Bimodal longitudinal investigation on changes in sentiments over social media interactions owing to covid-19 pandemic

Abstract

Changes in human sentiments over cyberspace with the emergence of a pandemic were unheard of before COVID-19, given that the last recorded pandemic occurred decades before interactive cyberspace existed. Accordingly, the opportunities and dimensionalities pertinent to the human sentiments and their changes that surfaced over social media interactions demand an in-depth analysis, which we perform in this study. Existing related research studies found to date focus on analyzing sentiments covering only text-based social media posts with limited contexts. These studies generally use out-of-the-box libraries to classify sentiments, which are often not suited for social media posts as they possess different styles compared to regular text bodies. To go beyond these studies, first, we collect public thoughts and images shared on Twitter by those who showed their interest in COVID-19. We then explore the existing sentiment classifier libraries and their potential blend for developing a new classification technique to better analyze sentiments over text-based tweets. Afterward, we perform exploratory data analysis on these collected thoughts and images to find the patterns inherently embedded within these changes of sentiments, owing to the COVID-19 outbreak, expressed over social media. Our findings through a bimodal investigation subsuming both text and images reveal a correlation between the two modalities of expression, pointing to changes in sentiments over two years spanning pre- and during COVID phases, identifying change-points for each type of sentiment during the different phases, etc. These findings unveil new dimensionalities of human interactions over cyberspace during a pandemic period.