Classification For Peer-To-Peer Collaboration

• US 20080077579A1
• Filed: 09/21/2007
• Published: 03/27/2008
• Est. Priority Date: 09/22/2006
• Status: Abandoned Application

• Current Assignee: LeapTag, Inc.
• Original Assignee: LeapTag, Inc.
• Inventors: Ozveren, Cuneyt, Kirazci, Ulas
• Application Number:
  US11/859,467
• Publication Number:
  US 20080077579A1
• Time in Patent Office: Days
• US Class Current: 707/5
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers

Abstract

A system and method for classifying content objects. The system includes a database and a classification application. The database is configured to store a plurality of content objects. The classification application is coupled to the database and configured to cluster content objects and implement at least one level of classification, including generating summary vectors formed of weighted sums of object vectors. The object vector includes a vector of numbers representative of a frequency of a superset of features potentially found in the content object.

Claims

1. A system for classifying content objects, the system comprising:

a database to store a plurality of content objects;

a classification application coupled to the database, the classification application to cluster content objects and implement at least one level of classification comprising generating summary vectors formed of weighted sums of object vectors; and

wherein each object vector comprises a vector of numbers representative of a frequency of a superset of features potentially found in the content object.

2. The system of claim 1, wherein the classification application is further configured to classify the content objects using an adaptive graph method to select a cluster of content objects and replace the summary vector with elements associated with the summary vector.

3. The system of claim 1, wherein the classification application is further configured to classify the content objects using a rapid fire method to classify clusters level by level until a leaf level of the cluster is classified.

4. The system of claim 1, further comprising a clustering application coupled to the classification application, the clustering application configured to cluster object vectors into high-level groups.

5. The system of claim 4, wherein the clustering application clusters object vectors into high-level groups using k-means clustering.

6. The system of claim 4, wherein the clustering application clusters object vectors into high-level groups using graph-based clustering.

7. The system of claim 1, wherein the classification application is further configured to classify dynamic content by taking a snapshot of the dynamic content, performing feature extraction on the snapshot of the dynamic content, performing hierarchical classification to select a group of dynamic content for inclusion with the dynamic content, and subsequently and repeatedly taking a snapshot of the dynamic content and performing feature extraction with each inclusion of new dynamic content.

8. The system of claim 1, wherein the classification application is further configured to optimize classification parameters by performing at least one of a maximum percentage classification of positive elements, a weighted sum classification of scores for positive elements at a leaf level, and a number classification of truly positive elements within a predetermined number of highest ranked elements.

9. The system of claim 1, wherein the classification application is further configured to modify negative and positive labeled elements of a training set at each level of a classification hierarchy and to drop at least one negatively labeled element from a one of the levels of the classification hierarchy.

10. A computer program product comprising a computer useable storage medium to store a computer readable program that, when executed on a computer, causes the computer to perform operations for classifying content, the operations comprising:

store a plurality of content objects;

cluster content objects and implement at least one level of classification comprising generating summary vectors formed of weighted sums of object vectors; and

wherein the object vector comprises a vector of numbers representative of a frequency of a superset of features potentially found in the content object.

11. The computer program product of claim 10, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to classify the content objects using an adaptive graph method to select a cluster of content objects and replace the summary vector with elements associated with the summary vector.

12. The computer program product of claim 10, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to classify the content objects using a rapid fire method to classify clusters level by level until a leaf level of the cluster is classified.

13. The computer program product of claim 10, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to cluster object vectors into high-level groups.

14. The computer program product of claim 13, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to cluster object vectors into high-level groups using k-means clustering.

15. The computer program product of claim 13, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to cluster object vectors into high-level groups using graph-based clustering.

16. The computer program product of claim 10, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to classify dynamic content by taking a snapshot of the dynamic content, performing feature extraction on the snapshot of the dynamic content, performing hierarchical classification to select a group of dynamic content for inclusion with the dynamic content, and subsequently and repeatedly taking a snapshot of the dynamic content and performing feature extraction with each inclusion of new dynamic content.

17. The computer program product of claim 10, wherein the computer readable program, when executed on the computer, causes the computer to perform an operation to optimize classification parameters by performing at least one of a maximum percentage classification of positive elements, a weighted sum classification of scores for positive elements at a leaf level, and a number classification of truly positive elements within a predetermined number of highest ranked elements.

18. A method for classifying content, the method comprising:

storing a plurality of content objects;

clustering content objects and implementing at least one level of classification comprising generating summary vectors formed of weighted sums of object vectors; and

wherein the object vector comprises a vector of numbers representative of a frequency of a superset of features potentially found in the content object.

19. The method of claim 18, further comprising classifying the content objects using an adaptive graph method to select a cluster of content objects and replacing the summary vector with elements associated with the summary vector.

20. The method of claim 18, further comprising classifying the content objects using a rapid fire method to classify clusters level by level until a leaf level of the cluster is classified.

21. The method of claim 18, further comprising clustering object vectors into high-level groups.

22. The method of claim 21, further comprising clustering object vectors into high-level groups using k-means clustering.

23. The method of claim 21, further comprising clustering object vectors into high-level groups using graph-based clustering.

24. The method of claim 18, further comprising classifying dynamic content by taking a snapshot of the dynamic content, performing feature extraction on the snapshot of the dynamic content, performing hierarchical classification to select a group of dynamic content for inclusion with the dynamic content, and subsequently and repeatedly taking a snapshot of the dynamic content and performing feature extraction with each inclusion of new dynamic content.

25. The method of claim 18, further comprising optimizing classification parameters by performing at least one of a maximum percentage classification of positive elements, a weighted sum classification of scores for positive elements at a leaf level, and a number classification of truly positive elements within a pre-determined number of highest ranked elements.