In an era increasingly affected by natural and human-caused disasters, the critical role of social media in disaster communication has become ever more prominent. Despite substantial research on social media during crises, a critical gap remains in understanding information- switching patterns during compound hazard events. This understanding is essential for detecting information deviations and curbing the spread of misinformation. This study introduces an information-switching model to identify dynamic perspective shifts in user-mention networks formed around crisis messaging on social media. It utilizes advanced natural language processing, network science, and census data to analyze geotagged tweets related to compound disaster events in Oklahoma in 2022. The study identifies distinct engagement patterns among different user types across various disaster stages. These patterns reveal: (a) how different disaster types distinctly affect public sentiments, (b) increased vulnerability of mobile home communities during disasters, (c) significance of education and transportation access in crisis response, and (d) variable impacts of user types and network properties on sentiment dynamics. Demographic factors, particularly gender differences, also show significant variance in perspective shifts, with females exhibiting greater sentiment variability compared to males during crises. These findings offer new policy insights related to disaster risk communication and demonstrate how the temporal shifts in social media crisis narratives align with real-world response to major disasters. These insights are especially valuable for emergency responders and policymakers to understand public sentiment changes and the impact of misinformation during disasters.