I'm wondering whether cleaning the bounce mail is done by parsing the returned message text and creating rules (if contains bla bla bla) or by extracting a certain error code number.
Please clarify.
Bounces are handled using VERP, i.e. a specially formed reverse-path. The reverse-path address is the one where to bounce will be sent, and it contains the original recipient in encoded form and maybe other informations too. Something like: bounce-john+example.com@list.example.org. See Variable envelope return path.
The analysis of the bounce mail is good for retrieving the exact cause of the bounce, especially if the bounce is sent in the standard DSN format, but otherwise it is not a reliable method (for example not all mail servers use DSN).