(bā´ zē-en fil´tër) (n.) A technique for identifying incoming e-mail spam. Unlike other filtering techniques that look for spam-identifying words in subject lines and headers, a Bayesian filter uses the entire context of an e-mail when it looks for words or character strings that will identify the e-mail as spam. Another difference between a Bayesian filter and other content filters is that a Bayesian filter learns to identify new spam the more it analyzes incoming e-mails.
Bayesian filtering is named for English mathematician Thomas Bayes, who developed a theory of probability inference. Bayesian filtering is predicated on the idea that spam can be filtered out based on the probability that certain words will correctly identify a piece of e-mail as spam while other words will correctly identify a piece of e-mail as legitimate and wanted. At its most basic level, a Bayesian filter examines a set of e-mails that are known to be spam and a set of e-mails that are known to be legitimate and compares the content in both e-mails in order to build a database of words that will, according to probability, identify, or predict, future e-mails as spam or not. Bayesian filters examine the words in a body of an e-mail, its header information and metadata, word pairs and phrases and even HTML code that can identify, for example, certain colors that can indicate a spam e-mail.
Bayesian filters are adaptable in that the filter can train itself to identify new patterns of spam and can be adapted by the human user to adjust to the user’s specific parameters for identifying spam. Bayesian filters also are advantageous because they take the whole context of a message into consideration. For example, not every e-mail with the word “cash” in it is spam, so the filter identifies the probability of an e-mail with the word “cash” being spam based on what other content is in the e-mail.
Proponents of Bayesian filters assert that the filters return less than one percent of false positives.
Other forms: Bayesian filtering (v.)