Computer-implemented systems and methods for taxonomy development

• US 9,116,985 B2
• Filed: 12/16/2011
• Issued: 08/25/2015
• Est. Priority Date: 12/16/2011
• Status: Active Grant

• Current Assignee: SAS Institute Incorporated
• Original Assignee: SAS Institute Incorporated
• Inventors: Mills, Bruce Monroe, Haws, John Courtney, Brocklebank, John Clare, Lehman, Thomas Robert
• Primary Examiner(s): Beausoliel, Jr., Robert
• Assistant Examiner(s): Rayyan, Susan F
• Application Number:
  US13/327,949
• Publication Number:
  US 20130159348A1
• Time in Patent Office: 1,348 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri

Abstract

Systems and methods are provided for generating a set of classifiers. A location is determined for each instance of a topic term in a collection of documents. One or more topic term phrases are identified, and one or more sentiment terms within each topic term phrase. Candidate classifiers are identified by parsing words in the one or more topic term phrases, and a colocation matrix is generated. A seed row of the colocation associated with a particular attribute is identified, and distance metrics are determined by comparing each row of the colocation matrix to the seed row. A set of classifiers are generated for the particular attribute, where classifiers in the set of classifiers are selected using the distance metrics.

Claims

1. A computer-implemented method for generating a set of classifiers, comprising:

determining, using one or more data processors, one or more locations of instances of a topic term in a collection of documents;

identifying, using the one or more data processors, one or more topic term phrases by parsing words in the collection of documents, wherein a topic term phrase includes one or more words that appear within a topic threshold distance of a topic term;

identifying, using the one or more data processors, one or more sentiment terms within a topic term phrase;

identifying, using the one or more data processors, one or more candidate classifiers by parsing words in the one or more topic term phrases, wherein a candidate classifier is a word that appears within a sentiment threshold distance of a sentiment term;

generating, using the one or more data processors, a colocation matrix including a plurality of rows, wherein a candidate classifier is associated with a row, and wherein the colocation matrix is generated using the locations of the candidate classifiers as they appear within the collection of documents;

standardizing, using the one or more data processors, each of the plurality of rows by dividing values in each row by a sum of the values in the row;

identifying, using the one or more data processors, a seed row, wherein the seed row is selected from among the plurality of rows, and wherein the seed row is associated with a particular attribute;

determining, using the one or more data processors, distance metrics by comparing rows of the colocation matrix to the seed row; and

generating, using the one or more data processors, a set of classifiers for the particular attribute, wherein classifiers in the set of classifiers are selected using the distance metrics.

2. The method of claim 1, further comprising parsing the collection of documents to identify locations of words within the collection of documents.

3. The method of claim 2, wherein stemmed versions of the words and the locations of the words are stored in one or more data structures.

4. The method of claim 3, wherein parts of speech are associated with the words and stored in the one or more data structures.

5. The method of claim 3, wherein noun groups are identified in the collection of documents, wherein individual noun records of the noun groups are removed from the one or more data structures and replaced by a noun group record.

6. The method of claim 1, wherein the sentiment terms are stored in a data structure stored in a computer-readable memory.

7. The method of claim 1, wherein the candidate classifiers are limited to nouns, verbs, or adjectives.

8. The method of claim 1, further comprising determining appearance counts of each of the candidate classifiers in the collection of documents, wherein the colocation matrix is limited to a top N candidate classifiers having highest appearance counts.

9. The method of claim 1, wherein the colocation matrix identifies a frequency that a particular candidate classifier appears within a colocation threshold of each other candidate classifier in the colocation matrix.

10. The method of claim 9, wherein the colocation threshold is equal to the topic threshold distance.

11. The method of claim 1, wherein a distance metric for a particular candidate classifier is determined based on a Euclidean distance calculation using values from the candidate classifier row in the colocation matrix and corresponding values in the seed row in the colocation matrix.

12. The method of claim 1, wherein candidate classifiers having associated distance metrics within a distance threshold are selected for inclusion in the set of classifiers for the particular attribute.

13. The method of claim 1, wherein the collection of documents includes one or more of:

websites, books, magazines, and social media postings.

14. The method of claim 1, wherein the collection of documents includes more than 1000 documents.

15. The method of claim 1, wherein the set of classifiers are provided to a downstream computer for processing.

16. The method of claim 1, wherein the collection of documents comprises text entered into a computer by a person using an input interface device.

17. The method of claim 1, wherein the topic term is a company name, and wherein the particular attribute is an attribute of the company.

