远见与创新: Collaborative Tagging

The Structure of Collaborative Tagging Systems
Scott A. Golder and Bernardo A. Huberman, HP

Collaboration is an interesting topic. BT is an excellent example. Now del.icio.us is another.

Tagging & Taxonomy is two different concepts. The difference is: Tagging is non-hierarchical and inclusive, so flexible, easy.

Flexibility brings possible challenges because of the characteristic of language:

Polysemy: a word can have many senses. e.g. "window" can be a hole in wall, or the pane of glass
Synonymy: multiple words share one meaning, e.g. television or tv
Basic level variation: the words descripting an item can vary from general to very specific, e.g. cat, animal

User activity is interesting, we can track it to find his/her interesting and find friends.

Section 4.2 shows, the tags for a URL can reach a random stable proportion after about 100 bookmarks. In figure 7a,b. the number of finally stable tags for a URL is about 10. the biggest tag proportion is 20% and 40%. The stability is reached at 50 bookmarks and maintains until 400 bookmarks. This just shows, a URL's tags don't have dramatic changes. 50 users' tagging can reach an agreement that what is this URL.

This pattern is explained by Eggenberger and Polya model. The remarkable property of this model is, the RANDOM stable point. From this analog, just like the read and black are same attributes, the 10 tags also indicate the property of this URL at same extent, although their final stable proportion are different value. We should treat these 10 tags as same importance.

The authors' analysis about imitation and shared knowledge is correct.

There are also two trivial problems in this paper:

The authors think user use del.icio.us primarily for their own benefit, but constitute a useful public good. In fact, many users publish their bookmarks just like their publish their blog. We all know most of Blog is written for others to read, rather than the private diary. Just like Blog is a place to show personal interest, idea, finding, and stimulate the communication among people, del.icio.us is also such a system. The name of "collaborative tagging system" has uncover this point, i.e. collaboration is the key. At some extent, it is more like Digg. In Digg, users votes for the best interesting topic.
The authors think figure 4a,b shows a retrospective sensemaking process. But in the figure, when tage 3 increases continuously, the tag2 doesn't drop. So tag3 means a new tag that user want to maintain, rather than he want to refine the tag2.

[Thought]

Chinese language may be good at Tagging. Each chinese word has its context. Is there any research about it?
Need have a look of the Gaming theory for cooperation gaming. It is interesting!

[Future Reading]