Deep&Cross network for CTR prediction

[WuLC]

This paper is from Google in 2017, similar to Wide&Deep, this network comprises of two parts: cross part and deep part. Deep network is normal deep neutral network, and cross network is for feature engineering. Compared with PNN, cross network is more efficient in learning certain bounded-degree feature interactions.

According to the paper, this network is superior over the state-of-art algorithms on the CTR prediction dataset and dense classification dataset, in terms of both model accuracy and memory usage