“Facebook Reactions” as Emotional Indicators: A Multi-Method Approach to Analyzing User Engagement with COVID-19 News on Indian Media Platforms.

This thesis investigates the role of Facebook reactions as indicators of public sentiment and engagement with COVID-19 pandemic-related news in India during different stages of the pandemic, ranging from March 24, 2020, to March 31, 2022. This work employs a mixed-methods approach combining time-series analysis with embedding-based topic modeling, GPT-4-assisted cluster labeling, and lexicon-based sentiment analysis to examine a dataset of 68,319 Facebook posts and a focused subset of 8,622 Facebook posts analyzed for the early stages of the pandemic (March 24 - April 14, 2020) to capture initial public responses. This data is derived from four major English-language Indian news outlets: The Times of India, The Hindu, Indian Express, and Hindustan Times. Facebook reactions such as "Love," "Haha," "Wow," "Sad," and "Angry" serve as unique paralinguistic digital affordances that enable users to express their feelings and emotions toward online content and posts. This research explores how these Facebook reactions correlate with news posts and user engagement with news content shared during the COVID-19 pandemic. By employing time-series analysis, the study identified significant spikes in user engagement, particularly during the early pandemic period, with forty-eight "unusual days" characterized by significant variations in Facebook reactions. Through embedding-based topic modeling, the analysis revealed twenty-five distinct thematic clusters for COVID-19-related news coverage, ranging from technological innovations and lockdown enforcement to global political responses and philanthropic efforts. These distinct clusters demonstrated different patterns of user engagement, with predominantly positive reactions ("Love," "Haha") associated with clusters focusing on community support and celebrity engagement, while negative reactions ("Angry," "Sad") were linked to clusters related to crisis impacts and enforcement actions. Additionally, these patterns revealed systematic variations in reactions across different news outlets. The analysis revealed that The Times of India achieved the highest engagement overall but showed negative sentiment scores, while The Hindu revealed the most positive sentiment scores despite lower user engagement. This analysis observed a moderate positive correlation (r=0.37) between sentiment and user reactions, indicating that the emotional tone in news posts influenced audience responses. Moreover, the study identified the exploitation of emotional engagement, commonly referred to as "rage-baiting" content, particularly around communal politics and posts related to China during the COVID-19 pandemic period. The findings of this study contribute to enhancing understanding of crisis communication in the digital era by illustrating how Facebook reactions can capture nuanced public emotional responses beyond traditional engagement metrics such as likes, comments, and shares. The innovative use of GPT-4 for cluster labeling presents a novel approach for interpreting large-scale social media datasets. The findings offer practical insights and guidance for news media organizations and policymakers to develop strategies for more effective crisis communication during public health emergencies. Keywords: Facebook reactions, COVID-19 pandemic, sentiment analysis, social media engagement, Indian news media, topic modeling, crisis communication, digital emotional indicators.