Application of Convexified Convolutional Neural Network in Text Classification

This paper introduces a type of convexified convolutional neural network (CCNN), introduced by Zhang, Liang and Wainwright. We applied this model on the classification of text-based online shopping reviews. This work makes an estimate on the error term brought by the low rank approximation. We also build our codes on the work done by Schaik. We make adjustments on the choices of kernel functions and further extend the application of the algorithm to multilayer CCNNs. The results show that Zhang’s model is practical on learning shallow CCNNs. However, there is no big improvement in multilayer CCNN. Analysis of likely causes are mentioned by the end of this section. Strengths, weaknesses as well as future work are discussed in the end.