Considering the current development trends within the sphere of digital media, personalization of recommendation services is one of the key challenges faced by modern recommender systems. Indeed, existing recommendation frameworks rely mostly on static user profiling and history-based analysis of their previous interactions, thus failing to adequately reflect the dynamics of emotional state changes and evolving preferences of the target audience. In order to solve the problem mentioned above, an innovative approach to short video content recommendation based on Emotion Driven Manifold Router framework is proposed in this paper. The Emotion Driven Manifold Router consists of three main components including Counterfactual Manifold Optimizer used to optimize content mapping based on simulated counterfactuals regarding user interactions, Agent Based Emotion Segmentation used to create dynamic clusters of users based on inferred emotion information, and finally Probabilistic User Perception Model used to estimate user preferences in terms of probabilities derived from emotion-behavior data. These three components operate through the mechanism of Policy Driven Coordination based on stochastic refinement of the entire process. As a result, the developed system proved to have significantly higher efficiency with an average increase of 23% in terms of interaction rates and a 17% growth of user satisfaction.
