Deep Interest Network for Click-Through Rate Prediction外文学习材料.pdfVIP

Deep Interest Network for Click-Through Rate Prediction Guorui Zhou, Chengru Song, Xiaoqiang Zhu Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, Kun Gai Alibaba Group {guorui.xgr,chengru.scr,xiaoqiang.zxq,zhuhan.zh,fanying.fy,maxiao.ma,yanghui.yyh,junqi.jjq,lihan.hl,jingshi.gk}@ ABSTRACT These methods follow a similar EmbeddingMLP paradigm: large Click-through rate prediction is an essential task in industrial scale sparse input features are first mapped into low dimensional 8 applications, such as online advertising. Recently deep learning embedding vectors, and then transformed into fixed-length vectors 1 based models have been proposed, which follow a similar Embed- in a group-wise manner, finally concatenated together to fed into 0 dingMLP paradigm. In these methods large scale sparse input fully connected layers (also known as multilayer perceptron, MLP) 2 features are first mapped into low dimensional embedding vectors, to learn the nonlinear relations among features. Compared with p and then transformed into fixed-length vectors in a group-wise commonly used logistic regression model [19], these deep learning e manner, finally concatenated together to fed into a multilayer per- methods can reduce a lot of feature engineering jobs and enhance S ceptron (MLP) to learn the nonlinear relations among features. In the model capability greatly. For simplicity, we name these methods 3 this way, user features are compressed into a fixed-length repre-