User Preference-Based Spamming Detection with Coupled Behavioral Analysis

Abstract

Nowadays, the explosive growth of unsolicited emails on Internet has been challenging the spam filtering systems when at the presence of big data. Current spam filters suffer from the following problems: (1) Not personalised; (2) Comparatively static association rules defined in the firewalls, or gateways; (3) Cannot identify the extremely hidden information that mixed in the syntax or semantics. To overcome these problems, we develop and implement a new email spamming system leveraged by coupled text similarity analysis on user preference and a virtual meta-layer user-based email network, we take the social networks or campus LANs as the spam social network scenario. Fewer current practices exploit social networking initiatives to assist in spam filtering. Social network has essentially a large number of accounts features to be considered.

We construct a new model called meta-layer email network which can reduce these features by only considering individual user’s actions i.e., replying network, reading network and deleting network. For the first time, these common user actions are considered to construct a social behavior-based email network. Further, a coupled selection model is developed for this email network, we are able to consider all relevant factors/features in a whole and recommend the emails practically to the user individually. The experiment data comes from the Enron email dataset, which has been recognized as a representative dataset for testing and validation. The experimental results show the new approach can achieve higher precision and accuracy with better email ranking in favor of personalised preference.