I'm doing a fuzzy matching on accounts and have their name, date of birth and citizenship as additional factors.

I am first evaluating the jarowinkler score of 2 names and if it is above a threshold of 0.8 ( to reduce the set the list has to iterate over) a weighted product model is then used for the jaro winkler scores of their name,d.o.b, citizenship to identify if 2 accounts are the same with different spelling of names.

My issue is that with a weighted product model the sum of the weights should = 1, if i do that with the jaro winkler scores (0-1) then lower scores will be multiplied far more than higher scores as: $$x^{0.y}-x> z-z^{0.y} $$ where z>x.

This would mean lower scores will be overestimated.

If there was a way to put the weights as 7 then that would propel jaro winkler scores of 0.7 and above and decrease the chances of scores <0.7 to be misrepresented.

What would you suggest to use a weighted product model on scores between 0 and 1 where higher scores are highlighted more than lower scores? Would the best way to multiply by 10 then use the model as it eliminated my issue of numbers<1 being raised to a power<1 ?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.