Risk factor identification for large scale amusement facilities using mixture of experts and fusion of multiple models

The safety of large-scale amusement facilities is an important part of public safety. Their operational risks are sudden and diverse, and efficient and accurate text recognition and analysis methods are urgently needed to support risk warning and intelligent supervision. In view of the problems of low recognition efficiency and poor adaptability of traditional methods, this paper proposes a risk text recognition model of MoE + Stacking multi-perspective fusion, which combines semantic embedding and keyword feature construction, and integrates clustering, graph modeling and multi-model integration strategies to achieve structured modeling and accurate recognition of amusement facility risk information. The results show that the MoE + Stacking model is significantly better than a single model in multiple indicators. The accuracy, precision, recall and F1 value are increased by 3.44, 3.50, 3.33 and 2.90 percentage points respectively compared with the original Stacking model, and the LogLoss is reduced by 0.41%, showing a stronger risk identification ability. The research results provide effective technical support for public safety text analysis and intelligent supervision of special equipment, and have good generalization and promotion value.