DeepFM is a neural network model that combines factorization machines (FM) and deep learning to predict click-through rate (CTR) in recommender systems. It addresses the limitations of existing models that focus on either low- or high-order feature interactions or require expert feature engineering. DeepFM learns both low- and high-order feature interactions end-to-end without requiring feature engineering beyond raw features. The model consists of two components: an FM component that models low-order interactions and a deep component that models high-order interactions. Both components share the same input and embedding vector, enabling efficient learning. DeepFM outperforms existing models like Wide & Deep, FNN, and PNN in terms of AUC and Logloss on benchmark and commercial datasets. It is efficient and effective for CTR prediction, with results showing significant improvements over state-of-the-art models. DeepFM is trained end-to-end without pre-training, and it shares feature embeddings between the FM and deep components, which enhances performance. The model is evaluated on two datasets: Criteo and Company*, demonstrating its effectiveness and efficiency. DeepFM's performance is further validated through hyperparameter studies, showing that it outperforms other models in terms of AUC and Logloss. The model is also compared with other deep learning approaches for CTR prediction, highlighting its advantages in capturing both low- and high-order feature interactions. Overall, DeepFM provides a more effective and efficient solution for CTR prediction in recommender systems.DeepFM is a neural network model that combines factorization machines (FM) and deep learning to predict click-through rate (CTR) in recommender systems. It addresses the limitations of existing models that focus on either low- or high-order feature interactions or require expert feature engineering. DeepFM learns both low- and high-order feature interactions end-to-end without requiring feature engineering beyond raw features. The model consists of two components: an FM component that models low-order interactions and a deep component that models high-order interactions. Both components share the same input and embedding vector, enabling efficient learning. DeepFM outperforms existing models like Wide & Deep, FNN, and PNN in terms of AUC and Logloss on benchmark and commercial datasets. It is efficient and effective for CTR prediction, with results showing significant improvements over state-of-the-art models. DeepFM is trained end-to-end without pre-training, and it shares feature embeddings between the FM and deep components, which enhances performance. The model is evaluated on two datasets: Criteo and Company*, demonstrating its effectiveness and efficiency. DeepFM's performance is further validated through hyperparameter studies, showing that it outperforms other models in terms of AUC and Logloss. The model is also compared with other deep learning approaches for CTR prediction, highlighting its advantages in capturing both low- and high-order feature interactions. Overall, DeepFM provides a more effective and efficient solution for CTR prediction in recommender systems.