基于注意力机制的兴趣网络点击率预估模型  

Click-Through Rate Prediction Model of Interest Network Based on Attention Mechanism

作　　者：许王昊 肖秦琨[1] XU Wanghao;XIAO Qinkun(School of Electronic and Information Engineering,Xi’an Technological University,Xi’an 710021,China)

机构地区：[1]西安工业大学电子信息工程学院,西安710021

出　　处：《计算机工程》2021年第1期101-108,共8页Computer Engineering

基　　金：国家自然科学基金面上项目(61271362,61671362);陕西省自然科学基础研究计划(2020JM-566)。

摘　　要：广告点击率(CTR)是互联网公司进行流量分配的重要依据,针对目前点击率预估精度较低的问题,结合通用的神经网络解决方案,构建一种基于注意力机制的深度兴趣网络(ADIN)模型。设计一个局部激活单元和自适应激活函数,根据用户历史行为和给定广告自适应地学习用户兴趣。引入注意力机制,区分不同特征对预测结果的影响程度,从而增强模型的可解释性。在3个公开数据集上的实验结果表明,相对LR、PNN等CTR预估模型,ADIN模型具有更高的AUC值和更低的LogLoss值,其预测效果更优。Advertising Click-Through Rate(CTR)is an important basis for Internet companies to allocate traffic.To address the inaccurate CTR prediction,this paper proposes an Attention mechanism-based Deep Interest Network(ADIN)model on the basis of the existing neural network solutions.This model has designed a local activation unit and an adaptive activation function to learn user interests adaptively based on the user’s historical behavior and the given advertisements.In addition,an attention mechanism is introduced to distinguish the contribution of different features to the prediction results,so that the interpretability of the model is enhanced.Experimental results on three public datasets show that compared with LR,PNN and other CTR estimation models,the proposed ADIN model has better prediction performance with a higher AUC value and a lower LogLoss value.

关 键 词：点击率预估 神经网络 局部激活 自适应激活函数 注意力机制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]