Methods for discovering and classifying situations that influence affective response

• US 9,076,108 B2
• Filed: 06/25/2011
• Issued: 07/07/2015
• Est. Priority Date: 05/11/2011
• Status: Active Grant

• Current Assignee: Affectomatics ltd.
• Original Assignee: Anat Thieberger Ben-Haim, Ari M. Frank, Gil Thieberger
• Inventors: Frank, Ari M., Thieberger, Gil, Thieberger Ben-Haim, Anat
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): FINK, THOMAS ANDREW
• Application Number:
  US13/168,968
• Publication Number:
  US 20120290516A1
• Time in Patent Office: 1,473 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 3/013 Eye tracking input arrangements G06F3/015 takes precedence
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation

Abstract

Described herein are methods for identifying situations. The methods receive samples, each comprising a temporal window of token instances to which a user was exposed and an affective response annotation. One embodiment clusters the samples into a plurality of clusters utilizing a distance function that computes a distance between a pair comprising first and second samples. Another embodiment utilizes an Expectation-Maximization approach to assign situation identifiers. And still another embodiment involves training, utilizing the samples, a machine learning-based classifier to assign situation identifiers.

Claims

1. A method for identifying situations, comprising:

receiving temporal windows of token instances to which a user was exposed and affective response annotations;

wherein the temporal windows of token instances comprise token instances that have overlapping instantiation periods;

generating samples from the windows and annotations;

clustering the samples into a plurality of clusters utilizing a distance function that computes a distance between a pair comprising first and second samples;

wherein the distance function assigns weight greater than zero to difference between token instances belonging to the first and second samples, and to difference between the affective response annotations belonging to the first and second samples;

assigning situation identifiers to the samples corresponding to the clusters; and

utilizing the samples and their corresponding situation identifiers to train a model that may be used by a machine learning-based predictor of situations;

whereby the machine learning-based predictor receives a sample comprising a temporal window of token instances and an affective response annotation, and utilizes the model to predict a situation identifier corresponding to the sample.

2. The method of claim 1, wherein the machine learning-based predictor comprises at least one of:

a neural network, a support vector machine, a random forest, and a decision tree.

3. The method of claim 1, further comprising clustering the samples into the plurality of clusters utilizing at least one of the following clustering approaches:

partitioning clustering and a hierarchical bottom-up clustering.

4. The method of claim 1, further comprising receiving a number of desired clusters and clustering the samples into the desired number of clusters.

5. The method of claim 1, wherein the affective response annotations comprise at least one of:

values representing emotional responses, values of a user measurement channel of the user, baselines value of an emotional response of the user, and baselines value of a user measurement channel of the user.

6. The method of claim 1, further comprising receiving situation identifiers for at least some of the samples, and clustering the samples utilizing an Expectation-Maximization algorithm.

7. The method of claim 1, wherein different samples corresponding to different clusters have at least one of:

a different characteristic sets of token instances, and different affective response annotations.

8. A method for training a model of situations, comprising:

receiving temporal windows of token instances to which a user was exposed and corresponding affective response annotations;

wherein the temporal windows of token instances comprises token instances that have overlapping instantiation periods;

generating samples from the windows and the annotations;

receiving situation identifiers corresponding to at least some, but not all, of the samples;

wherein the situation identifiers correspond to a certain number of situations; and

training a model with the samples and the situation identifiers utilizing an Expectation-Maximization approach;

wherein the model describes probabilities of the samples corresponding to situations from the certain number of situations.

9. The method of claim 8, further comprising utilizing the model to label samples with situations identifiers.

10. The method of claim 8, further comprising utilizing the samples and their corresponding situation identifiers for training a situation model that may be used by a machine learning-based predictor of situations;

wherein the machine learning-based predictor receives a sample comprising a temporal window of token instances and an affective response annotation, and utilizes the situation model to predict a situation identifier corresponding to the sample.

11. The method of claim 10, wherein the machine learning-based predictor comprises at least one of:

a neural network, a support vector machine, a random forest, and a decision tree.

12. The method of claim 8, wherein the affective response annotations comprise at least one of:

values representing emotional responses, values of a user measurement channel of the user, baselines value of an emotional response of the user, and baselines value of a user measurement channel of the user.

13. The method of claim 8, wherein certain number of situations comprise situations that are characterized by having corresponding samples comprising different characteristic sets of token instances.

14. The method of claim 8, wherein certain number of situations comprise situations that are characterized by having corresponding samples comprising different affective response annotations.

15. A method for training a classifier of situations, comprising:

receiving temporal windows of token instances to which a user was exposed and corresponding affective response annotations;

wherein the temporal windows of token instances comprise token instances that have overlapping instantiation periods;

generating samples from the windows and the annotations;

receiving situation identifiers corresponding to the samples;

wherein the situations identifiers correspond to a plurality of different situations; and

utilizing the samples and the situation identifiers to train a classifier;

wherein the classifier receives a sample comprising a temporal window of token instances and an affective response annotation, and selects a situation identifier corresponding to the sample.

16. The method of claim 15, wherein different situations are characterized by having corresponding samples comprising different characteristic sets of token instances.

17. The method of claim 15, wherein different situations are characterized by having corresponding samples comprising different affective response annotations.

18. The method of claim 15, wherein the classifier comprises at least one of:

a neural network, a support vector machine, a random forest, and a decision tree.

19. The method of claim 15, wherein the affective response annotations comprise at least one of:

values representing emotional responses, values of a user measurement channel of the user, baselines value of an emotional response of the user, and baselines value of a user measurement channel of the user.

20. The method of claim 15, further comprising receiving additional samples without corresponding situation identifiers, and utilizing a semi-supervised training approach for training the classifier based on the samples, the situation identifiers, and the additional samples.