Abstract
Spam has become one of the severest problems for today's network systems. In this paper, we present an adaptive spam filtering mechanism based on message fingerprinting. In our mechanism, each message is represented by a fingerprint vector, and two messages with a short distance in their fingerprint vectors are viewed as variants of each other. We present methods for fast matching a query message against a list of known spam messages, and methods for adaptive updating of the fingerprint vectors of known spam messages. Experiments on real spam data demonstrate the effectiveness of the proposed method.