This paper was presented at CSCW in Banff, Canada, Nov 4-8, 2006.

Authors:

  • Shilad Sen
  • Shyong K. (Tony) Lam
  • Al Mamunur Rashid
  • Dan Cosley
  • Dan Frankowski
  • Jeremy Osterhouse
  • F. Maxwell Harper
  • John Riedl
  • all from University of Minnesota

Tags can be characterized as “selfish work that benefits the community.” So they are a good thing and we should understand what gets people to contribute them.

Types of tags:

  • factual: e.g. del.icio.us
  • subjective: e.g. amazon.com (”good” “great”)
  • personal: personally meaningful, perhaps date related

Today, most of the tags are factual in MovieLens, a movie recommendation site and their research platform.

Experiment

Created 4 tagging communities to determine what sharing conditions influenced how tags were created. The four conditions:

  1. no shared tags
  2. randomly chose some tags to share
  3. shared the most popular tags to share
  4. inference algorithm that found similar movies and found most popular tags across all of these movies, shared those

These 4 conditions generated 4 different types of tagging practices:

  1. tags were split evenly between personal, subjective, and factual
  2. favored subjective tags with a fair share of factual too
  3. favored factual tags
  4. favored factual tags

2,3,4 had very few personal tags

Conclusions:

  • people only liked the subjective tags they agree with
  • people like factual tags a lot
  • personal tags of others are not useful at all, but personally, they are very useful

Their data is highly influenced by a few number of people contributing a high proportion of the tags, but Shilad didn’t think this skew influenced the results.