The Wonderful World of Metadata

It seems like a good majority of literature on Folksonomies and Taxonomies divides the “issue” into two camps. These two camps then proceed to lob grenades at each other. The folksonomy camp accuses the taxonomists of creating highly subjective and arbitrary semantic relationships that further cultural bias, force users into sometimes awkward or arcane jargon, and remain rigid while information changes. The taxonomic camp accuses the folksonomists of creating “sloppy” systems that are imprecise, rife with ambiguities, and poor for returning all relevant records. In regard to the charges leveled in both directions, I have the following comments:

• The moniker of “sloppiness,” quickly thrown at folksonomies, is grandly overstated. The research done in the article “Folksonomies: Tidying up Tags” clearly shows that some tags gain critical mass and can be identified as culturally agreed upon. One-time tags, be they junk, whitenoise, or just simply unpopular tags, get weeded out in organizational visualizations like tags clouds. They fade to the back. Core terms become recognized as such by the will of the group.

• More tags actually improve accuracy! The bigger the herd, the more apparent commonalities will be with a larger number of tags. Also, with more contributors, there will be a greater diversity of less-popular tags. This is not actually a bad thing. With more access points, users can think about an indexed item in more ways and still reach the same item. Clay Shirkey makes a good point in saying:

Tagging gets better with scale. With a multiplicity of points of view the question isn’t “Is everyone tagging any given link ‘correctly’”, but rather “Is anyone tagging it the way I do?” As long as at least one other person tags something the way you would, you’ll find it…

Not all one-time tags are junk. Some might actually be useful for someone else looking for the same thing.

• Tags simply cannot replicate the semantic relationships in a taxonomy. There is a world of value in BT, NT, and RT relationships. Although the value gained also introduces a healthy helping of subjective bias, these relations are quite useful and cannot be replicated with tagging alone. This is a huge reason why taxonomic structure (be it pre-set hierarchy or on-the-fly faceting) remains viable and important.

• As alluded to above, tags provide a democratic way of indexing. There is no inherent “power” doing the indexing. The power is us and reflects our naming. That said, there is a still a cultural bias within the group (a picture of snow will most likely be tagged “snow” in U.S.-based Flick’r rather than one of the 12 Inuit words for snow); however, the open format allows for group bias to be easily rectified by users (there is nothing saying that Inuit word tags can’t be added to the photo of snow in Flick’r).

So, the thoughts above bring me the following benefits outlined for each method of indexing:

• Semantically, a taxonomy is much richer
• Tagging is structurally more flexible
• Tagging creates more objective tags
• Tagging allows for many more search-access points
• Taxonomies can (conceivable) guarantee that all relevant records will be returned

So, why not have both?

Right now, I can think of only a handful of instances where both types of indexing are employed at the same time. Amazon does it, although the tagging side of their site isn’t featured very well and hasn’t attracted that many taggers. I believe (although I’m not sure) that the University of Pennsylvania’s PennTags system allow users, once they are logged in, to view communal tags in the catalog as well as normal LC subject headings. Perhaps the brute-force computing necessary to create a system that can do both simultaneously makes the option unviable. Also it might seem confusing to pair the two, especially if the interface is poorly created.

There is, however, a good case to be made from pairing taxonomic and folksonomic structures, especially in image and video indexing. As Christine Jacobs points out in her article “If a picture is worth a thousand words, then…” images and video are akin to the novel in that “aboutness” is incredibly harder to define and far far more arbitrary. Three different indexers might come up with three different ideas of “aboutness” for a photograph (or, in that case, for a novel….what is Moby Dick really about?). In this case, an image indexing database might be well served to try to reap the objectivity of tagging while, at the same time, implement the control of a taxonomy.

Slawsky’s article “Building a Keyword Library for Description of Visual Assets: Thesaurus Basics” describes the how a DAM system can create a controlled vocabulary that is rich in semantic relationships and captures all records within a database. Great! This could be a terrific starting point. Unfortunately, this DAM system’s semantic relationships and assigned vocabularies are arbitrary and rigid. (Many of Slawsky’s examples for “BT” relationships left me scratching my head. They certainly were not intuitive for me.)

Here is where tagging can help. If a searcher is not having luck in the controlled vocabulary, why not offer the option of searching and contributing tags? Searches could be side-by-side or integrated into the same interface. Providing a second way to search could benefit the imagined cultural outsider, the neophyte, or simply the person who is thinking about a topic differently. For a person that likes following semantic relationships, is looking to expand or narrow their research, or is familiar with the system’s terminology - then the traditional taxonomy still offers that structure.

If Amazon can pull this off for everything from novels to diapers, there has to be a way to customize and improve it for large image collections. I don’t know of any image indexing system as of right now that offers both the taxonomic and folksonomic approach, but maybe it does already exist. Does anyone know? If not, it needs to be created. It seems like the two approaches complement each other well and would benefit from being implemented in conjunction - especially in a image indexing environment.