A novel deep learning architecture for sentiment classification

Abstract

Evolution of plethora of e-commerce sites resulted in fierce competition among their providers. In order to acquire new and retain existing customers, various producers and market managers effectively employ online feedback analytics tools. Most of the online feedback analysis tools are built using sentiment analysis models. Sentiment analysis evolved in the last one and half decades for review mining process. An important sub-task of sentiment analysis called sentiment classification is used mainly to decide whether a written review is expressing either positive or negative sentiment towards a target entity. In order to have better sentiment classification accuracy, we proposed a hybrid deep learning architecture, which is a hybrid of a two layered Restricted Boltzmann Machine and a Probabilistic Neural Network. The proposed approach yielded better accuracy for five different datasets compared to the state-of-the-art.