ABSTRACT
The pandemic that arose due to the novel Corona Virus Disease of 2019 (COVID) has become the biggest challenge of all time. The entire world’s population has stormed social media to express their opinions, emotions, and sentiments. This manuscript implements classical machine and deep learning approaches with static and stacked word embeddings to identify the sentiments of the COVID-19 tweets extracted from Twitter. The problem we have tackled in this manuscript is the multi-class classification problem for three and five classes, respectively. Our proposed deep learning model with stacked word embeddings has outperformed the individual static pre-trained embeddings representation, classical machine, and deep learning approaches altogether. The proposed model has proven useful in complex classification tasks such as identifying classes belonging to the same group of sentiments namely Extremely Negative and Negative, Extremely Positive and Positive. The experimental results also show the superior performance of stacked word embeddings for the peculiar contextual semantic comprehension from small tweets and dealing with the unbalancedness of the experimental dataset. We achieved the accuracy with stacked embeddings with accuracy being 73.01% and 84.25% for three and five classes, respectively. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.