Iterative Active Feature Extraction

• US 20140180973A1
• Filed: 12/21/2012
• Published: 06/26/2014
• Est. Priority Date: 12/21/2012
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Lingenfelder, Christoph, Pompey, Pascal, Verscheure, Olivier, Wurst, Michael
• Granted Patent: US 9,292,798 B2
• Time in Patent Office: Days
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/025 Extracting rules from data
• CPC Class Codes: G06N 5/043 Distributed expert systems; Blackboards

Abstract

Techniques for iterative feature extraction using domain knowledge are provided. In one aspect, a method for feature extraction is provided. The method includes the following steps. At least one query to predict at least one future value of a given value series based on a statistical model is received. At least two predictions of the future value are produced fulfilling at least the properties of 1) each being as probable as possible given the statistical model and 2) being mutually divert (in terms of numerical distance measure). A user is queried to select one of the predictions. The user may be queried for textual annotations for the predictions. The annotations may be used to identify additional covariates to create an extended set of covariates. The extended set of covariates may be used to improve the accuracy of the statistical model.

Claims

1. A method for feature extraction, the method comprising the steps of:

a) receiving at least one query to predict at least one future value of a given value series based on a statistical model;

b) producing at least two predictions of the future value fulfilling at least the properties of

     1) each being as probable as possible given the statistical model and

     2) being mutually divert; and

c) querying a user to select one of the predictions.

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

d) querying the user for textual annotations for the predictions;

e) using the annotations to identify additional covariates to create an extended set of covariates; and

f) using the extended set of covariates to improve the accuracy of the statistical model.

3. The method of claim 1, further comprising the step of:

presenting the predictions to the user; and

querying the user to select which of the predictions the user believes is most probable.

4. The method of claim 2, wherein the textual annotations comprise tags.

5. The method of claim 2, further comprising the step of:

repeating the steps a-f in an iterative manner with the extended set of covariates.

6. The method of claim 5, further comprising the step of:

at each iteration, displaying the textual annotations from previous iterations with the predictions.

7. The method of claim 2, further comprising the step of:

obtaining the statistical model.

8. The method of claim 7, wherein the statistical model is obtained from a statistician.

9. The method of claim 1, further comprising the step of:

recommending textual annotations from past interactions.

10. The method of claim 1, wherein the statistical model comprises an auto-regressive model or a complex regression model.

11. The method of claim 1, wherein the predictions of the future value are mutually divert in terms of numerical distance measure.

12. The method of claim 1, wherein the numerical distance measure comprises a Euclidean distance measure.

13. The method of claim 8, further comprising the step of:

providing the textual annotations to the statistician for use in retraining the statistical model.

14. The method of claim 1, wherein the user is a domain expert.

15. The method of claim 1, further comprising the step of:

training the statistical model using a training data set.

16. The method of claim 15, wherein the training data set comprises a plurality of training data points, wherein each of the training data points comprises a set of covariates together with a known prediction.

17. The method of claim 16, wherein the known prediction is gathered from historical data.

18. The method of claim 16, further comprising the steps of:

clustering the training data points based on the covariates to form a set of data clusters; and

using the training data points in each of the data clusters to train the statistical model and thereby produce multiple prediction models.

19. The method of claim 18, further comprising the step of:

using the prediction models to produce the at least two predictions of the future value.

20. The method of claim 16, further comprising the steps of:

creating a regression tree using the training data points, such that each leaf node of the regression tree contains a set of the training data points;

clustering, for each of the leaf nodes separately, true values associated with the training data points to derive a set of prediction clusters for each of the leaf nodes of the regression tree.