Display advertisement click rate prediction method based on joint learning model

• Main Authors: LIU JINYU, LAI ZHI, LIU MENGJUAN, CAI SHIJIA, QIU LIZHOU, LI JIAXING
• Format: Patent
• Language: Chinese; English
• Published: 13.10.2020
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FORADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORYOR FORECASTING PURPOSES; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE,COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTINGPURPOSES, NOT OTHERWISE PROVIDED FOR

The invention discloses a display advertisement click rate prediction method based on a joint learning model, and aims to learn information of high-order features by using two different residual network substructures so as to improve the prediction accuracy of the model. The method comprises the following steps: firstly, establishing a sample based on an existing advertisement putting record, andpreprocessing features to obtain a training data set; secondly, establishing a fusion structure which comprises a first-order feature embedding layer, a hybrid embedding layer, a ResNet-Left substructure, a ResNet-Right substructure, a representation vector layer and an output node; thirdly, learning the parameters of the fusion structure by using the training data set to obtain final prediction model parameters; and finally, for a new advertisement display opportunity, calculating a predicted click rate based on the trained click rate prediction model. 本发明公开一种基于联合学习模型的展示广告点击率预测方法，目的是利用两个不同的残差网络子结构来分别学习高阶特征的信息，以提升模型的预