February 15, 2006
Recipe aggegators when?
Marc Canter predicts that recipes will soon step to the forefront of the world of microcontent. I know someone who might be interested in this. I wonder: how long before the beta launch of Foodr / scrum.ptio.us??
Web 3.0
I've rather pompously made some predictions on what Web 3.0 may constitute:
• Personal metadata ownership and management: initiatives such as AttentionTrust, Structured Blogging and Microformats are building infrastructure which supports the notion that data is the property of the user, not the platform owner. Data can be moved from one service or device to another at will, it can be exchanged for something of value, but the user has the right to know who is using it and how. Structured blogging enables user-generated microcontent to be stored in its appropriate schema and syndicated as an identifiable content-type to any third party that requires it. This new paradigm provides far greater user control, and enables a new class of web app; Edgeio, for example, aggregates specific microformats (in this case classified adverts) and builds a service around them; KritX does the same for reviews;
• Implicit data capture: architectures of participation are worthwhile in theory, but only a small percentage of users will actually add value through content submissions. The goal is to architect a system where participation happens by default. Applications such as OnLife, Octave, Last.fm and Root Markets have begun to address this problem by ensuring that metadata capture is invisible and requires no real effort on the user's part;
• Self-ethnography: some discussion is now taking place abound ‘datablogging’, ‘personal data mining’ and ‘meta-experiences.’ Once systems are capable of implicit data capture, what kinds of services will emerge for mining and presenting the resultant data? The value of these services, according to Ed Batista, is “the experience of comprehending and assessing that information is derived from a previous (or concurrent) experience”;
• Personalisation and attenuation: most mashups are currently ‘dumb’. The problem with such services is their inability to filter and personalise the huge quantities of available information. Attenuation (i.e. the reduction of ‘information noise’) is of such importance that it occupies the thematic core of O’Reilly’s 2006 Emerging Technology conference. The emergence of open-source inference engines will enable developers to build intelligence into web apps, and capitalise on the nascent ‘attention economy’.
August 15, 2005
Last.fm vs Pandora
Ian's been roadtesting Last.fm and Pandora, and rates both of them pretty highly. His head-to-head comparison of Cocteau Twins-based recommendations was particularly interesting. The main reason I prefer Last.fm / Audioscrobbler is a philosophical one - Last.fm uses the listening habits of other people to feed me recommendations; Pandora uses a system based on categorisations by expert musicologists. Music is a social activity, so I prefer my recommendations to come from a social network. The same argument applies to folksonomies (and taxonomic categorisation in general). Assuming the quality of recommendations from both are similar, the philosophical question remains: is everyone equally capable of making sense of the world, or are some better qualified to do it than others?
November 11, 2004
eBay Pulse
Some simple data mining by eBay has yielded this "daily snapshot of the World's Online Marketplace." Compelling stuff for trendwatchers and curious punters, it aggregates most popular searches, highest-priced items and most watched items onto one page. But what does it mean when a "Native Limestone Timken Kansas School Warehouse Storage" is the most watched item...?
July 27, 2004
All the tags
This is a fascinating page on Flickr (takes a while to load): an aggregate of every tag appended to every photo. Surely a marketeer's wet dream. No semantic framework though (likely to be an ongoing problem with user-generated metadata.) What Flickr needs is picture-recognition technology. I saw a demo once where the application initially required training; users must tag images with the names of subjects. Thereafter, the application began to ‘learn’ the names of subjects contained within new photos with an 80% accuracy level. I saw another demo that partly inferred from a camera phone's location data what the contents of the photo might be. Once this data has been obtained, it can then be fed into controlled vocabularies, making it much more valuable.
July 14, 2004
RDF bad...
I met with a certain London university today and the topic of conversation steered ultimately towards RDF, a metadata standard for which they have little time. Why? Because the so-called 'controlled vocabularies' of RDF are often uncontrolled, flawed, messy and poorly-structured. They asked the questions: who is building them, what is their agenda, and can they be trusted? It recalled Cory Doctorow's irreverant but still-relevant metacrap rant from 2001. It's probably true that schemas will be built by specific communities to cater for their local needs. And it would be bad if this results in duplication and increased heterogeneity. But to counter that charge, 'schema orchestration' is coming along nicely, and automatic harvesting from existing documents will remove the element of doubt from current human-generated metadata.
July 13, 2004
www.tomspetition.org
When an email concerning www.tomspetition.org fell into my inbox I initially thought it was a piece of sick spam. Not so. It is in fact a petition to renew the Assault Weapons ban, set up by Tom Mauser, whose 15-year old son was killed in the Columbine High School massacre. To support the cause, the site includes a cool graphical map of the US, illustrating the location and degrees of separation between signatories. When you sign the petition, you can see your position in the email trail from origin. This is an unlikely illustration of the power of metadata. From basic RDF graphs to GIS to Musicompass, structured metadata can make a lot more sense when it's represented graphically, particularly if orchestration between vocabularies is taking place. I'd like to see more more topographies and ontologies represented graphically; doing it really well so that the outputs are comprehensible will be the challenge I guess.