Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Contextual tag inference
Mandel M., Pascanu R., Eck D., Bengio Y., Aiello L., Schifanella R., Menczer F. ACM Transactions on Multimedia Computing, Communications, and Applications7S (1):1-18,2011.Type:Article
Date Reviewed: Jan 30 2012

Tagging digital music, video, images, and other media is typically frustrating. Most people attempt to either manually tag digital music files or use software to do this. At the end of the day, reliable and comprehensive manual tagging is time consuming and rarely efficient. The authors of this paper address the problem of online tagging solutions that are likewise not efficient in returning optimum results. The methods they describe perhaps go a little way toward improving algorithmic solutions for a tag inferencing system such as that developed by [1].

In this paper, Mandel et al. discuss a method to auto-tag or auto-generate tags across media (image, music, video, or other). Their focus is specific in that they discuss how context underpins the auto-tagging model described. Although the authors cite other tagging models relevant to their own deployment of context, it is not entirely clear exactly what context is, and if they have contributed to developing it. Their reference to the decision tree used by [1] sheds some light on how context works. The system seems to involve creating tag nodes, creating relations between tags, developing training data, and adjusting the semantics at all levels. It would have been helpful for the authors to give some examples using context. However, they did report some optimization in the interrogation of their own dataset for this system.

What interests me is their use of Amazon’s Mechanical Turk service to create their own datasets to test their learning algorithms for inference tagging of individual media entity data. The cost of accumulating these data was minimal, and the resultant dataset was apparently rich with human responses and idiosyncratic vocabularies. This seems a convenient and financially viable means of developing datasets.

This paper does not clearly define context for the reader. It does describe the background work done by others in attempts to develop an efficient tagging system, but there is minimal indication that the authors’ contribution will impact tagging data inferencing methods. However, the intelligent auto-tagging system advanced by [1] may gain ground over time with trials such as those advanced by the authors.

Reviewer:  Alyx Macfadyen Review #: CR139799 (1206-0610)
1) Aucouturier, J.; Pachet, F.; Roy, P.; Beuriv, A. Signal + context = better classification. In Proc. of the International Symposium on Music Information Retrieval. Austrian Computer Society, 2007, 425–430.
Bookmark and Share
  Featured Reviewer  
 
Human Information Processing (H.1.2 ... )
 
 
Connectionism And Neural Nets (I.2.6 ... )
 
 
Signal Analysis, Synthesis, And Processing (H.5.5 ... )
 
 
Sound And Music Computing (H.5.5 )
 
Would you recommend this review?
yes
no
Other reviews under "Human Information Processing": Date
The creative mind
Boden M., Basic Books, Inc., New York, NY, 1991. Type: Book (9780465014521)
Dec 1 1991
Logic and information
Devlin K., Cambridge University Press, New York, NY, 1991. Type: Book (9780521410304)
Aug 1 1992
Eye movement analysis system using fundus images
Kawai H., Tamura S., Kani K., Kariya K. Pattern Recognition 19(1): 77-84, 1986. Type: Article
Feb 1 1988
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy