KendraBaseProgress

Progress so far on the KendraBase project.

Test server up running:

• Debian unstable
• PostgreSQL?
• Python2.3 (have resolved pyPgSQL? issues...)
• pyPgSQL? 2.3
• Some simple web scripts

Implemented so far:

• URI-based triple store
• very simple prefix stripping and replacement...
• Simple user interface with triple editing
• DB dump and load
• Simple interfacing to Tim Berners-Lee's cwm engine
  • Via external process and pipes, at the moment...
• Various user-views: Wiki-style, RDF, triples, browsing...
• Everything is Unicode-transparent, so multilingual support is easy (check this: is cwm Unicode-transparent?)

Some checks:

• basic logical inference with cwm has been tested
• basic reasoning should only take a couple of seconds on a fast machine for limited inference worlds of a few hundred facts
• this should be the size of our ontologies...

Now it's time to start serious integration of the work done so far.

Lessons learned:

• See KendraBaseLessonsLearned

Things to do:

• at the moment, the web UI is a single fused mass of make-hook, do-action, and display-page code. This should be made pluggable ASAP.
  • make-hook: making a form or link that will activate a server action
  • do-action: what it says -- kicked off by hook URLs or form GET/POSTs?
  • display-page: display a page or part of a page, based on the aftermath of a hook/action sequence
• unify cookies etc into this framework

• Make external converters more easily added...

Daniel would like:

• Ability to have images and freetext be either Subject or Object
• Multiuser environment
• User preferences area
  • Dictionaries for translating between local and global terminology like:

Local	Global
has	kendra:has_a
David Bowie	http://www.musicbrainz.org/showartist.html?artistid=956
Daniel	people:Daniel_Harris_68467453

Restructuring

• Core HTML / CGI / URL / cookie / login services
• Database services Done
• Pluggable external data transformation (N-triples <-> N3, RDF) Done
• Page generation, actions, action dispatcher...
  • table&hook-driven rather than hard-coded

Questions

"Virtual document" idea:

• a single internal representation for
  • editable triple set
  • generated RDF
  • generated n3
  • etc...

• Do we add this as other fields (baseuri, linenumber) to the already-overloaded triple store?
• Or to build an external document table structure.

Fields for this are simply:

  baseuri linenumber triple_id

Note that this also should be possible to be reified (!).

See: KendraBaseLessonsLearned for more on this.

• Don't forget to add URLescaping to UTF-8 unicode chars...
• Is it yet time for full end-to-end native Unicode strings, rather than UTF-8?

Short-term goals

Just two short-term goals, for now:

1. Import example test database into software
2. Add auto-generation of edit/display form for object classes

Longer-term

• Direct database storage for generated datatypes, and XML generation from this...
• Logic / ontology-based SQL query generation (???)

• Adding a native DB interface to CWM?
• Finding an alternative processor module?
• Write our own?

• Replace underpinnings with something like [Redland]? (uses LGPL or MPL licenses)