Accurately modeling the spread of public opinion on social media requires capturing both the temporal dynamics and structural heterogeneity of user interactions. However, existing Susceptible-Infected-Recovered (SIR)-based approaches often overlook the hierarchical nature of interactions such as posting, commenting, and replying, as well as the evolving susceptibility of users. To address these limitations, we propose the Hierarchical Bayesian Logical Susceptibility-Infection-Recovery (HBLSIR) framework. This model integrates Bayesian inference and logical growth to dynamically estimate user susceptibility and simulates information diffusion across three interaction levels. It further incorporates emotion-conditioned propagation by instantiating emotion-specific SIR processes across different structural layers. Experiments on the VISTA dataset, covering 11 emotion categories, demonstrate the efficacy of HBLSIR, which outperforms SIR models in both cumulative and daily propagation simulations. These results highlight the effectiveness of hierarchical and emotion-aware modeling in capturing complex social media diffusion dynamics.