Spam Filtering Methods: Bayesian vs Fuzzy Logic
Fuzzy logic operates on the notion that items are members of different sets to different degrees.
A room with a temperature of 68 degrees could be said to be equally a member of the "too cold" and "too hot" sets. A 75 degree room might belong 70% to the "too hot" set and 30% to the "too cold" set.
This may simply sound silly - but it can be very useful trying to break subjective concepts into mathematical terms that can be implemented in a computer program. For example an air conditioner may turn on in proportion to how much the room is a member of the "too hot" set. Many car transmissions now use fuzzy logic to determine optimal shift points.
SpamButcher's anti-spam filtering code uses fuzzy logic to make observations about a message to determine what extent it's a member of the "spam set" and what extent it's a member of a the "non-spam set." If it crosses a certain threshold it's considered to be spam and is filtered. The SpamButcher spam detector is available as a free anti-spam software trial.
Bayes' Theorem is based on probability theory. Using collected data points it statistically attempts to determine the probability that a message is spam and whether it should be filtered or not.
If these spam filtering techniques sound similar - it's because they are. In practice the differences between Bayesian filters and a fuzzy logic filters spam stopping ability comes down to the data set.
A Bayesian spam detector is more likely to have a data set based strictly on mathematical observations about spam.
As a fuzzy logic filter, SpamButcher uses both machine collected data in addition to human selected data points to make its spam filtering decisions. By combining both machine and human intelligence SpamButcher can offer superior spam fighting performance.