(ICCC 2020) HFF: Hybrid Feature Fusion Model for Click-Through Rate Prediction

Abstract: "Deep neural network (DNN) which is applied to extract high-level features plays an important role in the Click Through Rate (CTR) task. Although the necessity of high-level features has been recognized, how to integrate high-level features with low-level features has not been studied well. There are some works fuse low- and high-level features by simply sum or concatenation operations. We argue it is not an effective way because they treat low- and high-level features equally. In this paper, we propose a novel hybrid feature fusion model named HFF. HFF modelconsists of two different layers: feature interaction layer and feature fusionlayer. With feature interaction layer, our model can capture high-level features. And the feature fusion layer can make full use of low- and high- level features. Comprehensive experiments on four real-world datasets are conducted. Extensive experiments show that our model outperforms existing the state-of-the-art models."

Authors: Yunzhou Shi (Tsinghua University, China); Yujiu Yang (Tsinghua University, Shenzhen, China)

Email: syz17@mails.tsinghua.edu.cn, yang.yujiu@sz.tsinghua.edu.cn