A CNN Model for SMS Spam Detection

A convolutional neural network (CNN) model is proposed for detection of Chinese short message service (SMS) spams. The Chinese Wikipedia corpus is used to train the word2vec model. On the basis of the training result, we conduct word segmentation on the sample short messages, convert the words into word vectors, and construct word-vector matrices of sentences as the input of the convolutional neural network. In the convolutional neural network, we first obtain a convolutional layer by performing convolution on the input layer, and then obtain a pooling layer by pooling (downsampling) the convolution layer; at last, we concatenate all pooling features on the pooling layer into a long vector which is used as the input of the Softmax classifier to detect SMS spams. This study also discusses the influence of hyperparameters on the CNN model and identifies the optimal combination of hyperparameters.