 This paper proposes a new hybrid recommendation algorithm called Light FM Deep Neural Network, LFDNN. It combines four modules, Light GBM, Shallow Model, Deep Model, and Fusion Module, to improve the accuracy of recommendation systems. The Light GBM module uses gradient boosting decision trees to process dense numerical features, while the Shallow Model and Deep Model use a Shallow and Deep Neural Networks respectively to capture higher-order feature interactions. Finally, the Fusion module combines the outputs from these three modules to produce a final recommendation result. Experiments show that LFDNN outperforms other state-of-the-art recommendation algorithms on two real-world datasets. This article was authored by Ho-choh-han, Yan Chunliang, Gabor Bella, and others.