SQL & PL/SQL :: High Volume Similarity Grouping

I am in the very early planning stages of a project the goal of which is to identify separate organizations which may in fact be the same organization.

Our first implementation of this task was a process designed to look for a few thousand organizations in a pool of a few hundred thousand organizations. To accomplish this we made heavy use of Oracle's Text index as well as a custom index type we created which utilized n-grams. This approach worked quite well for on-demand editing of the organizations, in which a user might log in and say in addition to what we already know about organization A we also know x, y and z does that change anything and worked acceptably well for the bulk processing we did on our "known" information once a week running for a couple of hours on the weekend.

We have now been tasked with reworking this initial implementation only now we want to look at a set consisting of several million organizations for potential matches which exist within the set. As in our initial implementation we will be breaking what we know about organizations into groupings so we aren't comparing a phone number to an email address and normalizing the data as much as we can so we ignore things like case and punctuation. Even after all this we are still talking about looking for similar values in a group which might be in the tens of millions (some types of data will have more than one value per organization).

My initial thought on the problem is to use n-grams though not in the way we did in the past. The basic idea here is that we break the search values up into all the substrings it is made of and look for other values which have a high number of those substrings in common.

SQL & PL/SQL was the best place for the question, but I could not think of a better one.