Knowledge discovery method with utility functions and feedback loops

• US 7,333,997 B2
• Filed: 08/12/2003
• Issued: 02/19/2008
• Est. Priority Date: 08/12/2003
• Status: Expired due to Fees

• Current Assignee: Viziant Corp.
• Original Assignee: Viziant Corp.
• Inventors: Campbell, Stanley V., Maren, Alianna J.
• Primary Examiner(s): Pham; Khanh B.
• Application Number:
  US10/604,705
• Publication Number:
  US 20050038805A1
• Time in Patent Office: 1,652 Days
• Field of Search: 707/1, 707 5- 6, 707/10, 707100-101, 706/45
• US Class Current: 1/1
• CPC Class Codes: G06F 16/435 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06F 2216/03 Data mining
• CPC Class Codes: Y10S 707/99936 Pattern matching access
• CPC Class Codes: Y10S 707/99942 Manipulating data structure, e.g. compression, compaction, compilation

Abstract

A knowledge discovery apparatus and method that extracts both specifically desired as well as pertinent and relevant information to query from a corpus of multiple elements that can be structured, unstructured, and/or semi-structured, along with imagery, video, speech, and other forms of data representation, to generate a set of outputs with a confidence metric-applied to the match of the output against the query. The invented apparatus includes a multi-level architecture, along with one or more feedback loop(s) from any Level N to any lower Level so that a user can control the output of this knowledge discovery method via providing inputs to the utility function.

Claims

1. A method for performing knowledge discovery comprising the steps of:

associating with each member of a data corpus one or more metatags through execution of a ranking function, wherein said ranking function is controllable through a parameter value;

selecting a first subset of members from said data corpus whose associated metatags are a match to a first set of criteria;

processing said first subset of members to produce a set of pairwise associations between elements of each of said first subset of members;

selecting a subset of said set of pairwise associations that reach a certain predefined or preset value;

identifying a second subset of members from said data corpus based on said subset of said pairwise associations;

computing a utility function to measure a utility of said second subset of members;

adjusting said parameter value to an adjusted parameter value based on said utility function;

associating one or more members of said data corpus with one or more retrospective metatags through execution of said ranking function controlled by said adjusted parameter value; and

selecting a third subset of members from said data corpus whose associated metatags are a match to a second set of criteria.

2. The method of claim 1, wherein said one or more metatags describe content of a respective member of said data corpus.

3. The method of claim 1, wherein said one or more metatags represent one or more concept classes contained in a respective member of said data corpus.

4. The method of claim 3, wherein said one or more concept classes are inferred through the presence of one or more feature vectors in one or more respective members of said data corpus.

5. The method of claim 1, wherein said first set of criteria are selected by a user.

6. The method of claim 1, wherein said pairwise associations are pairwise associations between concepts and further comprising the step of calculating a value, for each pairwise association, representative of the number of times that each respective pairwise association occurs across said second subset of members.

7. The method of claim 1, wherein said elements are nouns and/or noun phrases.

8. The method of claim 1, further comprising the steps of:

generating a second set of pairwise associations between elements of said third subset of members; and

calculating a value, for each pairwise association, representative of the number of times that each respective pairwise association occurs across said third subset of members.

9. The method of claim 8, further comprising the step of changing said first set of criteria to said second set of criteria based on said utility function.

10. The method of claim 8, wherein said elements are nouns and/or noun phrases and further comprising the step of identifying pairwise associations having a respective said value reaching a predetermined threshold.

11. The method of claim 10, further comprising the step of extracting elements of said third subset of members associated with said identified pairwise associations having a respective said second value reaching said predetermined threshold.

28. The method of claim 1, wherein said third subset of members is greater in number than said first set of members.

29. The method of claim 1, wherein one or more members of said data corpus are images.

30. The method of claim 1, wherein said parameter value comprises a vector of parameters.

31. The method of claim 1, further comprising the step of calculating a value, for each pairwise association, representative of the number of times that each respective pairwise association occurs across said first subset of members.

12. A method for performing knowledge discovery, the method comprising the steps of:

determining a first degree of correlation among a data corpus;

metatagging members of said data corpus with metatags according to a metatagging scheme, wherein said metatagging scheme employs a first level of knowledge representation for said first degree of correlation and employs at least a second level of knowledge representation for a second degree of correlation among data, wherein said step of metatagging is controllable through a parameter value, and wherein said first and second levels of knowledge representation are representative of different degrees of correlation among data;

determining said second degree of correlation among a first subset of said data corpus;

identifying a second subset of members from said data corpus based on said second degree of correlation among said first subset of said data corpus;

computing a utility function to measure a utility of said second subset of members;

adjusting said parameter value to an adjusted parameter value based on said utility function;

remetatagging said members of said data corpus with metatags according to said metatagging scheme, wherein said step of remetatagging is controlled through said adjusted parameter value and;

selecting a third subset of members from said data corpus whose associated metatags are a match to a first set of criteria.

13. The method of claim 12, further comprising the step of selecting said first subset of members from said data corpus whose metatags are a match to a second set of criteria.

14. The method of claim 13, further comprising the step of selecting a second subset of said data corpus whose metatags are a match to said second set of criteria.

15. The method of claim 12, wherein said step of determining a first degree of correlation comprises the step of identifying one or more concept classes contained in a respective member of said data corpus.

16. The method of claim 12, wherein said step of determining a first degree of correlation comprises the step of generating a set of pairwise associations of nouns and/or nouns phrases between elements of said first subset of members.

17. The method of claim 16, further comprising the step of calculating a value, for each pairwise association, representative of the number of times that each respective pairwise association occurs across said first subset of members.

18. The method of claim 12, wherein said step of determining a first degree of correlation comprises the step of generating a set of subject noun-verb-object noun associations drawn from a portion of said data corpus.

19. The method of claim 12, wherein said step of determining a first degree of correlation comprises the step of identifying context associations across a portion of said data corpus.

20. The method of claim 15, wherein said step of determining said second degree of correlation comprises the step of generating a set of pairwise associations of nouns and/or nouns phrases between elements of said first subset of members.

21. The method of claim 20, further comprising the step of calculating a value, for each pairwise association, representative of the number of times that each respective pairwise association occurs across said first subset of members.

22. The method of claim 12, wherein said step of determining said second degree of correlation comprises the step of generating a set of subject noun-verb-object noun associations drawn from a portion of said data corpus.

23. The method of claim 12, wherein said step of determining said second degree of correlation comprises the step of identifying context associations across a portion of said data corpus.

24. The method of claim 12, wherein said step of determining said second degree of correlation comprises the step of identifying semantic associations across a portion of said data corpus.

25. The method of claim 12, wherein said first level of knowledge representation is associated with a category of data selected from the group consisting of:

nouns and/or nouns phrases, noun-verb-object noun associations, context associations, or semantic associations.

26. The method of claim 25, wherein said second level of knowledge representation is associated with a category of data selected from the group consisting of:

noun-verb-object noun associations, context associations, or semantic associations.

27. The method of claim 25, wherein said second level of knowledge representation is associated with a higher level of abstraction than said first level of knowledge representation.

32. The method of claim 12, wherein said step of determining a first degree of correlation comprises the step of generating a set of pairwise associations of concepts between elements of said first subset of members.

33. The method of claim 32, further comprising the step of calculating a value, for each pairwise association, representative of the number of times that each respective pairwise association occurs across said first subset of members.