Incremental SVM Model for Spam Detection on Dynamic Email Social Networks

Early Detection of Outgoing Spammers in Large-Scale Service Provider Networks

“…Our data set SUNET [15] NTU [16] Kossinet et al [19] Enron [14,20] It can be seen that the rate of email traffic (number of emails sent or received per day) logged in our data set is between 1-3 orders of magnitude higher than that of the other data sets. The reason for this huge difference is that, whereas previous studies of outgoing spam mostly use small sized datasets taken from academic institutes [16,11,13,18] or from Enron's public data set [14], our study uses a data set collected by the mail servers of a large ESP.…”

“…The reason for this huge difference is that, whereas previous studies of outgoing spam mostly use small sized datasets taken from academic institutes [16,11,13,18] or from Enron's public data set [14], our study uses a data set collected by the mail servers of a large ESP. Table 2 provides additional information regarding our data set.…”

“…The number of outgoing messages sent by an account is used to compute some of the features we defined above. We use it also as an independent feature as was done in some prior works [13,14,16]. If a message was addressed to multiple recipients, we refer to each recipient as an additional outgoing email, e.g.…”

“…Previous studies of outgoing spam also made use of email traffic originating from academic institutes [13,11,16] and of log files collected by non-ESP organizations [14].…”

“…Whereas content-based filters consider the properties of individual messages, a different approach examines the social interactions of email accounts, reflected by inter-account communication patterns, since, in many cases, the social interactions of spammers and non-spammer users are significantly different [11][12][13][14][15][16].…”