Bayesian spam filtering library for Python

问题:

I am looking for a Python library which does Bayesian Spam Filtering. I looked at SpamBayes and OpenBayes, but both seem to be unmaintained (I might be wrong).

Can anyone suggest a good Python (or Clojure, Common Lisp, even Ruby) library which implements Bayesian Spam Filtering?

Thanks in advance.

Clarification: I am actually looking for a Bayesian Spam Classifier and not necessarily a spam filter. I just want to train it using some data and later tell me whether some given data is spam. Sorry for any confusion.

回答1:

Do you want spam filtering or Bayesian classification?

For Bayesian classification there are a number of Python modules. I was just recently reviewing Orange which looks very impressive. R has a number of Bayesian modules. You can use Rpy to hook into R.

回答2:

Try Reverend. It's a spam filtering module.

回答3:

RedisBayes looks good to me:

http://pypi.python.org/pypi/redisbayes/0.1.3

In my experience Redis is an awesome addition to your stack and can help process data at blazing fast speeds compared to MySQL, PostgreSQL or any other RDBMS.

import redis, redisbayes
rb = redisbayes.RedisBayes(redis=redis.Redis())

rb.train('good', 'sunshine drugs love sex lobster sloth')
rb.train('bad', 'fear death horror government zombie god')

assert rb.classify('sloths are so cute i love them') == 'good'
assert rb.classify('i fear god and love the government') == 'bad'

print rb.score('i fear god and love the government')

rb.untrain('good', 'sunshine drugs love sex lobster sloth')
rb.untrain('bad', 'fear death horror government zombie god')

Hope that helps a bit.

回答4:

Try to use bogofilter, I'm not sure how it can be used from Python. Bogofilter is integrated with many mail systems, which means a relative ease of interfacing.

回答5:

SpamBayes is maintained, and is mature (i.e. it works without having to have new releases all the time). It will easily do what you want. Note that SpamBayes is only loosely Bayesian (it uses chi-squared combining), but presumably you're after any sort of statistical token-based classification, rather than something specifically Bayesian.

回答6:

A module in the Python natural language toolkit (nltk) does naïve Bayesian classification: nltk.classify.naivebayes.

Disclaimer: I know crap all about Bayesian classification, naïve or worldly.