From today’s Irish Times comes a story which shows the real significance and impact of a common Information/Data Quality problem, transposition of letters or numbers.
A Swedish couple holidaying in Italy were looking forward to their visit to the lovely sunny island of Capri.
Unfortunately a “finger flub” on their GPS put them 650 kilometres north and inland of their intended destination in the lovely Italian industrial town of Carpi.
Oh dear.
Have you heard about the Syrian lorry driver going from Turkey to Gibralter who ended up at Gibralter Point, a nature reserve on the East Anglian Coast! Apparently Gibraltar is listed as being in the UK which is true politically but not geographically! This explains why the driver was not alerted by having to cross the channel!
This article has more interesting examples and an insight into the psychology of people who abrogate all responsibility to the machine, in the face of the evidence. There’s a lesson there for those responsible for driving companies in the right direaction!
I have been there – and I am glad I am gone now. I once was responsible for the search functionality a major international e-commerce site that used the RDBMS’ built-in SOUNDEX capabilities to match product names. It was dog-slow and seemed to produce more false positives and false negatives that desired hits.
At peak time (Holiday Season) all fuzzy matching capabilities had to be turned off because the 96 or so DB CPUs could not take the extra load.
Needless to say I replaced the technology there, and later I did the same at a sister company using the Netrics Matching Engine which is based on mathematically modelling the human notion of similarity and achieved a measurable ROI in 1-2 months.
BTW I would add metaphone, double metaphone, Levenshtein/string edit distance , Jaro-Winkler, Q-gram, skip-gram etc in the same category – they are ill suited to deal with all the matching challenges caused by data variations across domains in our modern multicultural times.
Naturally, I am biased towards a mathematical approach that overcomes the issues in both accuracy and performance inherent in the these methods.