Description:
DSPAM (as in De-Spam) is an extremely scalable, open-source statistical anti-spam filter. While most commercial solutions only claim a mere 95% accuracy (1 error in 20), a majority of DSPAM users frequently see around 99.95% (1 error in 2000) and can sometimes reach peaks as high as 99.991% (2 errors in 22,786, as with one particular user). DSPAM presently functions as both a server-side agent for UNIX email servers and a developer's library for mail clients, other anti-spam tools, and similar projects requiring drop-in spam filtering. DSPAM has been implemented on many large and small scale systems with the largest being reported at about 350,000 mailboxes and is presently being used or planned for use in multiple commercial spam filter solutions.

Author: Jonathan A. Zdziarski
Homepage: http://www.nuclearelephant.com/projects/dspam/

App rating details:
Total votes: 0
Overall rating: 0

How'd this get started?: Jonathan: "I originally wrote DSPAM to filter mail for some pastors in my church who were also running an online ministry/chatroom and using email on my server. They (and I) were getting upset at the massively increasing raunchiness of some types of spam. My only solution at the time was to change their email address every few months and set up a little procmail magic with word-matching, which really didn't work all that well.

...DSPAM's scalability and dedication to R&D are probably DSPAM's two strongest suits...

DSPAM was built to filter mail for large ISPs, universities, corporations, etc. To my knowledge, DSPAM's largest installations are up around 100,000 users. This helps address the spam problem on a grand scale rather than relying on the end-user to be savvy enough to understand filtering. In order to make any dent in spam, scalability is a must IMO.

From an R&D standpoint, I've dedicated much of my spare time to developing new algorithms and approaches that I feel are making a big difference in the quality of spam filtering, which adds flavor and personality to DSPAM setting it apart from "generic bayesian filter". Bill Yerazunis (CRM114) and I released an Internet-Draft in January outlining a message inoculation format allowing different users of different spam filters to "inoculate" eachother with new types of spam. In February, I released an algorithm called "Bayesian Noise Reduction Logic" which performs a type of Dolby on text (it was actually named Bayesian "Dobly" - an old spinal tap reference - until Dolby Labs made me change it). I've also begun to code some unique (but still experimental) neural networking functions into DSPAM and am working on some other ideas. One of the great things about this area of work is all the energy in the community. I've had the pleasure of being able to chat with some much smarter folks than I such as Gary Robinson and Paul Graham and have included much of their research into DSPAM as well.


[ Download DSPAM now! | Get Support for DSPAM at LinuxQuestions.org ]