An interesting problem. ;-)

It seems you need further criteria than just Levenshtein. I would implement some word-based checks additionally (whose results you have to combine somehow). My ideas are

• if one string is substantial longer than the other, check if all words of the shorter string are contained in the longer
• if some words contain mostly capital letters, expect them to be in both strings (and exclude them from the Levenshtein-check)

Of course you need to experiment with the weights of additional checks (what mean "substantial longer" (more words?), "mostly capital letters" (>=80%?)...), and of course you can easily construct examples where they won't help you (e.g. George W. Bush jr. is not the same as George W. Bush sr.). ... but maybe it helps.

Please keep us up-to-date with your solution!