I write it with more details as an idea.
Let say that each user is get a hidden score when they accept or reject an edit.
There let say that we give this score.
A, B -> if they agree they get both +1
A, C -> if they not agree they get both -1
On one edit if we have A,C,D the A disagree with C and D, so A gets -1, C D, did not get anything as score.
Remember we try to find the one that are act randomly, not get the high score.
If A, act randomly is probably gets a lot of -1 on some tags, and then you can calculate the percentage of -1, for example on 100 edits, if 50% or more are disagree with others (-1) then is probably act randomly - or in general is not in the same line with the community.
Now maybe A, did not know anything about that tag and is only make mistakes about that tag. I suggest to be eliminating per Tag. Eg, I know nothing about java and if I accept or not on java tag maybe my edits on that Tag are wrong, but not on asp.net that I know more.