TAG RANKING

• US 20100250190A1
• Filed: 03/31/2009
• Published: 09/30/2010
• Est. Priority Date: 03/31/2009
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Zhang, HongJiang, Hua, Xian-Sheng, Liu, Dong, Wang, Meng, Yang, Linjun
• Granted Patent: US 8,175,847 B2
• Time in Patent Office: Days
• US Class Current: 702/181
• CPC Class Codes: G06F 16/48 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually

Abstract

Technologies for generating a boosted tag ranking for a media instance, the boosted tag ranking based on probabilistic relevance estimation and tag correlation refining. Such boosted tag rankings may be used for search result ranking, tag recommendation, and group recommendation.

Claims

1. A system for generating a boosted tag ranking for a media instance, the system comprising:

a probabilistic relevance estimator operable to estimate a relevance score for each input tag of the target media instance, wherein each relevance score is associated with a corresponding input tag, wherein the relevance scores are ordered by relevance thus forming a relevance tag ranking; and

a tag correlation refiner operable to refine the relevance tag ranking based on tag similarity measurements derived from media instances that share a common tag with the media instance wherein the tag similarity measurements are based on tag exemplar similarity and tag concurrence similarity, wherein the media instances are comprised by at least one media repository.

2. The system of claim 1 wherein the tag exemplar similarity between tags t_i and t_j is defined as:

3. The system of claim 2 wherein the radius parameter σ

is the median value of all pair-wise Euclidian distances between the media instances.

4. The system of claim 2 wherein the tag concurrence similarity between tags t_i and t_j is computed in part by estimating the distance between the tags t_i and t_j:

5. The system of claim 4 wherein the concurrence similarity between tags t_i and t_j is defined as:

φ

_c(t₁,t_j)=exp(−

d(t₁,t_j)).

6. The system of claim 5 wherein the tag exemplar similarity and tag concurrence similarity between two tags t_i and t_j are combined as similarity value s_ij.

7. The system of claim 6 wherein the similarity value s_ij is defined as:

s_ij=s(t_i,t_j)=λ

·

φ

_e(t_i,t_j)+(1−

λ

)·

φ

_c(t_i,t_j)

where λ

belongs to [0, 1].

8. The system of claim 7 wherein the tag correlation refiner constructs a tag graph wherein an edge between two nodes of the tag graph, wherein the two nodes represent the two tags t_i and t_j, is weighted based on the similarity value s_ij.

9. The system of claim 8 wherein a random walk over the tag graph is performed, the random walk operable to promote tags of the relevance tag ranking that have many near neighbors and demote isolated tags of the relevance tag ranking.

10. The system of claim 1 wherein the media instance is a digital image.

11. The system of claim 1 wherein the media instance is a digital video.

12. The system of claim 1 wherein the each input tag is taken from surrounding text of a document including the media instance.

13. The system of claim 1 wherein a search result ranking is generated for each media instance returned in a search.

14. The system of claim 1 wherein a group recommendation is generated for the media instance based in part on keywords associated with various groups.

15. The system of claim 1 wherein a tag recommendation is generated for the media instance and wherein the each input tag is derived from neighbors of the media instance.

16. A method for generating a boosted tag ranking for a media instance, the method comprising:

estimating, by a probabilistic relevance estimator of a computer, a relevance score for each input tag of the target media instance, wherein each relevance score is associated with a corresponding input tag, wherein the relevance scores are ordered by relevance thus forming a relevance tag ranking; and

refining, by a tag correlation refiner of the computer, the relevance tag ranking based on tag similarity measurements derived from media instances that share a common tag with the media instance wherein the tag similarity measurements are based on tag exemplar similarity and tag concurrence similarity, wherein the media instances are comprised by at least one media repository.

17. The method of claim 16 wherein a search result ranking is generated for each media instance returned in a search.

18. The method of claim 16 wherein a group recommendation is generated for the media instance based in part on keywords associated with various groups.

19. The method of claim 16 wherein a tag recommendation is generated for the media instance and wherein the each input tag is derived from neighbors of the media instance.

20. One or more computer-readable media embodying computer-executable instructions that, when executed by a processor of a computer, cause the computer to perform a method for generating a boosted tag ranking for a media instance, the method comprising:

estimating, by a probabilistic relevance estimator of the computer, a relevance score for each input tag of the target media instance, wherein each relevance score is associated with a corresponding input tag, wherein the relevance scores are ordered by relevance thus forming a relevance tag ranking; and

refining, by a tag correlation refiner of the computer, the relevance tag ranking based on tag similarity measurements derived from media instances that share a common tag with the media instance wherein the tag similarity measurements are based on tag exemplar similarity and tag concurrence similarity, wherein the media instances are comprised by at least one media repository.