基于积注意力交互网络模型的点击率预测  

CLICK THROUGH RATE PREDICTION BASED ON PRODUCT ATTENTION INTERACTION NETWORK MODEL

作　　者：张安勤 王迎香 田秀霞 Zhang Anqin;Wang Yingxiang;Tian Xiuxia(College of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200135,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200135

出　　处：《计算机应用与软件》2024年第3期63-69,80,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61772327)。

摘　　要：如何提高广告点击率是对大数据网络营销的一个具有挑战的问题。考虑到用户点击行为的不确定性,提出一种基于积注意力交互网络模型的点击率预测模型。将用户的行为向量进行内积或外积,并根据广告自身的特征赋予交互后向量相应权重,然后进行点击率预测。在两个数据集上进行实验验证,结果表明该模型相对于传统的点击率预测模型在归一化基尼系数上提高了2%以上,预测效果更好。How to improve the click through rate of advertisement is a challenge to the network marketing in the era of big data.Considering the uncertainty of user s click behavior,a click-through rate prediction model based on product attention interactive network model is proposed.The model made the inner or outer product of the user s behavior vector,and gave the corresponding weight to the interactive vector according to the characteristics of advertising itself.Experiments were carried out on two data sets.The results show that the proposed model can improve the normalized Gini coefficient by more than 2%compared with the traditional hit rate prediction model,and can predict more accurately.

关 键 词：点击率 注意力机制 因子分解机 内积 外积 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]