Tag Clouds: A Response
Note: The following is a response to Tag Clouds II. Please be sure to read Mr. Zeldman's full post so as to better understand where I'm coming from in the text below.
Tags are like links. Shown in “raw popularity mode” they simply point to the content that's most popular. I fully agree with Mr. Zeldman that if you're running a many-authored site, tags are in fact a fantastic way to cut straight to content that's the most popular.
Where I think Mr. Zeldman's discomfort with tag clouds begins is the fact that “popularity” is about as far as tags (and also link-trackers) have come in terms of their sophistication.
When tag clouds are sliced yet again — by time, another tag or two, a keyword, an author — they start showing their power. Technorati and Blogdex know this about links, and slice their data according to time (typically one day). Del.ico.us, which serves a similar purpose to Blogdex and Technorati but works on a totally different principle, uses time to slice the information presented on their main page as well.
This use of tag clouds shows that they are just in their infancy. There's a lot more to them we haven't seen publicly “done” yet. We've really only just begun “tagging” things. (To “tag” something having a slightly more open ended meaning than simple “categorization.”) The best display method we've got at this point is simply showing “what's popular” in any given raw data cloud, much as do the afore mentioned link-tracking sites.
I agree with Mr. Zeldman that these “popularity meters” are not particularly useful except as a sort of general barometer of the large-scale network effects taking place in the blogosphere, and certainly that they do bury unique content over time.
But what about taking them to the next level? What about applying some network principles to them? What about “smart tags?” Tags that know “people who used this tag for this photo/link/post also used this other tag.” Or that “these three tags” tend to be used together when categorizing something.
Cutting the data in this way can turn a simple popularity cloud into a fantastic data mining technique. The relationships between the tags are what's important, not so much the tags themselves — in this way they're just a means to a greater end. If, of course, someone's willing to take them to the next level.
Update: I haven't worked it all out yet, but this chapter from Cluetrain fits in here somewhere, I'm sure of it.
