Re: Fuzzy string matching of product names

On Monday 5. April 2010 22.00.41 Peter Geoghegan wrote:
> similar they sound. How can that actually be applied to get the
> functionality that I've described?

I've got a similar problem in my 18th century research, when clerks usually
took pride in being able to spell a name in any number of ways. I've landed on
a solution where I'm sending search strings to SIMILAR TO. I usually get far
too many hits, but it's much easier to browse through 100 hits than the entire
dataset which is approaching 60,000 records.

Optimizing the search strings is based upon a lot of experience.

It would probably be better to add a column with normalized names, but the
amount of work involved with that is staggering. I eventually associate most
of the records to «persons» with normalized names, but the search process can
sometimes be very frustrating, and it would really help with some kind of
fuzzy search.

Just in case anyone should suggest it: Both Soundex and Metaphone are useless
for Norwegian 18th century names.

regards,
--
Leif Biberg Kristensen
http://solumslekt.org/