在线广告点击率预测方法的研究综述

A Review of Click-Through Rate Prediction Approaches for Online Advertising

在在线广告和推荐系统中,准确预测点击率(Click-Through Rate,CTR)是至关重要的。CTR是广告被点击次数与广告被展示次数的比值。过去,许多传统的机器学习算法,如逻辑回归、支持向量机,因为简单且易于实现而被广泛地应用于广告点击率预测工作。然而,这些传统算法往往需要复杂的特征工程。相较之下,深度学习模型能够有效自动提取高阶特征,可以较好地解决这一问题。此外,为了实现更高效、更准确的性能,融合了嵌入式和钦层感知器(Multilayer Perceptron,MLP)的优点的混合架构近年来被广泛地应用。该文对预测点击率的方法进行了全面的研究,不仅根据现有解决方案的架构将其分为三类,而且对每一类进行了详细的概述。最后,该文指出了该领域存在的挑战和未来发展方向,为进一步研究指明可能的途径。

In online advertising and recommender systems,it is critical to accurately estimate the click-through rate(CTR),a ratio of the number of clicks to the number of impressions.Many machine learning algorithms such as lo-gistic regression,SVM,etc.,are adopted to this task,which are all featured by the focus on feature engineering.Deep learning models that can effectively extract the high-order features automatically havea instinctive advantage in such work.To achieve more efficient and accurate performance,the hybrid architecture combining the embedding and MLP is more popular recently.This paper conduct a conscientious investigation into the approaches of estimating the click-through rate(CTR).Besides,we not only classify the existing solutions into three categories based on their architectures but also give a detailed overview of each category.We further present a comprehensive comparison of these state-of-the-art approaches.Finally,we summarize several open challenges and future directions in this field,which perhaps shed light on further studies.