Abstract
As generative artificial intelligence (AI) technologies become increasingly present in creative and professional domains, examining public discourse surrounding these tools is important for understanding their broader social implications. This study conducts a two-part analysis of the initial public reaction to Sora, the generative video model developed by OpenAI, by analyzing 23,543 English-language comments posted on YouTube between February and April 2024. Rather than relying on traditional positive–negative sentiment classifications, this study integrates fine-grained emotion detection with topic modeling to examine the relationship between emotions and topics in the discourse. Based on the residual analysis, the overall association between topics and emotions was weak; however, certain topics were associated with specific emotions. For instance, ethical discussions were more likely to be associated with sadness and anger, artistic settings were associated with fear, and benchmark discussions were associated with joy. Methodologically, this study utilizes an emotion–topic coupling through residual deviation with a hierarchical LDA-BERTopic approach, bringing together computational modeling and theories of emotion. This study provides a structured and theory-based way to explore the affective and thematic patterns in the public’s discourse surrounding Sora.