Is there a policy which tags to use when there is an obvious hierarchy? For instance, questions with tags random-matrices or matrix-analysis could also be tagged more generally matrices to make searches easier, and indeed they are more or less often. But sometimes this may seem too redundant. Likewise for extremal-combinatorics$\subset$co.combinatorics and obviously lots of other pairs.
How should that be handled best?
Two basic pieces of information:
One should always use at least one top-level tag (those with 2-letter prefix corresponding to arXiv categories).
One can use up to five tags; there seems no reason not to make use of all slots if there are enough relevant tags.
I think it is good practice to use both general and specific tags if they apply.
Like use riemann-zeta-function, analytic-number-theory, nt.number-theory
to then add zeta-functions and dirichlet-series maybe is a bit much (but there'd be no harm either).
A key-point to keep in mind is that one can follow and filter by certain tags. It thus makes sense to include general tags even if everybody seeing the more special tag would know the more general tag for if the general tag is missing they might never see the question.
More information along these lines can be found on the Frequently asked questions about tagging on MathOverflow
