基于xDeepFM模型的个性化智能推荐策略OA

The extended deep factorization machine(xDeepFM)model with an explicit feature crossing mechanism is applied to construct a personalized recommendation strategy for precise delivery of police-related online public opinion announcements.User attribute features and event features are combined with text semantic features extracted from user-generated content using the RoBERTa pre-trained language model.A Universal Transformer is employed to fuse features from multiple sources,followed by training and optimization of the recommendation model.Two cases,the"Hubei police car forcing a car to stop on the expressway"and the"Chongqing'Fat Cat'river incident,"are used for empirical evalua-tion.The model achieves push accuracies of 91.64%and 90.12%,demonstrating strong performance and generalization ability.Taking the Hubei case as an example,differential analysis of audience groups by age,authentication status,and regional distribution identifies distinct information acceptance preferences and response patterns,providing technical support and strategy references for personalized content delivery by public security authorities.