Semantic Web services

Bulletin of the American Society for Information Science & Technology, Apr/May 2003 by Parsia, Bijan

The World Wide Web allows people to follow bewildering paths through hundreds of documents on a variety of topics without losing equilibrium. People shop, plan and book travel, check accounts and pay bills, publish articles and artwork, share photos with family and friends and total strangers, build complex information systems and manage a variety of business interactions. Web programmers have done very well in building programs that help people write Web pages, build and maintain Web sites and develop sophisticated Web stores. It is much more challenging to develop programs that can use the Web with more human-like flexibility without a human being constantly in the loop. The Semantic Web and Web Services are two visions of how to make the Web more amenable to automated use.

The Semantic Web

"The Semantic Web is the web of connections between different forms of data that allow a machine to do something it wasn't able to do directly." (Weaving the Web, p. 185)

Programs can do a lot with the current Web, much of it critical to successful and pleasant human interaction with it. Web crawlers and spiders make link checking and site archiving easy and are absolutely essential for search engines. What makes Google the search engine for the Web is its scope (achieved by crawling the Web), the "goodness" of its results (achieved, in part, by automated reasoning about the links between pages) and the fact that it doesn't require massive, disciplined human intervention by the search engine operators or by Web authors. The last point needs a bit of explication. Google does require a staff of brilliant people who constantly enhance and tune the crawling, indexing, ranking, storage and retrieval programs. This is a significant amount of human effort, but it is dwarfed by the alternative: a team of catalogers to categorize three billion Web pages by hand. Google also doesn't require that page authors supply correct metadata for their pages above what they do naturally in writing link-rich hypertext pages. There's no need for an explicit cataloging step, either by page authors or by Google operators. Moreover, it's not clear that adding that step would produce better results (or even results as good).

A striking feature of this sort of automation is that it depends on the interconnectedness of the Web - its link structure. Web links are what make the Web work for human browsers, authors and, it turns out, for some content-sensitive programs. This suggests that a richer link structure could support not just better search, but other activities. Hyperlinking Web pages together for the convenience of human browsers turns out to be useful for building search engines that cater to the needs of those human browsers. What can we build if we have different kinds of links supporting activities besides human browsing? The best way to find out what we can do is to do it. The Semantic Web is driven by a specific vision: to explore what sorts of links between which kinds of representations supply the greatest achievable potential for the Web.

The Semantic Web is rooted in Tim BernersLee's original conception of the Web. Web links were seen not just as providing a navigatory connection for the reader, but also as (partially) constituting the meaning of the linked representations. On this view, the Web is a kind of semantic net with Web pages as nodes and hyperlinks as arcs. The machine processable meanings of the nodes are constituted by the patterns of arcs between them. The original Web was a hypertext system instead of a pure semantic net. The nodes were bits of text with quite a bit of meaning all on their own - meaning that is largely inaccessible to current programs, though quite handy for literate people. So, the extra meaning of the Web based on knowledge representation (KR) tends to take a back seat to that based on natural language. Web links only give us a little semantics, but it turns out that a little semantics goes a long way.

Still, it would be nice to have a little more semantics. If the original Web is a hypermedia system with aspirations toward KR, the Semantic Web seeks to be a KR system deeply integrated with global, distributed hypermedia. More precisely, the Semantic Web is a Web-like - global, distributed, universal, loosely coupled - KR extension of the human/hypermedia Web. There are no content-directed constraints on putting up a Web page or with making a Web link to anything else on the Web. These design desiderata are commonly expressed with the slogan, "Anyone can say anything about anything." It's not enough, therefore, to have a KR system with the same scope of subject matter as the Web (that is, any topic you care to write about), but that system must also accept input from practically anyone. This vision is as much a departure from traditional KR as the Web is from traditional hypertext. The Semantic Web requires a blending of KR and Web architecture. The difference between the current Web and the Semantic Web is that the Semantic Web asks people to say their anythings about anything in a way that's more amenable to significant, content-sensitive machine processing.