Human-Computer Etiquette

328 Human– Computer etiquette

later deletion. As machine learning lters improved, with advanced

similarity-matching methods and compression techniques (Goodman,

Cormack, and Heckerman, 2007), spammers sent more spam to

counter their losses, and found new ways to spam (Cranor and

LaMacchia, 1998). When machine learning lters identied spam

words like “free,” spammers wrote “f-r-e-e,” or inserted blank HTML

comments like “f<!---->ree” which became “free” when rendered.

Spammers can now bypass text detection entirely by sending spam

images inside random innocuous e-mails impervious to text lters. If

providers develop image-matching lters spammers can randomize

image content, or if providers block Web image e-mails, spammers can

embed ...