Lessons learned from playing with the KendraBase
Slots vs. ontology
- Defining object types and "slots" is a different job to defining ontology
- We are at the moment primarily concerned about
- first: defining object data types, and then Web interfacing
- secondly: gluing those slots to an ontology, as per first thoughts
"Document" vs. view
Dealing with the "sea of triples" world:
- the W3C Semantic Web model has an idea of "document" that groups things together
- it's a natural place to hook modification dates, author etc.
- think in terms of programming languages, with hierarchy of line, procedure, module, etc.
- what is our equivalent?
- we have an idea of "views" on a sea of information
- our views correspond very roughly to W3C-type documents
- but we want to have far more fine-grained attribution of information
W3C document idea:
- pro: n3 is quite readable by experts and can be written and edited like a programming language -- we lose this paradigm by going to a fine-grained model, unless we want to "flatten" the fine-grained data
- con: RDF is almost totally unreadable, even by experts, seems to be intended for generation from a database anyway, so are we losing anything?
- BUT: we can choose to "flatten" the data if we like, effectively choosing to "own" the assertions as our own. This makes cut-n-paste editing possible.. v. v. important.
- imagine "semantic rich text", which will not only hold metadata like font and style info, but also semantic information from its publisher...
- can we do this as part of the HTML/CSS framework, eg. by encoding the semantic info via CSS class attributes... so that ordinary HTML editors will pass it through, and we can recover it on the other side?
- If external triples are mutable, what happens to our data when referenced triples change?
- When we have multiple views containing the same triple, what are the semantics? Copying, or reference?
- Do we need/want complex version control?
Confusion between two concepts:
- caching as an accelerator, to eliminate need to follow URL links
- caching as an archival system, to keep authoritative copies of old information
- accelerator caching fails soft
- but archival caching needs to know how far to traverse...
- idea of document/view would be useful here???