Personally I have just been using the full URL as the identifier but
this doesn't work that well in the case of canonical URLs. Splitting
the URL up might have some advantages but you would really need two
parts of the identifier, one for the video itself (like UFnXm6cjGwU or
17853047) and one for the scope of that (youtube or vimeo). You could
maybe use provider_url for the latter.
Personally I would prefer just having a normalised link back to the
original web document. Then this can also be used to link back to the
source in a clean way without actually carrying through any extra
parameters such as "feature=topvideos" etc...