Fuzzy match - Githubissues

lhcb / Condorcet

3 stars 3 forks source link

Fuzzy match #24

Closed gdujany closed 3 years ago

gdujany commented 9 years ago

Attempt to mitigate issue #23

alexpearce commented 9 years ago

This seems very brittle and difficult to maintain. Is this the solution we're going with?

gdujany commented 9 years ago

Not necessary, but it is a possibility:

from my current understanding (I hope to be proven wrong) the only information we can retrieve from the membership database to identify a voter is his name and surname
Name and surname may be different in the two databases

IF the above is true I can imagine only two solutions:

correct the membership database
be less fussy when comparing

The 1 a priory is the way to go even if it's a lot of work. I wanted to try 2

betatim commented 9 years ago

Fuzzy matching is a nice idea. I think people will worry though that someone who is not allowed to vote will end up being allowed to vote. Can we think of a good way to estimate the probability for that happening? Does glance contain the authorlist of ATLAS/CMS? If so we could see if any of those would be classed as a "match"?

gdujany commented 9 years ago

I did a check with a csv file I got form Atlas' Glance and I obtained 22 truly different matches for 5071 attempts (~0.4%). There were also 13 true matches like Antonio Di Domenico. One should also consider that people not in the z5 group cannot see the voting page irrespective if they are in the author list or not.