A(RCNN) Neural Architecture for Document Classification

Frank Ferraro, Department of Computer Science and Electrical Engineering

Graduate Student: Sarthak Mehta, Department of Computer Science and Electrical Engineering

We propose a neural architecture for the document classification problem where we utilize a hybrid neural structure which is a three-layered structure – first, a character level convolutional neural layer, next layer is an attention layer to get word level representation from characters, the third layer is a recurrent neural network structure/(again convolutional neural network) in order to get a get sentence level representation and finally we have interconnected dense structure stacked to classify the document.