18. The method of claim 17, wherein the particular attribute is a customer service attribute, a product attribute, a product mix attribute, a product price attribute, or a company culture attribute.

19. The method of claim 1, wherein synonyms of selected classifiers are included in the set of classifiers.

20. The method of claim 1, wherein the colocation matrix is stored in one or more data structures store in a computer-readable medium.

21. The method of claim 1, wherein the one or more sentiment terms are identified by comparing words in topic term phrases to words in a sentiment term list.

22. The method of claim 21, wherein the sentiment term list contains words that connote human feeling.

23. The method of claim 21, wherein the sentiment term list includes the terms:

like, love, hate, and dislike.

24. A computer-implemented system for generating a set of classifiers, comprising:

one or more data processors;

a computer-readable medium encoded with instructions for commanding the one or more data processors to execute steps including;

determining one or more locations of instances of a topic term in a collection of documents;

identifying one or more topic term phrases by parsing words in the collection of documents, wherein a topic term phrase includes one or more words that appear within a topic threshold distance of a topic term;

identifying one or more sentiment terms within a topic term phrase;

identifying one or more candidate classifiers by parsing words in the one or more topic term phrases, wherein a candidate classifier is a word that appears within a sentiment threshold distance of a sentiment term;

generating a colocation matrix including a plurality of rows, wherein a candidate classifier is associated with a row, and wherein the colocation matrix is generated using the locations of the candidate classifiers as they appear within the collection of documents;

standardizing, using the one or more data processors, each of the plurality of rows by dividing values in each row by a sum of the values in the row;

identifying a seed row, wherein the seed row is selected from among the plurality of rows, and wherein the seed row is associated with a particular attribute;

determining distance metrics by comparing rows of the colocation matrix to the seed row; and

generating a set of classifiers for the particular attribute, wherein classifiers in the set of classifiers are selected using the distance metrics.

25. The computer-implemented system of claim 24, further comprising parsing the collection of documents to identify locations of words within the collection of documents.

26. The computer-implemented system of claim 25, wherein stemmed versions of the words and the locations of the words are stored in one or more data structures.

27. The computer-implemented system of claim 26, wherein parts of speech are associated with the words and stored in the one or more data structures.

28. The computer-implemented system of claim 26, wherein noun groups are identified in the collection of documents, wherein individual noun records of the noun groups are removed from the one or more data structures and replaced by a noun group record.

29. The computer-implemented system of claim 24, wherein the sentiment terms are stored in a data structure stored in a computer-readable memory.

30. The computer-implemented system of claim 24, wherein the candidate classifiers are limited to nouns, verbs, or adjectives.

31. The computer-implemented system of claim 24, further comprising determining appearance counts of each of the candidate classifiers in the collection of documents, wherein the colocation matrix is limited to a top N candidate classifiers having highest appearance counts.

32. The computer-implemented system of claim 24, wherein the colocation matrix identifies a frequency that a particular candidate classifier appears within a colocation threshold of each other candidate classifier in the colocation matrix.

33. The computer-implemented system of claim 32, wherein the colocation threshold is equal to the topic threshold distance.

34. The computer-implemented system of claim 24, wherein a distance metric for a particular candidate classifier is determined based on a Euclidean distance calculation using values from the candidate classifier row in the colocation matrix and corresponding values in the seed row in the colocation matrix.

35. The computer-implemented system of claim 24, wherein candidate classifiers having associated distance metrics within a distance threshold are selected for inclusion in the set of classifiers for the particular attribute.

36. The computer-implemented system of claim 24, wherein the collection of documents includes one or more of:

websites, books, magazines, and social media postings.

37. The computer-implemented system of claim 24, wherein the collection of documents includes more than 1000 documents.

38. The computer-implemented system of claim 24, wherein the set of classifiers are provided to a downstream computer for processing.

39. The computer-implemented system of claim 24, wherein the collection of documents comprises text entered into a computer by a person using an input interface device.

40. The computer-implemented system of claim 24, wherein the topic term is a company name, and wherein the particular attribute is an attribute of the company.

41. The computer-implemented system of claim 40, wherein the particular attribute is a customer service attribute, a product attribute, a product mix attribute, a product price attribute, or a company culture attribute.

42. The computer-implemented system of claim 24, wherein synonyms of selected classifiers are included in the set of classifiers.

43. The computer-implemented system of claim 24, wherein the colocation matrix is stored in one or more data structures store in a computer-readable medium.

44. The computer-implemented system of claim 24, wherein the one or more sentiment terms are identified by comparing words in topic term phrases to words in a sentiment term list.

