The paper "xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems" by Jianxun Lian et al. addresses the challenge of learning high-order feature interactions in recommender systems, particularly for web-scale systems where manual feature engineering is costly and inefficient. The authors propose a novel model called eXtreme Deep Factorization Machine (xDeepFM), which combines a Compressed Interaction Network (CIN) and a classical deep neural network (DNN). CIN is designed to learn explicit high-order feature interactions at the vector-wise level, while DNNs capture implicit interactions. This combination allows xDeepFM to effectively learn both types of interactions, reducing the need for manual feature engineering. The model is evaluated on three real-world datasets, demonstrating superior performance compared to state-of-the-art models. The paper also discusses the theoretical and practical aspects of CIN, including its space and time complexity, and provides insights into the hyper-parameter settings that optimize the model's performance.The paper "xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems" by Jianxun Lian et al. addresses the challenge of learning high-order feature interactions in recommender systems, particularly for web-scale systems where manual feature engineering is costly and inefficient. The authors propose a novel model called eXtreme Deep Factorization Machine (xDeepFM), which combines a Compressed Interaction Network (CIN) and a classical deep neural network (DNN). CIN is designed to learn explicit high-order feature interactions at the vector-wise level, while DNNs capture implicit interactions. This combination allows xDeepFM to effectively learn both types of interactions, reducing the need for manual feature engineering. The model is evaluated on three real-world datasets, demonstrating superior performance compared to state-of-the-art models. The paper also discusses the theoretical and practical aspects of CIN, including its space and time complexity, and provides insights into the hyper-parameter settings that optimize the model's performance.