SMS Classification Methods Based On Deep Learning

With the mobile internet's rapid development, mobile phone text messages(SMS) have been used widely in various fields and become an indispensable communication means in people's daily life for its advantages of low price, easy to use and reliable delivery. However, extensive use of text messaging has been bringing risks in mobile information security such as the overflow of 'spam' messages. This leads to a problem waiting to be solved urgently. How to maintain the SMS healthy development while avoiding the undesirable influence of the ‘spam' messages. As a new field in machine learning, deep learning can automatically extracts the objective characteristics and represents them explicitly. It shows excellent performance when processing large scale data. In order to reduce the negative influence of the proliferation of the 'spam' messages, this thesis studied the text messages classification problem by incorporating the deep learning and text classification methods.Concerned with the current inaccurate SMS classification problem, this research proposed a novel method to classify the mobile text messages, which was based on the text classification methods and combine the deep learning algorithms. This method firstly pre-processed the mobile messages in the collecting date to select the characteristics of the 'spam' messages. Then a deep belief neural network was built by taking the messages vectors as the input neurons and treating the spam characteristics as the output neurons. Finally, the classification results were obtained by training the sampling messages with the built deep belief neural network.Experiments shows deep learning can learn the mobile messages without supervision and make classification of the SMS. Moreover the larger the data scale is, the better the method performs.