Machine Learning Systems and Methods for Improved Natural Language Processing

• US 20090119095A1
• Filed: 11/04/2008
• Published: 05/07/2009
• Est. Priority Date: 11/05/2007
• Status: Abandoned Application

• Current Assignee: Enhanced Medical Decisions, Inc.
• Original Assignee: Enhanced Medical Decisions, Inc.
• Inventors: Smychkovich, Yuri, Beggelman, Marlene
• Application Number:
  US12/264,668
• Publication Number:
  US 20090119095A1
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Disclosed is a method to generate at least one new set of concepts to be used to perform natural language processing (NLP) on data. The method includes receiving one or more sources of input data, and determining, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the at least one new set of concepts.

Claims

1. A method to generate at least one new set of concepts to be used to perform natural language processing (NLP) on data, the method comprising:

receiving one or more sources of input data; and

determining, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the at least one new set of concepts.

2. The method of claim 1 wherein determining, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the new set of concepts comprises:

comparing attributes of the at least one initial set of concepts to the one or more sources of input data to identify non-matching attributes of the at least one initial set of concepts that do not match any portion of the one or more sources of input data.

3. The method of claim 2 wherein the non-matching attributes of the at least one initial set of concepts includes attributes that do not semantically or syntactically match any portion of the one or more sources of input data.

4. The method of claim 2 further comprising:

identifying non-matching portions of the one or more sources of data that do not match any of the attributes of the at least one initial set of concepts; and

replacing the identified non-matching attributes of the at least one initial set of concepts with the identified non-matching portions of the one or more sources of data to generate the at least one new set of concepts.

5. The method of claim 4 wherein identifying non-matching portions of the one or more sources of data comprises:

removing from the one or more sources of data one or more of;

matching portions that match at least one attribute of the at least one initial set of concepts and semantically neutral portions of the one or more sources of data.

6. The method of claim 1 further comprising:

generating one or more processing rules having one or more search constraints, the one or more processing rules adapted to be applied to input data, the one or more processing rules being associated with the determined at least one attribute representative of the type of information detail to be included in the new set of concepts.

7. The method of claim 6 wherein the one or more processing rules are each associated with one or more groups corresponding to respective levels of rule complexity, wherein at least one group of rules associated with a first level of complexity is a subset of another group of rules associated with another, higher, level of rule complexity.

8. The method of claim 1 further comprising:

identifying from an initial set of processing rules one or more processing rules that produce close matches with the one or more sources of data, each of the processing rules of the initial set including one or more searching constraints; and

modifying at least one of the one or more searching constraints of the identified processing rules.

9. The method of claim 1 wherein determining the at least one attribute representative of the type of information detail to be included in the new set of concepts comprises:

determining the at least one attribute representative of the type of information detail based on one or more inductive bias assumptions.

10. The method of claim 9 wherein the one or more inductive bias assumptions include one or more of:

an assumption that consistency and accuracy of a classification operations applied to the one or more sources of input data increases with increased complexity of processing rules applied to the one or more sources of input data, an assumption of maintaining complexity neutrality, an assumption that semantically related terms are interchangeable, an assumption that concept modifiers do not alter the semantic content of a concept, an assumption that a portion within the one or more sources of data input assigned to a function is not available to be assigned to another function and an assumption that syntax and the choice of semantic expression used in the at least one set of concepts are dependent.

11. The method of claim 9 wherein the one or more inductive bias assumptions comprise forward-chaining rules that specify allowable combinations of data portions within the received one or more sources of input data.

12. The method of claim 9 wherein determining the at least one attribute representative of the type of information detail based on one or more inductive bias assumptions comprises:

determining the at least one attribute representative of the type of information detail based on one or more inductive bias assumptions that process the one or more source of input data in a manner that simulates an operation of human knowledge acquisition and storage.

13. The method of claim 1 wherein the at least one initial set of concepts includes a continuously updated set of skeleton classifiers.

14. The method of claim 13 wherein the skeleton classifiers include skeleton ontologies.

15. The method of claim 1 further comprising:

determining from the at least one initial set of concepts an optimal set of concepts to apply to the one or more sources of data input to generate the new set of concepts.

