[rdfweb-dev] Re: Parsing FOAF in perl?

[Danny Ayers]

>Well one approach I've played with (in the thought experiment sense)
>is a kind of evolutionary approach. In this scheme each batch of
>statements is considered a source model. These source models are
>combined to produce inferred models provided there are no logical
>contradictions. As new source models are added they are merged into
>new inferred models. All the models co-exist and are independent. When
>the system is queried all models are consulted and some kind of
>fitness function is applied so that inferred models that are
>consistently wrong are gradually eliminated.

Like it!

>Its a sort of experience based reasoner. Like a human it will make
>mistakes while it is learning who to trust.

I've thought for a while that the logical semweb will benefit from (need!)
tools from old-fashioned AI, though I hadn't really thought much beyond
scraping. But the fitness function you describe can probably be lifted
straight out of a neural net textbook.
I guess until we have big flows of statements going on it won't be possible
to properly test, but the idea sounds absolutely feasible.