CrossSenti: A Zero Shot Cross Domain Sentiment Classification of Customer Reviews Using Deep Learning Frameworks

Abstract

Cross-domain text sentiment analysis (CDTSA) is the process of identifying the feeling emotion opinion attitude also called sentiment of the user comments or reviews written in text across multiple domains, such as reviews, social media posts, or news stories, when sentiment distribution and expression change dramatically between domains. Since conventional sentiment analysis models are usually trained and tested on data from a single domain. When sentiment analysis models are applied to domains other than those for which they were trained, their performance frequently deteriorates. In this paper, we propose a zero shot cross-domain sentiment analysis using a deep learning framework, where we trained the model on a source domain and evaluated in a distinct target domain. Our proposed method does not need to retrain the model for the target data. We demonstrate the global vector (GloVe) word embeddings method that quantifies the text data and combines with a convolutional neural network (CNN) and Multi-channel CNN-based models. We test the models on 12 pairs in cross-domain of amazon customer review datasets and perform an assessment on metric accuracy, precision, recall, and F1-score. And our proposed method outperforms the state-of-the-art results on cross-domain amazon review data sets. The experimental assessment of the cross-domain dataset indicates that the proposed strategy attains remarkable performance.