Method and system for the dynamic analysis of data

• US 6,636,862 B2
• Filed: 07/05/2001
• Issued: 10/21/2003
• Est. Priority Date: 07/05/2000
• Status: Expired due to Fees

• Current Assignee: Camo, Inc.
• Original Assignee: Camo, Inc.
• Inventors: Lundahl, David S., Plaehn, David C.
• Primary Examiner(s): Mizrahi, Diane D.
• Assistant Examiner(s): Mofiz, Apu M
• Application Number:
  US09/899,227
• Publication Number:
  US 20020107858A1
• Time in Patent Office: 838 Days
• Field of Search: 707/1, 707/6, 707/10, 707/5, 707/101, 707/102, 709/217, 250/339.12, 704/9
• US Class Current: 1/1
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06F 18/23 Clustering techniques
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: Y10S 707/99942 Manipulating data structure, e.g. compression, compaction, compilation

Abstract

In a method and system for the dynamic analysis of data represented in distinct matrices, if two data matrices X and Y are present in which corresponding rows of X and Y each refer to the same underlying object, a relationship can developed between the X and Y data matrices, which allows for a prediction of responses in Y on the basis of inputted X-data. And, if a third data matrix Z is present in which corresponding columns of Y and row of Z each refer to the same underlying object, a relationship can developed between the X Y and Z data matrices, which allows for link between X and Z through Y.

Claims

1. A method for the dynamic analysis of a first data matrix with a data structure comprised of row objects and column objects, wherein the row objects of said first data matrix can be related to a second data matrix with a data structure comprised of the same row objects and independent column objects, comprising the steps of:

performing a cluster analysis on said first data matrix to segment said data matrix into two or more row clusters;

creating two or more cluster membership prediction models, each of said cluster membership prediction models defining the relationship of the second matrix to a particular row cluster of the first data matrix, such that subsequent submission of data representative of an unknown row object with the data structure of the second matrix allows for a prediction of the particular row cluster of the first data matrix; and

dynamically updating said at least two cluster membership prediction models with one or more new row objects having the column data structure of said first data and second matrices, provided that said new row objects are not identified as outliers, and without performing a new cluster analysis on the existing row objects of said first data matrix.

2. The method as recited in claim 1, in which said cluster analysis comprises the following substeps:

generating at least distinct two cluster analysis models, each of which segment said first data matrix into two or more row clusters; and

identifying the cluster analysis model which most accurately predicts whether an unknown row object with the data structure of said second matrix has membership in particular row cluster or is an outlier.

3. The method as recited in claim 2, and further comprising the step of:

dynamically updating said at least two cluster analysis models with one or more new row objects having the data structure of said first data matrix, provided that said new row objects are not identified as outliers.

4. The method as recited in claim 1, and further comprising the step of:

identifying the cluster membership prediction model which most accurately predicts the particular row cluster of the first data matrix based on the submission of data representative of an unknown row object with the data structure of the second matrix.

5. The method as recited in claim 1, and further comprising the step of:

creating two or more value prediction models, each of said value prediction models defining the relationship of the second matrix to said first data matrix, such that subsequent submission of column values in said second data matrix of a row object allows for a prediction of the column values in said first data matrix.

6. The method as recited in claim 5, and further comprising the step of:

dynamically updating said at least two value prediction models prediction models with one or more new row objects having the column data structure of said first data and second matrices, provided that said new row objects are not identified as outliers.

7. The method as recited in claim 5, in which said computational steps are accomplished through execution of a digital computer program, each of said data matrices and subsequent data submissions being entered into said digital computer program.

8. The method as recited in claim 7, in which said value prediction models are accessed via a computer network, thereby allowing an end user to submit data representative of column values in said second data matrix of a row object allows for a prediction of the column values in said first data matrix.

9. The method as recited in claim 1, in which said computational steps are accomplished through execution of a digital computer program, each of said data matrices and subsequent data submissions being entered into said digital computer program.

