A team from the Deutsche Telekom Laboratories in Israel have devised an anti-spam tool that can reduce the torrent on junk mail by 94 per cent, without having to analyse the content of messages.
Eitan Menahem and Rami Puzis used a machine-learning algorithm to classify the reputation of mail transfer agents and identify the likelihood that messages coming from a sender would be spam.
The benefit of the system is that it can dramatically reduce the volume of spam being received without analysing the content of emails, the researchers said.
“While content-based filtering [methods] are considered as the most accurate spam mitigation methods, they are also the most computationally intensive and sometimes considered as privacy infringing,” the researchers said.
And while blacklists, such as those run by the Spamhaus group, are effective tools for the blocking of spam, spammers frequently change IP addresses or even steal legitimate to avoid the filters.
The other major technique used is the classification of senders by reputation, but many of the sender reputation mechanisms (SMR) have been built with a limited data set.
As the Deutsche Telekom researchers had access to an email log with 9.5 million entries, made up of email headers, which were received over the period of a week at one of its DT's internet service providers.
The researchers then set about using so-called machine learning algorithms to learn how to identify headers that were indicative of spam, and to analyse how the behaviour of the mail transfer agents behaved over time.
Their system was able to block 94 per cent of all spam messages detected, with only 0.5 per cent of messages flagged as spam turning out to be legitimate email.
The researchers were also able to show that their system could detect four out of five spam messages that had made it passed the blacklists used by the ISP.
The system could reduce the workload put on ISPs and large enterprises when it comes to filtering out spam, the researchers said.
Dust storm on Titan only the third Solar System body where such storms have been observed
New technique could enable quantum computers to scale-up to millions of qubits
Systrom and Krieger taking time off "to explore our curiosity and creativity"
Comcast's £29.7bn winning bid more than twice the £13.7bn Rupert Murdoch valued Sky at just eight years ago