As reported from the New Scientist, a team of researchers from IBM, and Cornell University developed a new algorithm for detecting spam in emails, called SMTP Path Analysis. The algorithm works by examining the path information (probably by looking at the Received headers), and detects patterns that are likely to be the route of a spammer. "...the algorithm is not meticulous enough to efficiently catch spam on its own, but works well in combination with content filtering tools.
Engineers at ActivSoftware recently announced their new algorithm called 'slow start outbound connection ramping.' This new server technology attempts to avoid becoming flagged as spam by automatically monitoring delivery success and failure rates and adjusting simultaneous connections to an email service provider based upon those parameters. It begins with a very low number of simultaneous connections to any one ESP for any one IP address. It monitors delivery failure to success ratios and slowly ramps up the number of connections to that ESP from that particular IP.
In another recent attempt to help legitimate email senders avoid becoming flagged as spammers, researchers at ActivSoftware, using a bayesian spam filter, sifted through over two hundred thousand words flowing through their email servers and itemized the top 50, or so, words most likely to trigger spam filters. The words are organized by their spam to ham ratio, or illegitimate to legitimate email ratio.
The team analyzed many factors within this data, but the most compelling was the spam to ham ratios. Words such as click and here don't rank as high, since they are used often in legitimate email. Whereas words like madam, rarely found in legitimate email, while readily found in spam email, had very high ratios. Using this method the team created, what they deemed, "A superior list of spam words."
The top twelve words follow:
homeowner
discreet
madam
materially
unclaimed
anticipates
soma
preapproved
unconditionally
beneficiary
refinance
intercourse