 35, wherein the segment is at least one of a sentence, a phrase, a paragraph or an entire communication.

42. A computer implemented method for tuning a polarity lexicon for use in classifying polar expressions, comprising:

providing a polarity lexicon;

determining with the assistance of a computer whether each communication in a plurality of communications is topical to a first topic;

for each communication determined to be topical to the first topic, separating the communication into at least one expression;

scanning the at least one expression for candidate words that have the potential to be added to the polarity lexicon;

with the assistance of a computer implemented graphical user interface providing a user with the candidate words for at least one of addition, subtraction or exclusion to the polarity lexicon; and

at least one of adding, subtracting or excluding each candidate word from the polarity lexicon according to input received by the graphical user interface.

43. The method of claim 42, wherein scanning includes a pattern based method that locates adjectives and adverbs that have a substantial chance of being polar.

44. The method of claim 42, wherein scanning locates candidate words by filtering the communication for words that appear at least a predefined number of times.

45. The method of claim 42, wherein scanning includes a pattern based method that locates adjectives and adverbs that have a substantial chance of being polar and locates candidate words by filtering the communication for words that appear at least a predefined number of times.

46. The method of claim 42 wherein providing with a graphical user interface a user with candidate words for at least one of addition, subtraction or exclusion to the polarity lexicon includes presenting each candidate word to the user with the word'"'"'s part of speech label and an example of that candidate word appearing in at least one electronic message collected for the topical domain.

47. A computer implemented method for obtaining topical sentiments from a plurality of electronically stored communications comprising:

determining with the assistance of a computer whether each communication in a plurality of communications is topical to a first topic;

for each communication determined to be topical to the first topic, determining with the assistance of a computer at least one topical expression in the communication;

locating with the assistance of a computer one or more polar expressions in the topical expressions; and

reporting an expression that is both a topical expression and a polar expression as containing a topical sentiment.

48. The computer implemented method of claim 47, wherein determining at least one topical expression and locating at least one polar expression are performed on the same communication.

49. The computer implemented method of claim 48, wherein determining at least one topical expression includes applying an automated text classifier on the communication and wherein locating at least one polar expression includes utilizing a domain-specific lexicon and shallow NLP techniques.