Stevan,
I would be seriously impressed by any algorithm that could tell me the core subjects of 'The Wasteland' by T S Eliot and then locate other works on the same themes.
Could your algorithms even tell me it was a poem?
I'm sure they would work well on certain kinds of document in certain subject areas and I would be happy to see them used. A good indexer needs to be quite intelligent but indexing is not a very satisfying job for an intelligent person.
I suspect your algorithms would be like many expert or rule-based systems, superior to most human beings in their specialist areas but weak to useless outside them.
Regards,
John.
> -----Original Message-----
> From: Repositories discussion list [mailto:JISC-
> [log in to unmask]] On Behalf Of Stevan Harnad
> Sent: 13 March 2008 14:08
> To: [log in to unmask]
> Subject: Re: Challenging assertions
>
> On Thu, 13 Mar 2008, Peter Crowther wrote:
>
> >> From: Stevan Harnad
> >>
> >> (3) Do you really believe that in the online age, any
> prefabricated
> >> classification system could ever beat boolean full-text search?
> >
> > Depends. How much automation am I allowed to have in my
> prefabrication,
> > how often can I update it, and at *exactly* which task(s) am I
> supposed
> > to beat Boolean full-text search? I can conceive of tasks at
> which
> > dynamically-generated classification systems can beat text
> search, but
> > these systems certainly aren't of the coarse grain of Dewey or
> LCC and
> > I wouldn't expect humans to enter or update the terms.
>
> I think Peter's point is relevant and correct:
>
> What I meant by "prefabricated classification" was something like
> Dewey hand-tagging of articles at source (the IR), a-priori.
>
> Of course boolean search alone is beaten by boolean search PLUS
> dynamically-generated supplements such as Google Page Rank,
> download
> stats, user tags or full-text-based classification algorithms.
>
> But not by Dewey-tagging at source; and I doubt that even boolean
> search
> PLUS Dewey pre-tagging produces incremental gains worth the pre-
> tagging
> effort.
>
> (I also bet you algorithms applied to the full-text database a-
> posteriori,
> as a whole, could generate a pseudo-Dewey classification scheme
> by computation alone that was at least as good as individual Dewey
> hand-tagging a-priori.)
>
> Stevan Harnad
|