A two-tiered user feedback-based approach for spam detection
School of Science
The current practice for spam detection works through binary classification of a message as either spam or ham. We propose a novel technique based on solicitation of user feedback in the spam classification process. The spam classifier proposed is semi-automated in nature, and is trained dynamically to include words and word-variants into the spam dictionary. Thresholds are defined to ascertain that spam and ham messages are accurately classified with highest probability. In addition, a set of messages that do not fall into the above two categories are tagged as grey messages. These messages are reclassified as ham or spam based on user feedback. Results obtained through experiments proved the superiority of the two-tier spam classifier over the single-tier spam classifier.