16. The method of claim 1 further comprising:

importing simple concept terms and synonyms from a remote source maintaining lists of simple concept terms;

constructing an exact term list from the imported concept terms and synonyms; and

generating for the at least one new set of concepts processing rules having search constraints by removing semantically neutral terms in the exact terms list and converting terms remaining in the exact term list to a normalized format.

17. The method of claim 16 further comprising:

augmenting a simple concept lexicon.

18. The method of claim 16 wherein generating processing rules comprises:

applying preposition and verb inflection rules and Placement Rules.

19. The method of claim 1 further comprising constructing a database for the at least one initial set of concepts.

20. The method of claim 19 wherein constructing the database for the at least one initial set of concepts comprises:

forming complex concept terms from an identified remote source of input data.

21. An apparatus, comprising:

a computer system including a processor and memory; and

a computer readable medium storing instructions for natural machine learning (ML) processing including instructions to cause the computer system to;

receive one or more sources of input data; and

determine, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the at least one new set of concepts.

22. The apparatus of claim 21 wherein the instructions that cause the computer system to determine, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the new set of concepts comprise instructions that cause the computer system to:

compare attributes of the at least one initial set of concepts to the one or more sources of input data to identify non-matching attributes of the at least one initial set of concepts that do not match any portion of the one or more sources of input data.

23. The apparatus of claim 22 wherein the non-matching attributes of the at least one initial set of concepts includes attributes that do not semantically or syntactically match any portion of the one or more sources of input data.

24. The apparatus of claim 22 wherein the instructions further comprise instructions that cause the computer system to:

identify non-matching portions of the one or more sources of data that do not match any of the attributes of the at least one initial set of concepts; and

replace the identified non-matching attributes of the at least one initial set of concepts with the identified non-matching portions of the one or more sources of data to generate the at least one new set of concepts.

25. The apparatus of claim 24 wherein the instructions that cause the computer system to identify non-matching portions of the one or more sources of data comprise instructions that cause the computer system to:

remove from the one or more sources of data one or more of;

matching portions that match at least one attribute of the at least one initial set of concepts and semantically neutral portions of the one or more sources of data.

26. The apparatus of claim 21 wherein the instructions further comprise instructions that cause the computer system to:

generate one or more processing rules having one or more search constraints, the one or more processing rules adapted to be applied to input data, the one or more processing rules being associated with the determined at least one attribute representative of the type of information detail to be included in the new set of concepts.

27. The apparatus of claim 21 wherein the instructions further comprise instructions that cause the computer system to:

identify from an initial set of processing rules one or more processing rules that produce close matches with the one or more sources of data, each of the processing rules of the initial set including one or more searching constraints; and

modify at least one of the one or more searching constraints of the identified processing rules.

28. The apparatus of claim 21 wherein the computer instructions that cause the computer system to determine the at least one attribute representative of the type of information detail to be included in the new set of concepts comprise instructions that cause the computer system to:

determine the at least one attribute representative of the type of information detail based on one or more inductive bias assumptions.

29. The apparatus of claim 28 wherein the one or more inductive bias assumptions include one or more of:

an assumption that consistency and accuracy of a classification operations applied to the one or more sources of input data increases with increased complexity of processing rules applied to the one or more sources of input data, an assumption of maintaining complexity neutrality, an assumption that semantically related terms are interchangeable, an assumption that concept modifiers do not alter the semantic content of a concept, an assumption that a portion within the one or more sources of data input assigned to a function is not available to be assigned to another function and an assumption that syntax and the choice of semantic expression used in the at least one set of concepts are dependent.

30. The apparatus of claim 21 wherein the at least one initial set of concepts includes a continuously updated set of skeleton classifiers.

31. The apparatus of claim 30 wherein the skeleton classifiers include skeleton ontologies.

32. The apparatus of claim 21 wherein the instructions further comprise instructions to cause the computer system to:

determine from the at least one initial set of concepts an optimal set of concepts to apply to the one or more sources of data input to generate the new set of concepts.

