"Sigfrid Lundberg, Lub NetLab" wrote:
...
> The only problem I have with this algorithm, is that it applies at
> indexing time (i.e., when compiling to triples) only. Would there be
> any problems algorithmic (or for performance) to recast it, for doing
> something similar at search time?
No. (Well, at least not in theory ... ;-)
This is simply the problem wether to include 'implicitly contained'
information
of a semantic net like RDF at modification time in the net or to
generate it on the
fly when searching. A few examples for 'implicitly contained'
information
are presented in the RDF Schema specification, e.g.
a -type-> A -subClassOf->B => a -type-> B
You can view this axiom as a transformation of labelled directed
graphs and you can decide whether you perfom this transformation
when new data enters the graph, or wether to generate it on the
fly for searching. You must only decide if you have a lot of
time (for searching) or a lot of disc space (for storing).
The described algorithm is simply defines a graph transformation
that can be executed whenever you want.
Greetings,
Stefan
|