The first step to fixing a problem, is to understand the problem.

I know, many people will possibly agree that this problem exists - but that sort of empirical observation isn't all that useful. If you want to try a method to fix it - any method at all - you need to first have a reliable way to measure whether it actually improves things.

And for that you first need to find a way to actually measure this - in a reliable, repeatable, unbiased way. You need to find a way to measure all of Wikipedia for this sort of problem. Not just the current state - but through the change history for any arbitrary point in the past as well.

Then do some thorough data analysis, trying to learn as much about the problem as you can. That's your best bet for actually solving the problem.