Identification of Spam Based on Dependency Syntax and Convolutional Neural Network

Abstract

Convolution Neural Network (CNN) is an algorithm which is more suitable for classify in images and natural language recognition. For Chinese spam processing identify, this paper proposed a hybrid DDTV-CNN model for short text classification that combines deep dependency trait vectorization (DDTV) with convolutional neural network. Parse the semantics of short texts by dependency parsing, we can get a binary tree, and construct a matrix through arc in a binary tree; then, nonlinear decomposing the matrix to get the eigenvector representation of semantic; finally, divide it into two categories by convolutional neural network. This article uses the performance evaluation index commonly used in the field of text classification and information retrieval to establish a evaluation system of spam identification. The evaluation system is used to evaluate the experimental data obtained from simulation experiments, and use performance evaluation index to evaluate that often used in text classification and domain of information retrieval, we construct evaluation system through it about spam identification; and then use it to evaluate experimental data that acquire from simulation experiment, and choice appropriate kernel functions and its parameters. Through the experiment contrasts, the classifier based on DDTV-CNN is more effective and rapid than traditional.