Study of spam short message filtering based on features selection of key words

Abstract

In spam SMS (Short Message Service) filtering system, key words frequencies are often used to measure the weights of key words. However, the behavior of this measurement is not very well. Therefore, we select the mutual information between the key word and the category, the length of it and the frequency of it as features of key words and figure out corresponding formula to measure the weights of key words. Our method is applied to our filtering system based on the Naïve Bayes algorithm which is also improved by the Lidstone algorithm to solve the unseen feature words problem. The results of our experiment based on the dataset built by ourselves show that the comprehensive evaluation index of our improved algorithm demonstrated a 19.61% increase in overall rating, compared to the filtering system using key words frequencies as features. © 2012 Springer-Verlag.