Sentiment Analysis using Deep Learning: A Domain Independent Approach

Abstract

The practice of finding emotion embedded in textual data is known as sentiment analysis, sometimes known as opinion mining. Various sentiment analysis algorithms, including classic Machine Learning models and Deep Learning models, have been suggested up until now. Some Machine Learning-based models, such as Naive Bayes, Decision Tree, SVM, and others, have demonstrated exceptional performance in sentiment categorization. Although Machine Learning algorithms have demonstrated high performance, they are constrained by the quantity of the dataset employed and include feature extraction tasks, which are time demanding. As a result, this study considers Deep Learning (DL)-based models, which include automated feature extraction and can handle massive amounts of data. One of the major issues with existing sentiment analysis models is that they are domain-dependent; hence, if there is a dataset available from a domain on which the model was not trained on, its accuracy is significantly reduced. To make the model domain agnostic, it is trained on datasets from three distinct domains: Twitter US Airline Review dataset, the IMDb Movie Review dataset, and the US Presidential Election dataset. The suggested sentiment analysis model is trained on five different deep learning models: CNN-GRU, CNN-LSTM, CNN, LSTM and GRU. The model's performance was evaluated using test data from three datasets on which the model was trained, as well as a fresh book review dataset scraped from the Amazon website.