Kind classification through emergent semantic analysis

• US 8,341,108 B2
• Filed: 06/09/2009
• Issued: 12/25/2012
• Est. Priority Date: 06/09/2009
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Thompson, Donald, Stojanovic, Alexander S.
• Primary Examiner(s): Chen, Alan
• Assistant Examiner(s): BROWN JR, NATHAN H
• Application Number:
  US12/480,821
• Publication Number:
  US 20100312743A1
• Time in Patent Office: 1,295 Days
• Field of Search: 706/52
• US Class Current: 706/52
• CPC Class Codes: G06F 16/68 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: H04L 51/066 Format adaptation, e.g. format conversion or compression
• CPC Class Codes: H04L 51/10 Multimedia information

Abstract

A goal of computer development is to understand the user and the data with which the user is engaged. If a better understanding of the user and their data can be accomplished, then additional information and features may be provided based upon the user'"'"'s intent and interests. Accordingly, as provided herein, Kinds may be created as digital representations of everyday things. Kind classifications may be created to characterize the Kinds. The Kind classifications may be updated based upon user interaction to further characterize Kinds with which the user has interacted. For example, when a user writes an email about using orange peels as an air freshener, an orange Kind classification may be updated to reflect that an orange may be used as an air freshener. Kind feature vectors and user feature vectors may be created to represent the probabilities that the Kind or user relates to particular characteristics.

Claims

1. A nonvolatile computer-readable medium comprising processor-executable instructions that when executed perform a method for updating a Kind classification based upon emergent semantic analysis comprising:

receiving usage data relating to a Kind classification;

associating a descriptor with the Kind classification based upon the usage data; and

associating a Kind feature vector with a Kind characterized by the Kind classification, the associating comprising;

creating one or more dimensions within the Kind feature vector; and

for respective dimensions within the Kind feature vector, assigning a probabilistic value to a dimension based upon a probability that the Kind relates to a characteristic of the dimension.

2. The nonvolatile computer-readable medium of claim 1, the usage data corresponding to user interaction with a data item associated with the Kind that is characterized by the Kind classification.

3. The nonvolatile computer-readable medium of claim 2, the data item comprising at least one of:

web content;

local content;

subscription content;

or shared content.

4. The nonvolatile computer-readable medium of claim 1, comprising:

creating a user feature vector for a user based upon one or more Kind feature vectors of Kinds associated with the user.

5. The nonvolatile computer-readable medium of claim 4, the creating a user feature vector comprising:

creating one or more dimensions within the user feature vector; and

for respective dimensions within the user feature vector, assigning a probabilistic value to a dimension within the user feature vector based upon the probability the user relates to a characteristic of the dimension.

6. The nonvolatile computer-readable medium of claim 1, the assigning a probabilistic value comprising:

assigning the probabilistic value to the dimension based upon one or more descriptors of the Kind classification.

7. The nonvolatile computer-readable medium of claim 1, comprising:

executing a classification module upon a data item associated with the Kind classification to associate one or more descriptors with the Kind classification.

8. The nonvolatile computer-readable medium of claim 7, the executing comprising:

executing an image Kind classification module upon an image data item associated with an image Kind classification; and

associating one or more descriptors with the image Kind classification based upon at least one of;

a color ratio feature, a color saturation feature, or a shape feature.

9. The nonvolatile computer-readable medium of claim 7, the executing comprising:

executing an audio Kind classification module upon an audio data item associated with an audio Kind classification; and

associating one or more descriptors with the audio Kind classification based upon at least one of;

a signal to noise ratio feature, a zero crossing feature, or an f domain analysis feature.

10. A method for creating a Kind feature vector, comprising:

creating a Kind feature vector for a Kind using a set of initial features and one or more descriptors of a Kind classification characterizing the Kind, the creating comprising;

creating one or more dimensions within the Kind feature vector; and

for respective dimensions within the Kind feature vector, assigning a probabilistic value to a dimension based upon the probability the Kind relates to a characteristic of the dimension.

11. The method of claim 10, comprising:

identifying the set of initial features based upon at least one of;

one or more descriptors within the Kind classification;

or

one or more features extracted from a data item associated with the Kind.

12. The method of claim 10, the assigning a probabilistic value comprising:

assigning the probabilistic value to the dimension based upon one or more descriptors of the Kind classification.

13. The method of claim 10, comprising:

updating the Kind feature vector based upon usage data relating to the Kind classification.

14. The method of claim 13, the updating comprising:

reassigning an updated probabilistic value to a dimension of the Kind feature vector.

15. The method of claim 10, comprising:

creating a user feature vector for a user based upon one or more Kind feature vectors of Kinds associated with the user.

16. The method of claim 15, comprising:

creating one or more dimensions within the user feature vector; and

for respective dimensions within the user feature vector, assigning a probabilistic value to a dimension based upon the probability the user relates to a characteristic of the dimension.

17. The method of claim 16, the assigning a probabilistic value comprising:

performing a weighted sum across the one or more Kind feature vectors to determine the probabilistic value.

18. The method of claim 17, the performing a weighted sum comprising:

summing respective probabilistic values of a dimension across the one or more Kind feature vectors.

19. The method of claim 15, the Kind associated with the user through user interaction with a data item corresponding to the Kind.

20. A system for creating one or more feature vectors, comprising:

a feature vector component configured to;

create a Kind feature vector for a Kind using a set of initial features and one or more descriptors of a Kind classification characterizing the Kind; and

create a user feature vector for a user based upon one or more Kind feature vectors of Kinds associated with the user.