An Ensemble Neural Network for the Emotional Classification of Text

In this work, we propose a novel ensemble neural network design that is capable of classifying the emotional context of short sentences. Our model consists of three distinct branches, each of which is composed of a combination of recurrent, convolutional, and pooling layers to capture the emotional context of text. Our unique combination of convolutional and recurrent layers enables our network to extract more emotionally salient information from text than formerly possible. Using this network, experiments classifying the emotional context of short sequences of texts from five distinct datasets, were conducted. Results show that the novel method outperforms all historical approaches across all datasets by 8.31 percentage points on average. Additionally, the proposed work produces results that are on average as accurate as state of the art methods, while using two orders of magnitude less training data. The contribution of this paper is a novel ensemble recurrent convolutional neural network capable of detecting and classifying the emotional context of short texts.