This document introduces the RDFWeb system, providing an overview of the technology, ambitions and architecture behind the project. While RDFWeb relies heavily on the W3C specifications for the Resource Description Framework (RDF), this document stands alone as a complete description of the system. A brief introduction of the RDF information model is provided for the curious.
This document is incomplete. RDFWeb is incomplete. The quick start guide hasn't been written yet. The vocabulary is in flux. Everything might change. We do however have two running RDFWeb implementations, and are now focussing on getting an initial stable RDF vocabulary defined so we can scale up the effort into a large distributed testbed. The vocab and quick start documentation are the most urgently needed. The current General Overview document should provide a reasonable pespective on what we're doing here, although it is patchy in several places.
Note: Two years later, I still haven't done much to finish this document off. For further documentation, see the FOAF namespace document or Edd Dumbill's IBM developerworks article.
see also: RDFWeb site, the co-depiction experiment
The basic idea behind RDFWeb is simple: the Web is all about making connections between things. RDFWeb provides some basic machinery to help us tell the Web about the connections between the things that matter to us.
Thousands of people already do this on the Web by describing themselves and their lives on their home page. Using RDFWeb, you can help machines understand your home page, and through doing so, learn about the relationships that connect people, places and things described on the Web. RDFWeb uses W3C's RDF technology to integrate information from your home page with that of your friends, and the friends of your friends, and their friends...
RDFWeb is best explained with an example. Consider a Web of inter-related home pages, each describing things of interest to a group of friends. Each new home page that appears on the Web tells the world something new, providing factoids and gossip that make the Web a mine of disconnected snippets of information. RDFWeb provides a way to make sense of all this.
Here's an example, a fragment from the mostly-fictional RDFWeb database. First we list some facts, then describe how the RDFWeb system makes it possible to explore the Web learning such things.
Dan lives in Zetland road, Bristol, UK with
Libby and Craig. Dan's email address is email@example.com.
Libby's email address is firstname.lastname@example.org. Craig's is
email@example.com. Dan and Libby work for an organisation
called "ILRT" whose website is at http://ilrt.org/. Craig
works for "Netgates", an organisation whose website is at
http://www.netgates.co.uk/. Craig's wife Liz lives in Bristol
with Kathleen. Kathleen and Liz also work at "Netgates".
Martin knows Craig, Dan and Libby quite well. Martin lives in
Bristol and has an email address of
Historical aside: this was true in summer of 2000, much of this is no longer true. I've left the example intact though.
This kind of information is the sort of thing typically found on Web home pages. The extract shown here indicates how short, stylised factual sentences can be used to characterise a Web of relationships between people, places, organisations and documents. In real life, this information would be most likely be distributed across variou s Web pages created by the individuals listed. Very likely, their pages will link directly or indirectly to the home pages of countless other friends-of-friends-of-friends.
We want a better way of keeping track of the scattered fragments of data currently represented in the Web.
We want to be able to find documents in the Web based on their properties and inter-relationships; we want to be able to find information about people based on their publications, employment details, group membership and declared interests. We want to be able to share annotations, ratings, bookmarks and arbitrary useful data fragments using some common infrastructure. We want a Web search system that's more like a database and less like a lucky dip. We need it to be be distributed, decentralised, and content-neutral. RDFWeb, if successful, should help the Web do the sorts of things that are currently the proprietary offering of centralised services.
RDF seems to offer a lot of promise in this area. While RDF is defined in terms of a rather abstract information model, our needs are rather practical. We want to be able to ask the Web sensible questions and common kinds of thing (documents, organisations, people) and get back sensible results.
All this sounds a bit ambitious (and it is), but we think we've a reasonable sense of how to build a linked information system with these capabilities.
This section can be skipped by readers who are not interested in the technological underpinnings of RDFWeb.
On the Web, one technology we can use to describe various kinds of things and their inter-relationships is called 'RDF', the Resource Description Framework. RDF is a relatively new specification produced by the World Wide Web Consortium (W3C). Like previous W3C specifications for HTML (Web documents), RDF is designed for widespread, decentralised use.
