A study on text classification based on stacked contractive auto-encoder
YangZhen Yu, Jing Hui · 2017
In order to improve the classification effect of text classification and reduce the error rate of classification, this paper proposes the thought of text classification based on Stacked Contractive Auto-Encoder (SCAE) by stacking network layer by layer, the SCAE constitutes the depth of the neural network by unsupervised training and learning text to improve the robustness of feature extraction, the network uses the gradient descent algorithm to optimize parameters of the network. This paper adopts an improved TFIDF method calculating the weight of feature words. Though experiments, the CAE and SAE (Sparse Auto-Encoder) are compared, the support vector machine (SVM) is used to classify text. Experiment result shows that the classification performance of single layer CAE is better than that of single layer SAE, and the classification performance of SCAE is better than that of SSAE (Stacked Sparse Auto-Encoder).