This study explores the moral coping strategies embedded in Korean Twitter discourse during the COVID-19 pandemic by employing deep learning methods grounded in Moral Foundations Theory (MFT). While much prior research has examined public values, ethical priorities, and emotional responses during the pandemic, comparatively little attention has been given to how moral categories are expressed dynamically in non-Western contexts. To address this gap, we constructed a Korean version of the Moral Foundations Dictionary (K-MFD), tailored to cultural and linguistic nuances, and applied it to more than 560,000 annotated COVID-19-related tweets collected between December 2019 and June 2020. Using Long Short-Term Memory (LSTM) neural networks, we conducted large-scale context-sensitive analyses to classify moral judgments and track their fluctuations over time.

Our findings reveal a dual moral strategy: individualizing moral foundations such as harm and cheating emerged as stable and long-term concerns, while binding foundations such as loyalty and degradation fluctuated in response to pandemic-specific events. Time-series analyses show that collectivist-oriented categories gained salience during moments of acute crisis—such as rising infection numbers or mask shortages—but receded as resources stabilized. In contrast, individualizing categories remained consistently central, suggesting that Korea’s high rate of mask-wearing and compliance may be explained less by collectivist loyalty and more by individualist motivations to avoid harm and prevent cheating. This interpretation challenges earlier studies that attributed East Asian compliance primarily to collectivist values.

Morpheme-level analysis further demonstrates how particular terms were strongly tied to moral categories: “confirmed cases” and “fake news” were salient within degradation, whereas “nation” and “integration” clustered around loyalty. Sentiment analysis revealed that moral expressions were overall more positive (57.1%) than negative (42.9%), with peaks of both occurring in March 2020 during the first major outbreak and subsequent stabilization of the mask supply. Such temporal variation underscores the importance of contextualizing moral attitudes within specific phases of a crisis.

This study contributes methodologically by combining a culturally adapted moral dictionary with deep learning, enabling a nuanced assessment of moral judgments beyond keyword counts. Substantively, it highlights the coexistence of individualist and collectivist orientations in Korean moral discourse, which shift according to situational demands rather than remaining fixed. These findings offer valuable insights for public health communication, policymaking, and disaster management by demonstrating that appeals to both individual and collective moral frameworks may be necessary depending on the stage and character of a crisis. More broadly, this work illustrates the significance of incorporating cultural specificity and temporal sensitivity into the computational study of moral psychology in digital environments