The high‐level risk perception diffusion caused by public health emergencies seriously threatens public mental health and social stability. Much scholarly attention focused on the traditional epidemic models or simply combined content and social attributes, overlooking the differences in public individual characteristics. This paper proposes an S1S2EIposIneuInegR model of risk perception diffusion by innovatively subdividing susceptible people and infectious people. Then, taking the Xi'an epidemic as an example (N = 105,417), this paper employs the sentiment analysis model of Word2Vec and Bi‐LSTM to calculate the emotional value of microblog text to quantify public risk perception. Finally, numerical experiments are conducted to explore the effects of cross‐evolution and emotional difference on risk perception diffusion under different scenarios. Findings reveal that a larger initial density of infectious people accelerates diffusion, with negative emotions playing a dominant role. In addition, the higher the risk perception level and the lower the heterogeneity, the greater the maximum impact and the final scale of diffusion. When the public health emergency deteriorates, the cross‐evolution tends to shift to a high‐risk perception. Otherwise, it tends to tilt to a low‐risk perception. These findings provide critical insights for developing precise risk perception guidance strategies and enhancing public health governance capabilities.