Filtering Spam by Using Factors Hyperbolic Trees

Most of current anti-spam techniques, like the Bayesian anti-spam algorithm, primarily use lexical matching for filtering unsolicited bulk E-mails (UBE) and unsolicited commercial E-mails (UCE). However, precision of spam filtering is usually low when the lexical matching algorithms are used in real dynamic environments. For example, an E-mail of refrigerator advertisements is useful for most families, but it is useless for Eskimos. The lexical matching anti-spam algorithms cannot distinguish such processed E-mails that are junk to most people but are useful for others. We propose a Factors Hyperbolic Tree (FHT) based algorithm that, unlike the lexical matching algorithms, handles spam filtering in a dynamic environment by considering various relevant factors. The new Ranked Term Frequency (RTF) algorithm is proposed to extract indicators from E-mails that are related to environmental factors. Type-1 and Type-2 fuzzy logic systems are used to evaluate the indicators and determine whether E-mails are spam based on the environmental factors. Additionally, weights of factors in a FHT database are continuously updated according to dynamic conditional factors in a real environment. Simulation results show that the FHT algorithm filters out spam with high precision. Furthermore, the FHT algorithm is more efficient than other methods when it filters E-mails with complex influencing factors. The main contribution of this paper is that the FHT based algorithm can filter E-mails based on influencing factors instead of matched words to allow dynamic filtering of spam E-mails.