Published 29 Jan 2004
As I’m currently working on finishing my dissertation, I stumpled upon some very interesting spam message recently.I’m writing about macchiato, a spam filter library, written in Java, where I’m using a naive bayesian filter to detect spam. Its an adaptive filter which gets trained by the user’s messages. As I’m close to finish this thing, I had to draw a line where I won’t go further optimizing the library - at least as part of my dissertation. Development will of course continue - Milestone M3 of Columba will have this integrated.Now to my recent discoveries in the faszinating world of penis enlargements, horny women and nigerian millionars…Most spam message tend to be written using HTML. Now take a look at the word Free in html:
Fr<font size=0>&#38nbsp;</font>eeDid you see the trick? &nbsp; is actually just a space, but it has a zero font size, so it looks correct to the user, but my text tokenizer won’t recognize this word. It will recognize Fr and ee as independent tokens, though. Which makes no sense :-(I’ll have to further improve the message tokenizer to be more aware of html than it is currently.