> -----Original Message----- > From: Marc Perkel [mailto:[EMAIL PROTECTED] > Sent: Monday, December 27, 2004 7:35 AM > To: Spamassassin Dev List > Subject: Re: A Feature I've always wanted - Test for multiple hits on > same rule > > > I'm not sure you would have to take this into account - but it would be > interesting to test the results. I would want just a couple of system > settable commands like: > > TwoHitFactor = 1.2 > ThreeHitFactor = 1.3 >
This isn't germane to the scoring algorithm per se, but perhaps more having to do with fine-tuning the Bayesian algorithm, somewhat along the lines you outline above: http://www.andrew.cmu.edu/user/dgovinda/pdf/multinomial-aaaiws98.pdf Abstract Recent approaches to text classifcation have used two different first-order probabilistic models for classification both of which make the naive Bayes assumption. Some use a multi-variate Bernoulli model, that is, a Bayesian Network with no dependencies between words and binary word features, e.g. Larkey and Croft, Koller and Sahami Others use a multinomial model, that is, a uni-gram language model with integer word counts, e.g. Lewis and Gale, Mitchell. This paper aims to clarify the confusion by describing the differences and details of these two models and by empirically comparing their classification performance on five text corpora. We find that the multi-variate Bernoulli performs well with small vocabulary sizes, but that the multinomial performs usually performs even better at larger vocabulary sizes providing on average a 27% reduction in error over the multi-variate Bernoulli model at any vocabulary size.
