When I've seen:
- obvious tag errors (e.g.,
baysian
instead of bayesian
),
- where there is already a tag pattern (e.g.,
distributions
instead of distribution
; at the point I changed it, I think there were 6 distributions
and only the 1 distribution
), or
- where there is another tag that would be appropriate (e.g.,
fundamentals
),
I've changed or added the appropriate tag. So obviously my vote in those areas is to streamline. Anything beyond those three scenarios, I'd be hesitant to do without running it by the community.
Along those lines, if you do streamline/clean-up tags, be warned that there may be almost a daylong timelag before the daily "janitor" cleans up the incorrect or orphaned tags.
As for your data
versus dataset
example, I'd suggest you start a specific meta question. Personally, I'm on the line and would wait to see what emerges over the next day or so.