Abstract
Addressing the risk of uncontrolled dissemination of AI deepfake videos in entertainment scenarios, this study constructs an explainable ensemble learning prediction framework from an entertainment computing perspective, systematically revealing the diffusion mechanisms of technology-enabled entertainment content. Guided by information ecosystem theory, the study first identifies nine core factors influencing deepfake video propagation through multidimensional feature decomposition. It innovatively proposes the RFECV-GA-PSO-RF hybrid feature selection algorithm to achieve efficient dimensionality reduction of entertainment computing features. Subsequently, the study employs a PSO-GA-XGBOOST ensemble model-fusing particle swarm optimization (PSO) and genetic algorithm (GA)-to achieve precise predictions of deepfake video propagation on real-world Chinese video platforms. This approach significantly outperforms existing models, demonstrating average improvements of 42.95% across four evaluation metrics (RMSE reduced to 1.230, MAPE reduced to 0.280, MAE reduced to 1.063, R² reaching 0.818). Finally, leveraging the interpretability of this predictive model, the study quantifies the importance of each feature and feature dimension. The proposed integrated prediction model not only provides novel predictive tools for the field of entertainment computing but also offers quantitative decision support for dissemination regulation and content ecosystem optimization in the era of intelligent entertainment, expanding the theoretical boundaries of interdisciplinary research in entertainment technology.