Re: Neural net - guessing literature kind

From: osmium (r124c4u102_at_comcast.net)
Date: 11/19/03 

Date: Tue, 18 Nov 2003 23:30:23 GMT

Borys Musielak writes:

> I have to write an algorithm which guesses the literature kind
> having 100 words of text.
> Of course the natural way to design it is by using a neural net
> and back propagation.
> The question is - how to design the net and what function to use
> to make it work well. And even more important task is: what does really
> differentiate poems from fiction, or criminals in terms of words (i
> think of using some syntax loops like verses with rhymes).
> Enough would be to differ poetry from prose. Better but i dont know if
> easy would be guessing prose kind: like love story, whodoneit, fantasy
etc.
> I dont expect solutions, I rather need some ideas and maybe links to
> similar projects.

It seems to me that poetry and prose differ mostly in line length and
punctuation. I was taught that Shakespeare was poetry but I never quite
bought that. It seemed more like an assertion than anything else.

For a prose classification scheme, I would think of something a bit like
the Minnesota Multi-Phasic Personality Inventory (MMPI), whcih is probably
well documented. As I see it it is a correlation problem. Discard the
prepositions and articles and focus on the nouns mostly. If the word
"tomahawk" is used it is one indicator of a western.

Relevant Pages