Implicit Feature Interactions for Recommender Systems Overview xDeepFM is a powerful recommender system model that aims to capture both explicit and implicit feature interactions in a unified manner. This model is designed to automatically learn patterns of combinatorial features, providing a robust solution for web-scale systems without the need for manual feature crafting.

Introduction In the realm of recommender systems, combinatorial features play a crucial role in achieving success. However, crafting these features manually can be costly due to the diverse, voluminous, and rapidly changing nature of raw data in web-scale systems. Factorization-based models, measuring interactions through vector products, offer an automated approach to learn patterns of combinatorial features and generalize to unseen features.

The xDeepFM model goes beyond traditional deep neural networks (DNNs) by introducing a novel Compressed Interaction Network (CIN). This network explicitly generates feature interactions at the vector-wise level, providing a unique perspective on interaction modeling. The CIN shares functionalities with convolutional neural networks (CNNs) and recurrent neural networks (RNNs). By combining the CIN with a classical DNN, the xDeepFM model achieves a balance between learning bounded-degree feature interactions explicitly and arbitrary low- and high-order feature interactions implicitly.