Emotional Analysis and Strategy Optimization of Live Streaming E-Commerce Users Under the Framework of Causal Inference

Abstract

Current emotion recognition models largely rely on similarity-based approaches, which limits their ability to explain the underlying causes of emotional changes and reduces the reliability of subsequent strategy adjustments. To address these limitations, this study proposes a unified framework that integrates multiple forms of emotion modeling with dynamic strategy regulation. Within this framework, novel methods such as reverse reasoning and causal embedding are introduced within graph neural networks to explicitly capture the relationships between intervention variables and emotional states. By incorporating causal control attention mechanisms and the NOTEARS algorithm, the framework enhances the fusion of heterogeneous information and improves the discrimination between critical influencing factors. Furthermore, a causal evaluation and cyclic regulation mechanism is constructed to enable continuous assessment and adjustment of emotional interventions. This comprehensive approach not only provides a robust computational foundation for emotion modeling but also offers practical guidance for developing intelligent, controllable, and adaptive emotion management systems. The proposed framework demonstrates potential for applications in personalized mental health support, human-computer interaction, and affective computing systems, where understanding the causal dynamics of emotion is essential for effective intervention.