Media tag recommendation technologies

• US 8,706,674 B2
• Filed: 06/29/2012
• Issued: 04/22/2014
• Est. Priority Date: 03/03/2009
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Yang, Linjun, Wu, Lei, Hua, Xian-Sheng
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): OLUDE AFOLABI, OLATOYOSI
• Application Number:
  US13/537,802
• Publication Number:
  US 20120265772A1
• Time in Patent Office: 662 Days
• Field of Search: 706/12, 707/4, 707/9
• US Class Current: 706/50
• CPC Class Codes: G06F 16/958 Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
• CPC Class Codes: G06Q 10/10 Office automation; Time management

Abstract

Technologies for recommending relevant tags for the tagging of media based on one or more initial tags provided for the media and based on a large quantity of other tagged media. Sample media as candidates for recommendation are provided by a set of weak rankers based on corresponding relevance measures in semantic and visual domains. The various samples provided by the weak rankers are then ranked based on relative order to provide a list of recommended tags for the media. The weak rankers provide sample tags based on relevance measures including tag co-occurrence, tag content correlation, and image-conditioned tag correlation.

Claims

1. A method comprising:

generating, by a computer, a first ranking feature based on a first measure of semantic correlation between a first media collection corresponding to a first tag and a second media collection corresponding to a second tag that co-occurs with the first tag, the first measure based on tag similarity in a semantic domain;

generating, by the computer, a second ranking feature based on a second measure of visual correlation between the first media collection corresponding to the first tag and the second media collection corresponding to the second tag that co-occurs with the first tag, the second measure based on media instance similarity in a visual domain;

combining the first ranking feature and the second ranking feature, where the combined first and second ranking features indicate a relative order between the first tag and the second tag but do not indicate a distance between the first tag and the second tag.

2. The method of claim 1 where the tag similarity is based on a co-occurrence frequency of the second tag with the first tag, the co-occurrence frequency normalized by a total frequency of the second tag.

3. The method of claim 1 where the tag similarity is based on a number of interactions between the first tag and the second tag divided by a union of the first tag and the second tag.

4. The method of claim 1 where the media instance similarity is based on a visual distance between the first tag and the second tag.

5. The method of claim 1 further comprising generating a third ranking feature based on a third measure of image-conditioned tag correlation between the first media collection corresponding to the first tag and the second media collection corresponding to the second tag that co-occurs with the first tag, the third measure based on media content similarity in a visual domain.

6. The method of claim 5, the combining further comprising combining the third ranking feature with the first ranking feature and the second ranking feature, where the combined first, second, and third ranking features indicate the relative order between the first tag and the second tag but do not indicate the distance between the first tag and the second tag.

7. The method of claim 5 where the media content similarity is based on a visual similarity between first media content corresponding to the first tag and second media content corresponding to the second tag.

8. At least one computer-readable media storing computer-executable instructions that, when executed by a computer, cause the computer to perform a method comprising:

generating a first ranking feature based on a first measure of semantic correlation between a first media collection corresponding to a first tag and a second media collection corresponding to a second tag that co-occurs with the first tag, the first measure based on tag similarity in a semantic domain;

generating a second ranking feature based on a second measure of visual correlation between the first media collection corresponding to the first tag and the second media collection corresponding to the second tag that co-occurs with the first tag, the second measure based on media instance similarity in a visual domain;

combining the first ranking feature and the second ranking feature, where the combined first and second ranking features indicate a relative order between the first tag and the second tag but do not indicate a distance between the first tag and the second tag.

9. The at least one computer-readable media of claim 8 where the tag similarity is based on a co-occurrence frequency of the second tag with the first tag, the co-occurrence frequency normalized by a total frequency of the second tag.

10. The at least one computer-readable media of claim 8 where the tag similarity is based on a number of interactions between the first tag and the second tag divided by a union of the first tag and the second tag.

11. The at least one computer-readable media of claim 8 where the media instance similarity is based on a visual distance between the first tag and the second tag.

12. The at least one computer-readable media of claim 8, the method further comprising generating a third ranking feature based on a third measure of image-conditioned tag correlation between the first media collection corresponding to the first tag and the second media collection corresponding to the second tag that co-occurs with the first tag, the third measure based on media content similarity in a visual domain.

13. The at least one computer-readable media of claim 12, the combining further comprising combining the third ranking feature with the first ranking feature and the second ranking feature, where the combined first, second, and third ranking features indicate the relative order between the first tag and the second tag but do not indicate the distance between the first tag and the second tag.

14. The at least one computer-readable media of claim 12 where the media content similarity is based on a visual similarity between first media content corresponding to the first tag and second media content corresponding to the second tag.

15. A method comprising:

generating, by a computer, a first measure of semantic correlation between a first media collection corresponding to a first tag and a second media collection corresponding to a second tag that co-occurs with the first tag, the first measure based on tag similarity in a semantic domain;

generating, by the computer, a second measure of visual correlation between the first media collection corresponding to the first tag and the second media collection corresponding to the second tag that co-occurs with the first tag, the second measure based on media instance similarity in a visual domain; and

recommending a co-occurring tag that corresponds to the first tag based on the first measure or the second measure.

16. The method of claim 15 where the tag similarity is based on a co-occurrence frequency of the second tag with the first tag, the co-occurrence frequency normalized by a total frequency of the second tag.

17. The method of claim 15 where the tag similarity is based on a number of interactions between the first tag and the second tag divided by a union of the first tag and the second tag.

18. The method of claim 15 where the media instance similarity is based on a visual distance between the first tag and the second tag.

19. The method of claim 15 further comprising generating a third ranking feature based on a third measure of image-conditioned tag correlation between the first media collection corresponding to the first tag and the second media collection corresponding to the second tag that co-occurs with the first tag, the third measure based on media content similarity in a visual domain.

20. The method of claim 19 where the media content similarity is based on a visual similarity between first media content corresponding to the first tag and second media content corresponding to the second tag.