The paper introduces the Wide & Deep learning framework, which combines the strengths of both wide linear models and deep neural networks to enhance recommendation systems. Wide linear models use cross-product feature transformations to memorize frequent co-occurrences of features, while deep neural networks learn low-dimensional dense embeddings to generalize to unseen feature combinations. The framework is designed to achieve both memorization and generalization in a single model, making it suitable for large-scale sparse input problems like app recommendations in the Google Play store. The authors evaluate the Wide & Deep model on Google Play, a commercial mobile app store with over one billion active users and one million apps. Online experiments show that the Wide & Deep model significantly improves app acquisition rates compared to wide-only and deep-only models. The implementation of the Wide & Deep model is open-sourced in TensorFlow, and the system is optimized for high throughput and low latency, serving over 10 million apps per second with a latency of 14 ms. The paper also discusses related work in machine learning and recommendation systems, highlighting the importance of combining linear models and deep neural networks to address the challenges of sparse and high-dimensional data in recommendation tasks.The paper introduces the Wide & Deep learning framework, which combines the strengths of both wide linear models and deep neural networks to enhance recommendation systems. Wide linear models use cross-product feature transformations to memorize frequent co-occurrences of features, while deep neural networks learn low-dimensional dense embeddings to generalize to unseen feature combinations. The framework is designed to achieve both memorization and generalization in a single model, making it suitable for large-scale sparse input problems like app recommendations in the Google Play store. The authors evaluate the Wide & Deep model on Google Play, a commercial mobile app store with over one billion active users and one million apps. Online experiments show that the Wide & Deep model significantly improves app acquisition rates compared to wide-only and deep-only models. The implementation of the Wide & Deep model is open-sourced in TensorFlow, and the system is optimized for high throughput and low latency, serving over 10 million apps per second with a latency of 14 ms. The paper also discusses related work in machine learning and recommendation systems, highlighting the importance of combining linear models and deep neural networks to address the challenges of sparse and high-dimensional data in recommendation tasks.