On the Radar: The Relationship Web

Why semantic web challenges may often be about relationships.

NewsTeam | AO [AlwaysOn] | POSTED: 10.30.04 @15:32

All of this human-centered content is published in a way that intersects right in the middle of the semantic web. A lot of these services are at a minimum generating meta data in XML, but a lot of them are using RDF. And while there isn't a lot of agreement on ontologies and the ontologies being used are not perfect, there's a huge amount of data out there. We're seeing people really interested in creating machine-readable information about themselves and their relationships, and FOAF is a way of representing who they are and who they know in RDF, and there's a huge explosion right now in the amount of FOAF data.

FOAF was started in 1998 or 1999 by Dan Brickley and Libby Miller from the W3C, and it was an experiment. A few years later, I remember seeing a neat website called They Rule, where they had generated RDF data that represented corporations and their boards of directors. And you could input to corporations and see how they were connected.

Today, people are generating, and major boinging services, too, millions and millions of files in RDF that describe how people relate. People are just starting to get the hang of this and build some interesting applications. It's enabling people to write applications that focus on people, their interests, and who they know. I think in the future we'll see a lot of really innovative business uses of this kind of technology. Relationship management is something that can certainly use this technology. And because RDF is great at providing you a mechanism for combining different kinds of data, if you have data about people and data about, say, records and albums and music, then you can ask with a simple query what albums would I like today that my friends are also listening to. So you can build these services very rapidly that are actually quite powerful.

In the future we'll see corporations adopt this and come up with some innovative new uses, and on the internet, you will actually be able to surf for people who are good lawyers or who are good accountants or who can walk your dog in your neighborhood. This sort of technology is happening now. There's a lot of data, and certainly probably in the next year or two you'll actually see internet-wide services that play to search on people.

Something worth mentioning, though, is privacy and trust. These are really the two challenges of using RDF to describe people's personal information. You don't want to give your good contacts to someone you don't know. You don't want to publish what you like to do on the weekend or what you did at Burning Man last year to just anyone who can find you on Google.

There needs to be systems where you can protect your personal information and own it, and this is a huge issue in existing social networks, because you're giving up valuable information about yourself to somebody else. They can essentially use it for a variety of other things. You need some way of selectively disclosing information about yourself. While I think that FOAF as a vocabulary in RDF is still at its very early stages, it's evolving very quickly, it's customizable, and a lot of the problems in this area are going to get solved soon.

Fromm: Radar Networks is building a lot of semantic web tools and basically a platform. Nova, do you want to talk about some of the important technologies, where they are, and where they're likely to go?

Nova Spivack: (Mr. Spivack is the CEO and president of Radar Networks.)  One general point that helps where to focus in this space is that one of the big shifts that's taking place is that we're moving from a focus on information to relationships. And you could say that anything you can express in one form you could express in the other: it's a philosophy. The semantic web is all about processing and forming relationships and doing that efficiently, finding ways to represent relationships between things, to define what those relationships really mean, to evaluate the reliability of those relationships, and all sorts of other technologies. The first generation of the web was the information web, and now we're starting to make the relationship web.

If you work with that type of data, you have to have tools for storing, or first for describing relationships, and then secondly for storing and retrieving them. You need to be able to do interesting queries on these big graphs of relationships. One of the first needs is a good way of just handling the data. You could use RDBMS and just convert these big graphs of relationships into tabular structure. That turns out to be a little bit of—because the relationships are semi-structured and are always changing—a database maintenance issue.

RDF allows you to just keep adding relationships without messing with the previous one, so it doesn't map precisely to the tabular structure. When you try to do this on top of a relational database, you quickly find that for various reasons, one is simply that you have to change this data representation, but also because storing and retrieving this data yields exponential numbers of queries as you start going through the graph, inexpensive relational databases don't scale well to huge graphs of data. If you want to buy the most expensive Oracle database, it will probably be okay for FOAF today. But in the future, even that may be a challenge because the relationship data structures that we see emerging are growing exponentially and they're very hard to manage.

