The demand for personalized recommendation systems growing rapidly during the recent years especially in entertainment and e-commerce fields. These recommendation systems face specific data consistency problems particularly when dealing with specialized groups such as young audiences. The proposed research develops an anime recommendation platform for children featuring a hybrid Deep Factorization Machines (DeepFM) and negative sampling structure which resolves data sparsity problems. The system creates a better dataset for model learning through negative sampling techniques alongside child-friendly genre dataset filtering. The implementation of DeepFM models incorporates Factorization Machines (FM) which detect low-order feature combinations with deep neural networks (DNN) for understanding high-order non-linear patterns. Anti-overfitting measures and generalization improvements are achieved through Regularization layers together with those based on Dropout. In this paper three metrics including precision, recall and Root Mean Squared Error (RMSE) are used for simulation of the model, that validate the proposed model increases recommendation accuracy while substantially decreasing user-item interactivity matrix sparsity levels. The model’s ability to improve user coverage emerges through visual inspections of sparsity reduction which demonstrates its efficiency in generating specific recommendations for children. The research shows that DeepFM coupled with negative sampling represents a strong method for addressing data sparsity problems in recommendation systems because it has demonstrated potential use in additional recommendation areas.