A New Click-Through Rates Prediction Model Based on Deep&Cross Network

Abstract

With the development of E-commerce, online advertising began to thrive and has gradually developed into a new mode of business, of which Click-Through Rates (CTR) prediction is the essential driving technology. Given a user, commodities and scenarios, the CTR model can predict the user’s click probability of an online advertisement. Recently, great progress has been made with the introduction of Deep Neural Networks (DNN) into CTR. In order to further advance the DNN-based CTR prediction models, this paper introduces a new model of FO-FTRL-DCN, based on the prestigious model of Deep&Cross Network (DCN) augmented with the latest optimization technique of Follow The Regularized Leader (FTRL) for DNN. The extensive comparative experiments on the iPinYou datasets show that the proposed model has outperformed other state-of-the-art baselines, with better generalization across different datasets in the benchmark.

Click-through rate prediction in online advertising: A literature review

“…Moreover, LR can be combined with other methods to predict clicks, e.g., FMSDAELR (Jiang et al, 2018), FO-FTRL-DCN (Huang et al, 2020) and GBDT+LR (He et al, 2014), where it forms the top layer after feature selection and feature interactions. We will discuss in detail FMSDAELR and FO-FTRL-DCN in Section 4.3 and GBDT+LR in Section 4.4.…”

“…In order to represent both low-order and high-order feature interactions in a unified framework, it is of necessity to combine DNN with low-order modeling frameworks discussed in Sections 4.1 and 4.2. A notable effort in this direction is DeepFM (Guo et al, 2017; Logloss (Huang et al, 2020).…”

“…Moreover, we can observe that PNN is superior to FNN. Specifically, among 14 studies conducted on 4 public datasets and 1 proprietary dataset, most reported that PNN performs better than FNN in AUC, Logloss, RMSE and RIG, with a few exceptions (Guo et al, 2018;Yan et al, 2020;Huang et al, 2020). However, the literature provides controversial results on performance comparison between DeepFM and PNN (or its extensions).…”

“…In this research branch, researchers have developed a few CTR prediction models similar to the modeling structure of DeepFM, e.g., Wide & Deep(Cheng et al, 2016), Wide & ResNet, attention stacked autoencoder (ASAE), eXtreme Deep Factorization Machine (xDeepFM)(Lian et al, 2018), hybrid feature fusion (HFF) and FO-FTRL-DCN(Huang et al, 2020). Wide & Deep is an interesting attempt for the sparse problem, which joins LR with a feed-forward neural network; however, it depends on business knowledge as LR requires manual feature engineering.…”

A CTR prediction model based on session interest

“…The model leverages the strength of CNN to generate local patterns and recombine them to generate new features. Huang et al [ 24 ] introduced a new model based on Deep&Cross Network [ 25 ], the model can get better feature interaction. Cross Network [ 26 ] further replaces the cross vector in Cross Network into a cross matrix to make it more expressive.…”