"Learning" filter engines -

Are there any "intelligent" or "learning" engines, which are capable of identifying "bad" phrases in texts (For example, if I want to filter texts with mail address:

  asdasd asd as d dgfdgfdgfdg sadasd (as used in thirdbird)  
 Asfsdf.com  

The first tool will not be identified as an email address ... but if the user is "taught" (click Or "got a mail address in the text" for example) for tooltips, many times, that text contains phrases like "xxxxx (at) xxxxx.xx" is suspicious, it "learns" that in the future This text should be automatically marked ...

Question is something like that on the market? I am wasting some libbs (like spam assassin, etc.) but these are "special" on email ...

The general idea you are talking about is one. Maybe he'll help you in your searches.

Edit: Some other examples:

< / Ul>