The first talk I was able to attend at Hack Day was by Flickr’s Aaron Straup Cope and Dan Catt, about Machine Tags. I’m really interested in this because it adds another layer of metadata to tags, allowing them to be read by machines. I’ve heard them described as triples, and in a way I suppose that’s true, but these are not like RDF triples. Basically, a machine tag consists of a namepace, a predicate, and a value organized in a certain syntax. It’s pretty simple, but should allow services to make use of the additional data pretty easily. I scribbled a note on my paper that says:
- folksonomy :: taxonomy
- machine tags :: RDF
That’s a simplification, of course, but it seemed to be a good way to describe the relationship. The two main issues that will affect the adaption of machine tags are:
- What can you do with them? I think the answer to this one is pretty wide open. You can make apps that will use machine tags to express relationships between content, people, etc, and trigger all kinds of behaviors. Flickr’s API lets you query machine tags, and basically what you do with it is just limited by your imagination.
- Where is the data coming from? The question is, though, will anyone aside from you, be adding the kind of machine tags that will make your application work? This is really two questions.
- What’s going to make me go back into nearly 1500 photos and add more tags to them? Something needs to be done to make this a little easier or people will never do it. I’m a pretty dedicated information geek. I’ve spent an hour disambiguating two names on Wikipedia. But I’m already dragging my feet adding my backlog of Flickr photos to the map, I can’t see sift through all of them again.
- Even if I do, what’s to say my machine tags will be compatible with your application? Do we need some kind of standards? Or are we expecting people to add new machine tags to their content for each application they want to contribute to?
Clearly, what’s needed is something that will assist users by automatically generating suggested machine tags that they can then revise, approve, or decline. Interesting things to think about at Hack Day…
One of the big winners of the day was a hack that used machine tags – Flickr Tunes by Steffan Jones. Basically, it was a Mac OX widget that used the BBC Muiscbrain database (I think) and the Flickr API to match machine tagged photos with a song. So, if a person took a photo that they felt illustrated a particular song, and they used the appropriate machine tags to capture the song name (and even a time code), then the images would display while the song played, as a sort of slide show, even keying to the specific moment in the song, if indicated.
Pretty cool, but as I mentioned, how much data would need to be entered to make it a valuable experience for fans of all different kinds of music?
5 thoughts on “Hack Day: Machine Tags”
Good writeup. Just a small correction, though; MusicBrainz isn’t a BBC project.
Opps, thanks for the info and the link. There were a lot of interesting APIs. services, and datasets flying around, and the details of that one were a little hazy to me. I thought that hack had something to do with BBC music, but maybe there was another aspect of it that I missed.
Ah-ha – the connection seems to be that the BBC Music site is also using the MusicBrainz IDs. So, I’m guessing Steffan used those IDs in his machine tags to access the songs availble on BBC Music.
Yes, I just noticed that myself and came back to correct the earlier comment. I’ll have to have a proper look later. I also reckon it’s worth emailling Steffan to get him to put his widget somewhere.
Pingback: Hack Day London « Meaningful Data