Occasionaly I have to de-dup mailing lists we receive from the government. Since my lists often include duplicates of the same address and duplicate names (different addresses) I use a variety of methods.

First I get the list with identical names and cities (uppercaseing both to avoid case issues). I suppose that probably gets false positives but we would prefer that in our case.

Then I take all the addresses and pull ones that the first 7 characters match on. It is pretty hard to have the same first 7 and note be a duplicate. Agian that is with full uppercasing of all fields. This also produces false positives but combining it with a name match makes it prety efficient. Depending on the list and the source I vary the number I choose. Normaly choosing a couple numbers and hand sampling until I reach what i feel is a happy medium.

These methods have helped me narrow 15k addresses down into a more accurate 10k. I think any time you try matching like this you are going to get false positives, but if you are looking for a way to flag some for human intervention then this can be a pretty good test. Best of Luck.

Sorry I meant to mention that the benifit of these is no regex, just plain old DB functions that are easy and quickly handled. For better results you could create a new colum and do some normalization like suggested by some above.