Metadata for the Masses

Over at Adaptive Path, Peter Merholz is talking about bottom-up classification systems like the tags used to organise data on del.icio.us and Flickr. If you're not familiar with these sites - they deal with internet bookmarks and photos, respectively - one of the main features is the ability to add multiple tags (like keywords), to your data, so that you can find things easily. The tags are entered as free text, so there's very little effort involved in adding them, and they aren't intended to be complete or unique.

In his article, Merholz brings the term ethnoclassification to our attention - defined as "how people classify and categorise the world around them" - and compares the use of free tagging systems to the landscape designer's use of "desire lines" to place paving (see On The Beaten Path for a good look at emergent paths) .

He also speculates about where these bottom-up classifications are headed next,

Use the tags to understand how people consider the content at hand. Then you can “pave” the best paths to ensure findability — say, by explicitly linking “nyc,” “newyork,” and “newyorkcity.” You can also align these tags with more formal schemes, thus enhancing the utility of both.

This raises some interesting issues, not least of which is the fact that Joshua Schachter over at del.icio.us and Stuart Butterfield over at Flickr seem hostile towards anything which might be seen as an attempt to standardise tagging systems. Merholz isn't suggesting standardisation here, but it's easy to get onto a slippery slope. Once we realise "nyc", "newyork", and "newyorkcity" are similar then the temptation is to merge them, but for all we know, the distinction may be important to some users. The solution is to offer browsing of multiple tags as if they were one (a union of tags) as an optional view of the data.

This is why the emergent paths comparison is a good one, especially in the case of del.icio.us where it's easy to see how similar tags could be suggested through usage, because different people will be adding and tagging the same URLs. In the case of Flickr though, tag consensus will be harder to reach unless tagging is opened up to everyone, perhaps to tag their collections of favourites. That way, when people search for a particular tag, Flickr could use the favourites tags to offer related tag suggestions. Because it is an optional query refinement rather than a unification of terms, it then becomes an interface issue and not a complex and unwanted database normalisation task. Over at del.icio.us, Joshua is already experimenting with user/tag similarity suggestions, hopefully Flickr will soon.