10. The method as recited in claim 9, in which said cluster membership prediction models are accessed via a computer network, thereby allowing an end user to submit data representative of an unknown row object with the data structure of the second matrix for a prediction of the particular row cluster of the first data matrix.

11. A method as recited in claim 1, wherein:

each of the row objects of said first data matrix is a consumer respondent to a research study; and

each of the column objects of said first data matrix is a product concept statement that one or more of the consumer respondents have evaluated.

12. A method as recited in claim 1, wherein:

each of the row objects of said first data matrix is a consumer respondent to a research study; and

each of the column objects of said first data matrix is an actual consumer product that one or more of the consumer respondents have evaluated.

13. A method for the dynamic analysis of a first data matrix with a data structure comprised of row objects and column objects, wherein the row objects of said first data matrix can be related to a second data matrix with a data structure comprised of the same row objects and independent column objects, and wherein the column objects of said first data matrix can be related to a third data matrix with a data structure comprised of the same column objects and independent row objects, comprising the steps of:

generating a family of piecewise, non-linear prediction models defining the response surface model relationships between said second and third data matrices and said first data matrix over respective domain spaces covering possible values from the combinations of said row objects and said column objects;

using at least one of the piecewise, non-linear prediction models to selectively predict;

an optimal value of the column objects of the second matrix, or

an optimal value of the row objects of said third data matrix, that maximizes or minimizes a predicted scalar value of said first data matrix while holding constant one or more selected values of the column objects of said second matrix;

or one or more selected values of the column objects of said third matrix;

dynamically generating new piecewise, linear models extending the domain space of possible values of the combinations of the column objects of said second data matrix and the row objects of said third data matrix with new data having;

the same column data structure of said second data matrix,

the same row data structure of said third data matrix, and

a response value representing the same measure from the elements of said first data matrix; and

dynamically updating at least one of the piecewise, non-linear prediction models with new data having;

the same column data structure of said second data matrix,

the same row data structure of said third data matrix, and

a response value representing the same measure from the elements of said first data matrix.

14. A method as recited in claim 13, and further comprising the steps of:

generating at least two distinct two cluster analysis models, each of which segment said first data matrix into two or more row clusters;

identifying the cluster analysis model which most accurately predicts whether an unknown row object with the data structure of said second matrix has membership in particular row cluster or is an outlier;

using one or more of said cluster analysis models to predict the cluster membership of an unknown row object with the data structure of said second matrix has membership in particular row cluster or to identify said unknown row object as an outlier;

using said cluster membership prediction in the generation of said piecewise, non-linear response surface models; and

dynamically updating said cluster analysis models with one or more new row objects having the same column data structure as said first data matrix, provided that said new row objects are not identified as outliers, and without performing a new cluster analysis on the existing row objects of said first data matrix.

15. A method as recited in claim 13, wherein:

each of the row objects of said first data matrix is a consumer respondent to a research study; and

each of the column objects of said first data matrix is a product concept statement that one or more of the consumer respondents have evaluated.

16. A method as recited in claim 15, wherein:

said first data matrix represents common numerical measures from the evaluation of each product concept statement by each consumer respondent;

said second data matrix represents common numerical measures that describe each consumer respondent; and

said third data matrix represents common numerical measures that describe the evaluated product concept statements.

17. A method as recited in claim 13, wherein:

each of the row objects of said first data matrix is a consumer respondent to a research study; and

each of the column objects of said first data matrix is an actual consumer product that one or more of the consumer respondents have evaluated.

18. A method as recited in claim 17, wherein:

said first data matrix represents common numerical measures from the evaluation of each actual consumer product by each consumer respondent;

said second data matrix represents common numerical measures that describe each consumer respondent; and

said third data matrix represents common numerical measures that describe the evaluated actual consumer products.

19. The method as recited in claim 13, in which said computational steps are accomplished through execution of a digital computer program, each of said data matrices and subsequent data submissions being entered into said digital computer program.

20. The method as recited in claim 19, in which said value prediction models are accessed via a computer network, thereby allowing an end user to submit data representative of column values in said second data matrix of a row object allows for a prediction of the column values in said first data matrix.