cygri’s notes on web data

Ginseng is a natural language query facility for RDF. The inspiration is “fridge poetry for the Semantic Web.” (a great tagline!)

“The markup prisoner’s dilemma” – why should I do markup if there are no tools making use of it, and why should I build tools when there is no markup?

“People do not understand formal logic. Full stop.” And they don’t even understand simple boolean logic.

So how should users do queries? Possible answers are query by example, graphical query languages, or full natural language processing. In practice, none of these really work for the semantic web.

Ginseng lets users query an RDF dataset with a limited form of natural language. They demo it with a geographical dataset. The text input field has autocompletion, so when you start with “W”, it will offer “What”, ”Where” and so on. If you type “Where is the m” it will offer “mississipi” and a bunch of other locations.

The dataset must have good terms or it won’t work well. Ginseng defines an RDF property for adding synonyms. So if your term is geo:isMountainOf, you could add a synonym “lies in” to the dataset.

Ginseng has a simple question grammar. It is combined with terms from the dataset into a full grammar that specifies all possible questions.

Queries are translated into a bunch of SPARQL queries that are executed against the dataset.

They did an informal user evaluation. People did reasonably well, but they didn’t like it because it restricts what they are allowed to type.

They intend to put it out as open source “after we clean it up.”

This is cool because you could potentailly point it at any SPARQL endpoint anywhere on the web and it would just work. (Well, somewhat, because most datasets don’t have good terms or synonyms.)

I’m at the Jena User Conference in Bristol. For me, this is a return to the place where I lived and worked for half a year in 2005, and I’m catching up with a lot of folks I haven’t seen for a while.

The organizers handed out USB drives containing the conference proceedings and the latest Jena release. Well, that was the plan. Turns out the drives – delivered by the vendor at the last minute – contain proceedings for another conference. So we have a bunch of papers about business management now.

Cool, that was about time: maps.google.de, and a bunch of other European countries got street data too.

This is where I live.

Weird: The new data doesn’t show up in applications using the Google Maps API. In my FOAF geolocation thingy, most of Europe is still a big blank territory. (Update, it works now, thanks to Alex for the hint!)

Funny: Google Maps can’t find Europe, Africa, or Asia. It does find America (but thinks it’s only North America). It does find Australia. It finds most countries, and most cities for the countries that have street data.

Good stuff.

Yesterday I returned from the Semantic Desktop Workshop.

My intention was to live blog. I didn’t find the time. Days started early and ended late. In between was way too much high-information goodness to let any of it pass.

The two previous workshops were centered around Leo Sauermann‘s gnowsis. This time we had a wider focus, with Giovanni Tummarello and Christian Morbidoni from DBin present, as well as Max Völkel from Semantic MediaWiki and a couple of Leo’s DFKI colleagues and some of his project partners in the Nepomuk project.

I did a short presentation on user interface issues of Semantic Web applications (weird photo here). Executive summary: Semantic Web developers should decide if they target RDF enthusiasts, power users, or casual users, because UI design requirements for each group are quite different. I may write up the talk in the next days if I can find the time.

We did a lot of hacking – the hands-on part is an important component of this workshop series. Giovanni, Chrisitan and I added a simple RSS export facility to DBin. This was the first time I worked on an Eclipse plugin.

The best part of every workshop and conference is meeting people and chatting about this pretty exotic stuff we care about. I spent quite a bit of time talking web architecture, wikis, SPARQL and personal knowledge management with Max Völkel. If you read any interesting ideas on this blog over the next few weeks, you can assume it’s me fleshing out something that Max mentioned off-handedly over beer.

Anyway, hanging out with fellow geeks is always fun, and I’m seriously looking forward to the next installment of this workshop series.

The photos are by Anja, who provided photographic coverage.

The first talk is by Leo Sauermann. He gives an update on the state of the semantic desktop engine gnowsis.

“Personal information management is the main use of a PC, why is it not in the operating system?”

Gnowsis is a desktop operating system enhancement. It’s an RDF layer on top of the existing structures and resources of a desktop PC: filesystem, Outlook and so on. You can build cool applications on top of the gnowsis layer.

The main use cases are:

• taking notes on complex structures, kind of like mind mapping
• connecting emails, files, web pages etc. by tagging
• using text classification to move files – the system learns from the tags and can suggest to move new stuff to where existing similar stuff is stored

A couple of ontologies are used to integrate PIM information from different applications and sources. There’s a stack of ontologies. A standardized upper-level ontology provides concepts like Person, a mid-level ontology is organization dependent (e.g. there’s one for the DFKI), the low level is domain- and application-dependent.

A user also builds his own personal information model (PIMO), which is a personal ontology kind of thing. The user has to map his personal concepts to the ontology.

Gnowsis uses Aperture as infrastructure. Using Aperture, developers can build custom adapters that pull RDF out of different data sources (e.g. desktop apps) and feed it into gnowsis.

There’s a nice idea: use Jabber for communication between different gnowsis instances. There would be URIs like gnowsis://sauermann@server123.dfki.de/resources/pimo/leo, the gnowsis protocol is a variant of Jabber, the sauermann@ part is a Jabber ID, the path is a local resource name. This isn’t implemented yet.

They don’t use OWL for ontologies (way too slow, inuntuitive inferencing), but RDFS is not expressive enough (no validation), so they have some kind of “RDFS + X” which aims to be natural for software developers instead of logicians.