At the heart of the RDF is an information model based around the idea of simple 3-part sentences such as "dan hometown-name bristol", or "martin homepage-url http://www.glandscape.com/". It turns out that this somewhat stilted way of describing things can be very expressive, since RDF allows anyone to define new terms such as 'homepage', 'hometown', 'bestFriendHomepage' and so on. The RDFWeb starter vocabulary provides some basic terms such as these to provide a common language for machine-readable homepages. RDF allows multiple such languages to be mixed together - for example, you can use the "Dublin Core" vocabulary to describe documents you've written or contributed to.
The RDFWeb system itself doesn't care much about what you say, but about the way in which you say it. By using a carefully designed file format and a simple common information model, we can build a distributed database in which anyone can say anyth ing about anything... By squashing our information into the simple 3-part-sentence framework, we can build systems that automatically combines such data from multiple sources.
One page might tell use that "firstname.lastname@example.org works-at http://ilrt.org/". Another might tell use that "http://ilrt.org/ based-in bristol". On this basis, RDF-aware tools such as the RDFWeb system could conclude that the person whose email address is email@example.com works for an organisation based in Bristol. Other pages might tell us more about that individual, or about the city of Bristol , or describe other things related to Bristol or firstname.lastname@example.org
By adopting a common style of representing this kind of data to Web-based computer systems, we can start to treat the Web as a distributed database containing arbitrary facts (as well as lies and half-truths...). RDFWeb was created to explore this view of the Web, and to show how practical tools can be built around these sometimes abstract ideas. There are a number of subtle technical points surrounding the practicalities of identifying people, organisations etc. in a standard way on the Web. These are discussed in more detail when we introduce the RDFWeb starter vocabulary.
The starter vocabulary for RDFWeb combines some basic RDF vocabulary we invented for the project, plus other useful existing vocabulary such as the Dublin Core metadata elements.
The Friend of a Friend (FOAF) vocabulary provides the basics. This XML namespace defines RDF properties useful for RDFWeb applications. Currently these are inadequately documented, and there is already some variation and ambiguity in the few testbed data files we already have.
Q: How do I get started?
A: (It Should Be) Easy. Except that we haven't created the appropriate HOWTO documents explaining how to contribute. Our bad. Asking in the RDF Interest Group IRC channel might shame us into finishing these.
Q: What happens then?
A: Various RDFWeb tools watch the Web for new files, and combine your data into a database along with many other similar files. These are available for searching...
Q: Um... how's it work then?
A: Everything is represented as fragments of a huge data graph, with Web identifiers for relationships between things (eg. wife, livesWith etc), and Web identifiers for anything we can readily identify, such as homepages, mailboxes. This provides an organising principle for blending together data from different sources. We can't always tell when two people are talking about the same thing, but we can try.
How can we represent people to the RDFWeb? Personal mailboxes
are a good example. They can be used as a way of picking out
some individual for unambiguous description: for any
email address, there's only ever one proper flesh and blood
owner (theoretically). So RDFWeb can use this kind of fact to
Here's a graphical
representation of an RDF description for the person who
has a personal mailbox
mailto:email@example.com. Note that
we don't have stable IDs for the people, so RDFWeb software
will have to do a bit of extra work to integrate and
normalise this data when it merges it with factoids from the
RDF home pages of the other people mentioned in the graph. We
can achieve this through knowing that for any specific value
of some uniquely identifying property, there can be at
most one Web-identifiable resource with that property
value. This trick works for people and email addresses,
organisations and their home pages, restaurants and their
telephone numbers and so on.
This document was written by dan brickley and libby miller with lots of ideas and suggestions from members of the rdfweb-dev mailing list.
(FAQ stuff here...)
Privacy/exhibitionism. Who owns what? (idea, database, software). Nobody owns the dataset, tools should be opensource, idea is trivial. This kind of application is pretty obvious once you buy into RDF. Database: each RDF page should be freely distributable, though legal nightmare with storage of personal data -- need to investigate. IPR/copyright over each page? (fair use statement to stop spammers harvesting the entire thing?). Basic vocab (see notes). Roadmap (event log / lifestream; dig sigs). recommendations. sw as scaffolding. How to say 'i like this restaurant...' etc. Motivating inference stuff: can we find real world uses for transitive relationships, symmetric etc. Eg. family tree stuff.
last updated: $Id: index.html,v 1.3 2002/02/09 22:01:05 danbri Exp $