Spam filtering in SMS using recurrent neural networks

Abstract

Short Message Service (SMS) is one of the mobile communication services that allows easy and inexpensive communication. Producing unwanted messages with the aim of advertising or harassment and sending these messages on SMS have become the biggest challenge in this service. Various methods have been presented to detect unsolicited short messages; many of which are based on machine learning. Neural Networks have been applied to separate the unwanted text messages (known as spam) from normal short messages (known as ham) in SMS. To the best of our knowledge, Recurrent Neural Network (RNN) has not been used in this issue yet. In this paper, we proposed a new method which utilizes RNN to separate the ham and spam with variable length sequences; even though we used a fixed sequence length. The proposed method achieved an accuracy of 98.11, indicates a considerable improvement compared to Support Vector Machine (SVM), token-based SVM and Bayesian algorithms with accuracies of 97.81, 97.64, and 80.54, respectively.