First you need something called a triple store—a new type of database that stores triples, which are basically relationships between things. You need to store these and retrieve them and be able to inference along really large chains of triples very quickly. They have been developed in the research arena primarily, and there have even been some attempts at early commercial triple stores. I think you would probably agree that performance isn't really there yet.

That was one of the first problem sets that we had to solve in trying to build a semantic applications platform handling large amounts of data. We've actually built what we believe is the currently optimal triple store, and we'll be LPGLing that to Java. That should go a long way to enabling a new class of applications that can work with graph data.

Above that, however, you need a lot more in order to build semantic applications easily. You have to be able to create and store ontologies, and then you need to be able to make instances of these ontologies, which are essentially your database. The ontologies are schema, and the data is the various things that you're building on top of it. You need a way of doing that, converting that into triples, and doing queries on the system.

Above that, unless you're a Ph.D., you're probably going to need a number of layers for building applications and doing queries, so we've been focusing primarily on adding these higher levels of value at this point to make it easy for non-specialist, non-Ph.D. coders to build enterprise applications that make use of ontologies and graph data structures.

In terms of our business model, we will be providing a higher-end version of the platform for enterprises with a further range of tools around it, similar to the Java business model, and then some services around that.

This text is excerpted from SDForum's Semantic Technologies Seminar, cohosted by AlwaysOn, TopQuadrant, and Enterprise Architect. Part Two of three in Series Two of four.

+ HOME

Tell us what you think of this post using our On or Off rating system. Only your most recent vote will count.

WAY OFF								ON THE $
	[1]		[2]		[3]		[4]		[5]

Join the Discussion

NOTIFY?

Releationships are indeed at the heart of the semantic web. Simplistic relationships such as those implied in FOAF have correspondingly limited value (mostly for semantic search, browsing and personalization, but not for semantic integration and analytics). We can say that one thing is related to another, but cannot exactly say why or how. Relationships that organize information or are based on set memberships or relate to the value restrictions are similarly valuable, but to a limited level. Relationships that capture "implicit semantics" via clusering, vector space, etc also provide some help, but to a limited extent.
For enterprise applications it is more valuable to have named relatiohsips that are codified in an ontology, and have explicit semaitc metadata. E.g. a Person lives at an Location which has an Address, or a Persons Works for an Organization in a given Role/position, etc. (with corresponding typing and value information). And being able to represent and preferably automatically understand that Company A buys Company B, and Company A acquires Company B, are the same or very close relationships is very valuable. Also for semantic analytics (Semantic Web/Technology Applications that support decision making, analysis or discovery), chaining relationships is very important-- as in say discovering possible suspecious chain of (indirect, multi-step) relatiohships between two entities in an Anti-money Laundering application. This typically require analyzing RDF graphs with typed relationships and semantic metadata (with domain ontoloiges providing additoinal context and the speher of interpretation and agreements). Some more thoughts are explored in:
"Relationships at the Heart of Semantic Web" (Google for a paper on the topic) where the issues of Modeling, Discovering, and Exploiting Complex Semantic Relationships are investigated.

Amit Sheth, CTO, Semagix, Inc www.semagix.com

re: AI Nostalgia (gruffly paraphrasing Jerry Feldman & others of the era)

"... Instead of Artificial Intelligence, AI really stands for 'Always Impossible' ... If we knew what we were doing, then it wouldn't be AI. ... Perhaps a more positive view is 'Always Improving' ..."

My fav is 'Always Intuitive' but that's an entirely different story ...

So, the immediate biz focus is on tangible "tool building" rather than biz model assumption fundamentals that provide the critical difference between incremental improvement vs. >10x value boost thru Tek innovation ('Semantic Web' in this case).

Perhaps this a pattern we've seen before?

:-)