A Study of Neighbor Users Selection in Email Networks for Spam Filtering
Yongchao Wang, Yuyan Chao, Lifeng He · 2018
Due to the protection of data security and personal privacy in email networks, it is difficult to select the neighbor user using email user's personalized attributes and behavior data. In daily life, the more interactive the people are, the more similar they are. So many researchers use the interaction strength to select the neighbor users. But this method has not been experimentally verified. Receiving the same email is the most basic condition that a neighbor user can provide a valid recommendation to the target user. So, we build an email corpus using real email data to verify the selection effect of different methods. The emails come from Enron email data set, log files of a Chinese corporate email server, and volunteers. This corpus contains more than 870,000 users and more than 5.14 million emails. With this corpus, we conducted experiments on seven different methods of neighbor user selection. The experimental results show the effect of user interaction strength is far from being as good as expected, and far inferior to user social attributes. In addition, the recipient performs far better than the sender. This paper provides a useful reference for the research of spam filtering based on user-based collaborative recommendation filtering in mail network.