Text Classification Using Scope Based Convolutional Neural Network
Jiaying Wang, Jing Shan, Yaxin Li, Yi Cheng Ren, Duo Zhai, Jinling Bao · 2019
Text classification is a classic and important task in Natural Language Processing (NLP). Convolutional Neural Network (CNN), as an important model for deep learning, has shown its advantages in the field. For a CNN model, the convolution operation is the key to success. Based on convolution operation, it can effectively capture the local conjunctions of features from high dimensional features. Most of existing approaches utilize a sliding window to conduct convolution operation. Different from existing approaches, we present a scope based convolutional neural network (SCNN) in this paper. Instead of using sliding windows, we propose a concept of scope to capture the local information. Different from a window, a scope only imposes constraint on distance between words. It is more flexible and can effectively handle complicated local feature. We utilize max pooling to keep the most valuable information. In this way, it can capture deeper local information which is hidden from windows based CNN. We conduct extensive experimental study to demonstrate the competitive performance of our model against state-of-the-art approach on real datasets. The result shows the effectiveness of our proposed method.