Adaptive e-mails intention finding system based on words social networks

Although many anti-spam techniques have been proposed till date, a foolproof solution for overcoming spam has not been found yet. Spammers still spread spam by using invariant intentions such as advertising and phishing; these intentions are difficult to detect using signature-based or content-based spam filters. In this study, we have proposed an adaptive e-mail intention finding system based on the E-mail Word Social Network (EWSN) that can detect the e-mails' intention and can adaptively and continually learn. EWSN is a data structure used for profiling a user's intentions through solicited and unsolicited e-mails. The EWSNs are constructed on the basis of the information in the user's mailbox and the expanded social relations of words obtained via search engines on the World Wide Web. Unlike previous approaches of spam filters, our system only requires a small amount of training data and it can be trained through feedback incrementally. Experimental quantitative results demonstrate that the misclassification rate, precision rate, and recall rate are better than several content-based filtering methods using a limited amount of training data. The quantitative results also demonstrate that the proposed method has good detection ability in the case of novel spam e-mail detection, without constantly updating the pattern of novel spam e-mails. The proposed method - capable of intention profiling and continual adaptation - is robust for detecting spam e-mails.