A Deep Learning Strategy to Sentiment Analysis on Low-volume Dataset

Abstract

The potency of Deep Learning techniques builds on the amount of labeled data accessible for training Deep Neural model. The lack of training data becomes a common barrier for using Deep Learning technique to perform different NLP task. Transfer learning or Domain adaptation is a required technique to settle the deficiency of categorized reviews by studying a model's ability of transferring knowledge across domains to perform Sentiment Classification. In this paper, we have suggested a transfer learning approach stand on a combination of Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) model, designated as C_LSTM. This proposed model is used to resolve the issue of data insufficiency in a particular domain by introducing a large-scale supplementary dataset from more or less similar domains for knowledge transferring. These experimental results of intra-domain and inter-domain Sentiment Classification on multi-domain product review indicate that utilizing a high-volume corpus on related domain can remarkably boost the accuracy.