Email Classification using Behavior and Time Features

Yequin Shao, Quan Shi, Yanghua XIAO, Nik Bessis, Peter Norrington

Research output: Contribution to journalArticle (journal)peer-review

67 Downloads (Pure)


The various forms, flexible sending tricks and tremendous number of spam emails have brought great challenges to accurate email classification. In this paper, we present a behavior- and timefeature- based email classification method. Based on email logs, email social networks are built through the extraction of entities and relations from the email records using the MapReduce model. By combining behavior features from social networks and time features from email sending intervals, we adopt a Support Vector Machine based classifier to identify spammers and non-spammers. Compared with the current email classification methods, the advantages of our method are: 1) in addition to the behavior-based features, our method integrates the time feature to facilitate email classification; 2) to efficiently handle the vast number of emails, we employ the MapReduce model to extract the behavior- and time-based features on the email social network. Experiments on real email data of three years show that the proposed method achieves better classification accuracy.
Original languageEnglish
Pages (from-to)463-472
Number of pages10
JournalJournal of Internet Technology
Issue number3
Early online date31 May 2017
Publication statusE-pub ahead of print - 31 May 2017


  • Classification
  • Email spam
  • Social network
  • Support vector machine


Dive into the research topics of 'Email Classification using Behavior and Time Features'. Together they form a unique fingerprint.

Cite this