Abstract: The rapid dissemination of information during public health crises, such as the COVID-19 pandemic, necessitates robust automated systems to monitor public sentiment. This research presents an exploratory study of various Bidirectional Gated Recurrent Unit (BiGRU) architectures specifically tailored for Thai-language sentiment classification. Using a balanced dataset of Thai news tweets, the study systematically investigated eight distinct BiGRU configurations—ranging from simple bidirectional layers to deep-stacked architectures—to identify optimal frameworks for crisis management.