Re: suggestions with confidence levels

On 4/25/07, Amit Khemka <[email protected]> wrote:
>
> On 4/25/07, Amit Khemka <[email protected]> wrote:
> > On 4/24/07, Ethan Bradford <[email protected]> wrote:
> > > It doesn't do this out of the box, but if you implement patch 1489981,
> it
> > > will.
> > >
> >
> > Thanks :-), let me get a look !
> >
> > cheers,
>
> Hi, finally I got the patch working with python bindings !


Very good!

I have another query:   Given a word and its possible suggestions how
> can I decide that anything with less than score 50, is a good
> replacement candidate ! Or say if the score of the first suggestion is
> 200 rather than picking that as a possible alternative i rather ignore
> it.
>
> May be knowing the way these scores are calculated would help, or else
> i guess i will have to use some heuristics.


Empirically, just from looking at several  lists of spelling suggestions, I
came up with a cut off of 130.

The biggest advantage I got from the score was in knowing when the top
suggestions had the same score, and thus could be reorganized based on other
criteria (I have a rough estimate of frequency from another source).  When
the scores are different, I almost always use their ranking.  The fudge
factor I put in to allow frequency to override this is to raise the ratio of
the scores to the 20th power and compare that to the ratios of the
frequencies.

cheers,
> amit.
>
> ----
>
>
> _______________________________________________
> Aspell-user mailing list
> [email protected]
> http://lists.gnu.org/mailman/listinfo/aspell-user
>

_______________________________________________
Aspell-user mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/aspell-user