Active Learning for Matching Problems