Sentiment Analysis of COVID19 Reviews Using Hierarchical Version of d-RNN

Abstract

In recent years understanding person's sentiments for catastrophic events has been a major subject of research. In recent times COVID19 has raised psychological issues in people's minds across world. Sentiment analysis has played significant role in analysing reviews across wide array of real-life situations. With constant development of deep learning based language models, this has become an active investigation area. With COVID19 pandemic different countries have faced several peaks resulting in lockdowns. During this time people have placed their sentiments in social media. As review data corpora grows it becomes necessary to develop robust sentiment analysis models capable of extracting people's viewpoints and sentiments. In this paper, we present a computational framework which uses deep learning based language models through delayed recurrent neural networks (d-RNN) and hierarchical version of d-RNN (Hd-RNN) for sentiment analysis catering to rise of COVID19 cases in different parts of India. Sentiments are reviewed considering time window spread across 2020 and 2021. Multi-label sentiment classification is used where more than one sentiment are expressed at once. Both d-RNN and Hd-RNN are optimized by fine tuning different network parameters and compared with BERT variants, LSTM as well as traditional methods. The methods are evaluated with highly skewed data as well as using precision, recall and F1 scores. The results on experimental datasets indicate superiority of Hd-RNN considering other techniques.