33. The apparatus of claim 21 wherein the instructions further comprise instructions to cause the computer system to:

import simple concept terms and synonyms from a remote source maintaining lists of simple concept terms;

construct an exact term list from the imported concept terms and synonyms; and

generate for the at least one new set of concepts processing rules having search constraints by removing semantically neutral terms in the exact terms list and converting terms remaining in the exact term list to a normalized format.

34. A computer program product residing on a computer readable medium for machine learning (ML) processing, the computer program product comprising instructions to cause a computer to:

receive one or more sources of input data; and

determine, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the at least one new set of concepts.

35. The computer program product of claim 34 wherein the instructions that cause the computer to determine, based on the one or more sources of input data and on at least one initial set of concepts, at least one attribute representative of a type of information detail to be included in the new set of concepts comprise instructions that cause the computer to:

compare attributes of the at least one initial set of concepts to the one or more sources of input data to identify non-matching attributes of the at least one initial set of concepts that do not match any portion of the one or more sources of input data.

36. The computer program product of claim 35 wherein the non-matching attributes of the at least one initial set of concepts includes attributes that do not semantically or syntactically match any portion of the one or more sources of input data.

37. The computer program product of claim 35 wherein the instructions further comprise instructions that cause the computer to:

identify non-matching portions of the one or more sources of data that do not match any of the attributes of the at least one initial set of concepts; and

replace the identified non-matching attributes of the at least one initial set of concepts with the identified non-matching portions of the one or more sources of data to generate the at least one new set of concepts.

38. The computer program product of claim 37 wherein the instructions that cause the computer to identify non-matching portions of the one or more sources of data comprise instructions that cause the computer to:

remove from the one or more sources of data one or more of;

matching portions that match at least one attribute of the at least one initial set of concepts and semantically neutral portions of the one or more sources of data.

39. The computer program product of claim 34 wherein the instructions further comprise instructions that cause the computer to:

generate one or more processing rules having one or more search constraints, the one or more processing rules adapted to be applied to input data, the one or more processing rules being associated with the determined at least one attribute representative of the type of information detail to be included in the new set of concepts.

40. The computer program product of claim 34 wherein the instructions further comprise instructions that cause the computer to:

identify from an initial set of processing rules one or more processing rules that produce close matches with the one or more sources of data, each of the processing rules of the initial set including one or more searching constraints; and

modify at least one of the one or more searching constraints of the identified processing rules.

41. The computer program product of claim 34 wherein the computer instructions that cause the computer to determine the at least one attribute representative of the type of information detail to be included in the new set of concepts comprise instructions that cause the computer to:

determine the at least one attribute representative of the type of information detail based on one or more inductive bias assumptions.

42. The computer program product of claim 41 wherein the one or more inductive bias assumptions include one or more of:

an assumption that consistency and accuracy of a classification operations applied to the one or more sources of input data increases with increased complexity of processing rules applied to the one or more sources of input data, an assumption of maintaining complexity neutrality, an assumption that semantically related terms are interchangeable, an assumption that concept modifiers do not alter the semantic content of a concept, an assumption that a portion within the one or more sources of data input assigned to a function is not available to be assigned to another function and an assumption that syntax and the choice of semantic expression used in the at least one set of concepts are dependent.

43. The computer program product of claim 34 wherein the at least one initial set of concepts includes a continuously updated set of skeleton classifiers.

44. The computer program product of claim 43 wherein the skeleton classifiers include skeleton ontologies.

45. The computer program product of claim 34 wherein the instructions further comprise instructions to cause the computer to:

determine from the at least one initial set of concepts an optimal set of concepts to apply to the one or more sources of data input to generate the new set of concepts.

46. The computer program product of claim 34 wherein the instructions further comprise instructions to cause the computer to:

import simple concept terms and synonyms from a remote source maintaining lists of simple concept terms;

construct an exact term list from the imported concept terms and synonyms; and

generate for the at least one new set of concepts processing rules having search constraints by removing semantically neutral terms in the exact terms list and converting terms remaining in the exact term list to a normalized format.