Dong and Fu, Cultural difference in image tagging, SIGCHI, Atlanta, Georgia: ACM, 2010, pp. 981-984

From Cohen Courses
Jump to navigationJump to search

The Paper

ACM Portal Listing

Citation

@inproceedings{1753472,
 author = {Dong, Wei and Fu, Wai-Tat},
 title = {Cultural difference in image tagging},
 booktitle = {CHI '10: Proceedings of the 28th international conference on Human factors in computing systems},
 year = {2010},
 isbn = {978-1-60558-929-9},
 pages = {981--984},
 location = {Atlanta, Georgia, USA},
 doi = {http://doi.acm.org/10.1145/1753326.1753472},
 publisher = {ACM},
 address = {New York, NY, USA},
 }

Abstract from the paper

Do people from different cultures tag digital images differently? The current study compared the content of tags for digital images created by two cultural groups: European Americans and Chinese. In line with previous findings on cultural differences in attentional patterns, we found similar cultural differences in the order of the image parts (e.g., foreground or background objects) that people tag. We found that for European Americans, the first tag was more likely assigned to the main objects than that by Chinese; but for Chinese, the first tag was more likely assigned to the overall description or relations between objects in the images. The findings had significant implications for designing cultural-sensitive tools to facilitate the tagging and search process of digital media, as well as for developing data-mining tools that identify user profiles based on their tagging patterns and cultural origins.

Summary

Synopsis

In this paper, Dong and Fu discuss the results of a lab study that looks at how native "European Americans" (e.g. Americans of European descent) and recently immigrated Chinese use different types of tags for images. In general, they find that images of different types tend to tag different parts of images in different orders - European Americans are significantly more likely to initally tag portions of elements in images that are coded as the "main object" than are the newly immigrated Chinese.

Introduction

The authors point to the general propagation of media creation tools, as well as some past research that has gone into tagging. They primarily argue that past work has been into why and how people tag images, while their interest is particularly in what people tag. Their primary hypothesis is that cultural difference will play a significant role in just what people choose to tag.

  • ...
  • ...
  • ...

Method

The primary method of the experiment was to gather two culturally different groups, then have them both practice a procedure of tagging images; tags would then be checked against previously coded images.


Participant Demographics

All participants (Chinese and American) were recruited from a university community; most had college education and some had graduated from college. All of them had familiarity with searching the web for images and some had experience tagging images. Outside of that, their demographics were as follows.

Gender Age Months outside of native country
Male Female Mean SD
European Americans 9 12 21.71 3.33 15 Max
Chinese 8 15 22.38 2.35 9 Max, 3.17 Mean

Procedure

The authors assembled 60 different pictures based on five criteria; there were 20 images from each of three foreground objects listed below.

  • Photos of real-life objects
  • No language or cultural iconic content
  • The image contained at least one clear foreground main object and a number of distinguishable background objects
  • The main objects were human, animal, and still objects
  • For a good proportion of images, there were similarities in the main objects or background objects such that multiple tags would be needed to distinguish them.
Sample images with the three types of main objects

All participants were told to create a set of tags, with the following instructions (given in English to the European Americans, in Chinese to the Chinese):

The purpose to create these tags is to make it easier to search for a particular image
in the future. Please imagine that weeks or months later, when you come back to look
for a certain image, you can only use tags as your searching cues. The tags you have
created should be able to help you find the correct image faster.


Coding Procedure

...

Results

...

Conclusion and Discussion

...

Notes and complaints

  • Dong and Fu have a very balanced gender ratio, but don't report any information on how tagging styles differed across gender. THere may have been no significant effects, but it would certainly be worth confirming.
  • Ahn and Dabbish stress that they asked the players of the ESP game to make tags for the benefit of other individuals. Dong and Fu, in contrast, explicitly told test subjects to make tags for their own benefit. This makes comparisons with their work potentially dubious.