This paper constructs a unified computing framework driven by emotional interaction, focusing on user behavior modeling and persuasive design optimization of digital platforms in real-time interaction scenarios. Based on clickstream, dwell time, scroll trajectory, touch pressure, facial cues and text feedback, a multimodal data set consisting of 4320 users and 18,600 interaction sessions is established. The model describes emotional fluctuations and behavior changes through time convolution, cross-modal alignment, state aggregation and transition estimation, and discriminates four types of states: participation, hesitation, resistance and acceptance. On this basis, the adaptive intervention module dynamically adjusted the information expression, interface density, feedback timing and recommendation strength by combining the state results. The experimental results show that the state recognition accuracy of the proposed framework reaches 93.1%, and the intervention matching accuracy reaches 90.4%, so that the compliance interaction rate is increased by 14.8%, and the average response delay is controlled within 86 ms. The results show that the framework can maintain stable state understanding, policy generation and front-end execution capabilities in continuous interaction scenarios, and has practical application capabilities.
