Tabulator: AJAX Generic RDF browser

Browsing data on the Web
Software is released at 0.7, almost beta 0.8
Firefox only, Opera widget for rel 0.7
Open source & rippable
http://dig.csail.mit.edu/2005/ajar/ajaw/tab
tabulator@w3.org - developer team

Tabulator people

Tim Berners-Lee

The Tabulator was Tim's idea and he coded up the original version at odd times in November and December 2005. See Links on the Semantic Web from Dec 2005

Undergraduate Research Opportunity Program (UROP) over June-August 2006:

Yushin Chen

"Joyce" wrote the calendar views, and incorporated the Simle timeline.

mugshot

Lydia Chilton

Lydia is working on statistical analysis, charts, etc.

mugshot

Ruth Dhanaraj

Ruth worked on the Tabulator in January 2006, adding the asynchronous fetching of documents during queries, etc.

Adam Lerer

Adam works on the back-end -- the query system, and generic stuff around the query UI. He implemented SPARQL protocol as a internal layer and as an external protocol.

mugshot

Jim Hollenbach

Jim is responsible for the map view. He also worked on the framework for multiple queries per view.

mugshot

David Sheets

David wrote the RDF parser, and does a lot of architecture and release engineering. He also installed the bug tracker.

mugshot

Thanks also to Dan Connolly and Rlph Swick for co-supervising students and for ideas, support, testing and encouragement

Interesting things

Generic, not app-specific browser
Avoiding rdf, URIs and RDF jargon in your face
Not using circles and arrows, which take up too much space
AJAX to be developer-hackable
Looks up data on web, follows links: Linked Data
Outline mode (for a web?)
Query by example -> table
Lenses and views (to come)

Features

Does GRDDL and link rel=meta on xHTML pages
RDF/XML parser (not complete), no N3 yet
Smush on owl:sameAs, [inverse]FunctionalProperty
Display with subproperties of rdfs:label
Load all ontologies (recursively)
Load data on browse in outline view
Load data as query binds new URIs
Provenance: click on any fact -> highlight source in list
Provenance: double-click on source -> look at source meta inc.

Practicalities

Firefox only as we only use firefox, research project.
Must set the codebase principals support (see tabulator help)
Charles McCN made an opera widget
SVN trunk much improved after the summer of 2006
HTTP 303 handling not standard in WebAPI XMLHTTPRequest() but managed in Tabulator

Multi-query Views

Table view (only one query at a time)
Google map (lat, long) (multiple)
Calendar (datetime) (multiple)
Simile timeline (datetime) (multiple)
User-added views

Linked data

New DesignIssues note: Linked Data

Use URIs as names for things
Use HTTP URIs so that people can look up those names.
When someone looks up a URI, provide useful information.
Include links to other URIs. so that they can discover more things.

Ideally: all links out and in
Reasonable size of data returned
Not knowledge bases in zip/tar files!
Link: The use in document d of URI which is in document d'

AJAR Library

Fully tested RDF/XML parser
Indexed triple store
Asynchronous query engine
SPARQL client and server (limited)
Smushes identical nodes (owl:sameAs, owl:InverseFunctionalProperty, owl:FunctionalProperty)
Looks up data
Tracks history if HTTP results

Lookup Algorithm

Conventions - should the whole set be a standard?

What to look up?
- document from URI (up to #)
- link rel=meta
- rdfs:seeAlso links
Follows 303 redirection
Content-type is definitive
- Text/plain: Guess with a warning
- application/rdf+xml Parse as RDF/XML
- text/n3+xml Parse as N3 (*not imp.)
- text/html
  - link rel=meta in HTML
  - GRDDL profile
  - RDF/a ??
- *+xml*:
  - GRDDL profile

What to do about large number of links

Pointer to document for each predicate?
- e.g.
Pointer to SPARQL endpoint?
How to describe what a SPARQL service can answer?

Tabulator-friendly data

linked data plus:

Use rdf:label to label things
Declare usable name properties to be subproperties of rdfs:label
Predicate labels are role nouns e.g. 'parent'
For predicates, also label unnamed inverse predicate, e.g. 'child' to void 'is parent of'
Not 'has parent' or 'is parent of'
Lower case for predicate labels is fine, tabulator can uppercase where needed
Inverse Function Properties (IFPs) of external items, e.g. homepage, email, etc for merging with other sources