I worked on some of the large scale fuzzy matching projects. Based on my experience you are in the right track.

You already have a table to which you load your fuzzy results. Add a derived column named IsMatchFlg and default the value to 0. Once you finish loading the data into the table run a SQL script to flag you aceptable record as IsMatchFlg=1.

This flag should be applied to 1 record per batch which is the record having maximum similarity score.

Then a select from this table where IsMatchFlg=1 will give your matching records.

I guess this helped. Let me know