45. The computer-implemented system of claim 44, wherein the sentiment term list contains words that connote human feeling.

46. The computer-implemented system of claim 44, wherein the sentiment term list includes the terms:

like, love, hate, and dislike.

47. A non-transitory computer-readable medium encoded with instructions for commanding one or more data processors to execute method for generating a set of classifiers, the method comprising:

determining one or more locations of instances of a topic term in a collection of documents;

identifying one or more topic term phrases by parsing words in the collection of documents, wherein a topic term phrase includes one or more words that appear within a topic threshold distance of a topic term;

identifying one or more sentiment terms within a topic term phrase;

identifying one or more candidate classifiers by parsing words in the one or more topic term phrases, wherein a candidate classifier is a word that appears within a sentiment threshold distance of a sentiment term;

generating a colocation matrix including a plurality of rows, wherein a candidate classifier is associated with a row, and wherein the colocation matrix is generated using the locations of the candidate classifiers as they appear within the collection of documents;

standardizing, using the one or more data processors, each of the plurality of rows by dividing values in each row by a sum of the values in the row;

identifying a seed row, wherein the seed row is selected from among the plurality of rows, and wherein the seed row is associated with a particular attribute;

determining distance metrics by comparing rows of the colocation matrix to the seed row; and

generating a set of classifiers for the particular attribute, wherein classifiers in the set of classifiers are selected using the distance metrics.

48. The non-transitory computer-readable medium of claim 47, further comprising parsing the collection of documents to identify locations of words within the collection of documents.

49. The non-transitory computer-readable medium of claim 48, wherein stemmed versions of the words and the locations of the words are stored in one or more data structures.

50. The non-transitory computer-readable medium of claim 49, wherein parts of speech are associated with the words and stored in the one or more data structures.

51. The non-transitory computer-readable medium of claim 49, wherein noun groups are identified in the collection of documents, wherein individual noun records of the noun groups are removed from the one or more data structures and replaced by a noun group record.

52. The non-transitory computer-readable medium of claim 47, wherein the sentiment terms are stored in a data structure stored in a computer-readable memory.

53. The non-transitory computer-readable medium of claim 47, wherein the candidate classifiers are limited to nouns, verbs, or adjectives.

54. The non-transitory computer-readable medium of claim 47, further comprising determining appearance counts of each of the candidate classifiers in the collection of documents, wherein the colocation matrix is limited to a top N candidate classifiers having highest appearance counts.

55. The non-transitory computer-readable medium of claim 47, wherein the colocation matrix identifies a frequency that a particular candidate classifier appears within a colocation threshold of each other candidate classifier in the colocation matrix.

56. The non-transitory computer-readable medium of claim 55, wherein the colocation threshold is equal to the topic threshold distance.

57. The non-transitory computer-readable medium of claim 47, wherein a distance metric for a particular candidate classifier is determined based on a Euclidean distance calculation using values from the candidate classifier row in the colocation matrix and corresponding values in the seed row in the colocation matrix.

58. The non-transitory computer-readable medium of claim 47, wherein candidate classifiers having associated distance metrics within a distance threshold are selected for inclusion in the set of classifiers for the particular attribute.

59. The non-transitory computer-readable medium of claim 47, wherein the collection of documents includes one or more of:

websites, books, magazines, and social media postings.

60. The non-transitory computer-readable medium of claim 47, wherein the collection of documents includes more than 1000 documents.

61. The non-transitory computer-readable medium of claim 47, wherein the set of classifiers are provided to a downstream computer for processing.

62. The non-transitory computer-readable medium of claim 47, wherein the collection of documents comprises text entered into a computer by a person using an input interface device.

63. The non-transitory computer-readable medium of claim 47, wherein the topic term is a company name, and wherein the particular attribute is an attribute of the company.

64. The non-transitory computer-readable medium of claim 63, wherein the particular attribute is a customer service attribute, a product attribute, a product mix attribute, a product price attribute, or a company culture attribute.

65. The non-transitory computer-readable medium of claim 47, wherein synonyms of selected classifiers are included in the set of classifiers.

66. The non-transitory computer-readable medium of claim 47, wherein the colocation matrix is stored in one or more data structures store in a computer-readable medium.

67. The non-transitory computer-readable medium of claim 47, wherein the one or more sentiment terms are identified by comparing words in topic term phrases to words in a sentiment term list.

68. The non-transitory computer-readable medium of claim 67, wherein the sentiment term list contains words that connote human feeling.

69. The non-transitory computer-readable medium of claim 67, wherein the sentiment term list includes the terms:

like, love, hate, and dislike.