When reviewing, I was very pleasantly surprised to receive questions that match my most active tags. I suppose it's "by design" and it's a great thing.
But I also receive a lot of questions that really do not match my knowledge (I am a C++/computer vision guy and I am asked to review Json, php, ruby, etc). Some are easy to interpret - they are plain crap, but others need some sort of knowledge in the area to make a clear judgement. Especially the "Suggest improvement" part is compromised.
So, I suggest to limit more aggressively the questions whose tags do not match the reviewer's expertise.
I see other were concerned about the topic, on both pro and contra arguments
But I think that a good formula would be a "biased random order" towards the implicit user preferences. It should limit the "Not sure" clicks my a measurable amount.
To make a clear distinction between the related questions/requests that handle manual sort by tags, I am talking about automatic filtering based on reviewer's most active tags.