Hi Peto,
And welcome to the forum.

Like Geert, I have no hands-on experience with the Data Miner API -- it's pretty new -- but I do a lot of EA extensions, and one of the common tasks there is to modelify data from an external source.
The key issue when maintaining such a model is to know what constitutes an identity. In EA, each package, element, connector, diagram, attribute, operation etc are identified by their GUID. Names can chage, but if the GUID doesn't change it's the same element.
The question is: what is the identity concept in your data source?
Once you're clear on that, if it's something like a GUID or a numeric identity, the best thing to do is to store that as a tagged value in each corresponding element in your model. If there is no separate identity concept you'll have to use names, in which case you need to make sure that there are no duplicates of (stereotype+name) in your target model before (re-) importing: that constitutes a broken model.
HTH,
/Uffe