Re: [Gnumed-devel] patient search functionality revisited

[Horst Herb]

> There is also a function levenshtein () which I found useful:
> it gives the number of substitutions required to match two strings, so
> is language-agnostic.

problem is that the Levenshtein distance is expensive to calculate. Since you
cannot even assume that the first letter is spelled correctly (it may or may
not begin with a silent letter etc), in the worst case you would need to
traverse the full list of names calculating the distance for each single one
of them.

Due to the possible number of permutations, it is not practical either to have
lookup tables for comon cases etc in order to speed up the search.

Horst



On Fri, 3 Jan 2003 21:19, Ian Haywood wrote:


> On Thu, 2 Jan 2003 01:19:17 +0100
>
> Karsten Hilbert <address@hidden> wrote:
> > > Don't you have any phonetic sear ch algorithms? At least for English
> > > there are
> > > a number of reasonable phonetic search algorithms available
> > > ("soundex"), = even in postgres/contrib
> >
> > I know. I am not immediately aware of any phoneme reducers for
> > German. I will do some research on those in due time, though.
> > Theoretically, soundex should be trainable for all languages
> > reasonably phoneto-structurally close to English.
> >
> > However, soundex will happen in a third level query or in a
> > second level one at the earliest if the user so desires. I am
> > trying to catch the common cases first in 1st/2nd level
> > queries. Of course, 1st level queries that don't return any
> > hits will be escalated to second level automatically. Same for
> > second to third level.
>
> There is also a function levenshtein () which I found useful:
> it gives the number of substitutions required to match two strings, so
> is language-agnostic.
>
> It would also be good to support nicknames, as some people (Australians in
> particular) are notorious for using them to exclusion.
>
> Ian