With the rapid advancement of the internet, social media has become an integral part of human life. It provides a soothing platform for people to share their opinions. Twitter became the most popular social platform during the COVID-19 pandemic to connect people internationally and support the fight against the infectious epidemic. Nevertheless, the increasing use of social media generates a massive amount of meaningless information, which researchers need to refine to interpret the material. Recently, investigators have proposed various approaches to gain insights from unstructured data on social platforms. Still, researchers face innumerable challenges in dealing with vague content. Meticulous polarity detection of people's opinions is an exhilarating and ongoing problem. Therefore, a hybrid lexicon-based dictionary and a CNN-LSTM-based approach have been proposed to analyze tweets related to COVID-19 vaccines. This model is enriched with enhanced feature transformations via the attention mechanism and an extensive lexicon. However, the lexicon approach is very domain-specific; on the other hand, neural networks are less interpretable and dependent on content quality. Consequently, the proposed hybrid model combines the strengths of lexicon-based dictionaries and neural networks while mitigating their limitations. The experimental results demonstrated the outperformance of the proposed hybrid model, achieving 83.44% accuracy, 88% F1-score, 91% recall, and 86% precision on COVID-19 tweets. It indicates that the proposed hybrid solution is robust and practical for addressing similar problems and other epidemic situations.
Tiwari et al. (Tue,) studied this question.