I've been playing with Google Base and also tracking Amazon because I've been trying to find a Book equivalent of last.fm, settled on librarything.com and was looking at the Amazon API, lists and recommendation system. What's interesting in all this is categorization systems for general SKU#s This also got raised in the Microformats group. The underlying problem is that a job, an event, a book, an iPod, a piece of music, and so on, all have different sets of attributes.

So back to Google Base. They've got their own set of top categories. Each category has a set of attributes that can be provided as part of an RSS or Atom upload using namespace extensions (Why isn't Dave Winer all over this? Somebody should tell him.). You can then also provide arbitrary tags that you can use in Google Base searches as pivots. These are kind of like Flickr/del.icio.us tags but the user interface is very different and there's no community feedback at all.

Now, inside Ecademy (and like Tribes) we've got our own copy of Craigslist. We use user-entered tags (unlike Tribes and Craigslist) to navigate these and to avoid me having to constantly maintain a category hierarchy. I'm working on uploading these automatically into Google Base. But if I'm going to do this really effectively I have to duplicate Google's top level categories and category attributes.

And so I finally get to the point. Google, Amazon and eBay with their APIs and uploads are forcing a kind of category imperialism on the rest of us. As much smaller developers, we have to match their category-attribute schemes in our systems or maintain lookup tables of the differences. And I really can't decide if this is a bad thing or a good thing. We do actually need standards for how to describe common SKUs. And the Microformat people could do worse than create an hGoogleJob format to match Google's idea of job. But then you get into the whole issue that Google's descriptions aren't always very good. For instance, they currently only support USA addresses and telephone numbers (surprise!). And Google are pretty much impossible to have a conversation with to try and improve them. There's no standards process here at all. We get to use whatever Google decides with no reference to us.

Much to ponder here.

Separately, Danny Ayers (I think) made a comment (which of course I can no longer find) about OPML, XOXO and RDF. He pointed out that a hierarchy is just a special case of a mesh. And that all real world problems are meshes. I'd take that further, we as humans have a hard time understanding and groking meshes. It's as though they're multi-dimensional when our brains want to work in 3D and preferably in 2D. And we work best manipulating symbols on 2D paper. Which is partly why we insist on reducing complex mesh problems into 2D Hierarchies. And it's why I consider Outliners to be harmful. Somewhere in there is a big thesis on why we seem to like hierarchical command structures that support an Alpha Male. But I digress! One of the things that makes tag based folksonomies interesting is that they represent a way of navigating and understanding multi-dimensional meshes in a 2D user interface.


[ << 2005 Notebook Drive Roundup ] [ Amazon.co.uk: XCP: Search Results Music >> ]
[ 30-Nov-05 6:45pm ] [ , ]