Approach for Hierarchical Global All-In Classification with application of Convolutional Neural Networks

This paper describes the application of convolutional neural networks adapted for hierarchical text classification task. Even though CNN models already been shown to be efficient for text classification, but not really previously explored in the context of hierarchy. Therefore, more detailed evaluation of experiments with CNN models were required. Our conducted experiments are compared with already existing multiple strategies that use Linear Regression and Support Vector Machines. The source of training data set is a collection of top 20 News Group data. We are curious to learn that our proposed methods achieve better results than existing state of art solutions. Furthermore, CNN hides the complexity of the hierarchical model and requires less resources for prediction. We find there are much more of unexplored space for improvements and optimizations of CNN application for hierarchical text classification.