A General Method For Automatic Discovery of Powerful Interactions In Click-Through Rate Prediction

被引:13
作者
Meng, Ze [1 ]
Zhang, Jinnian [2 ]
Li, Yumeng [3 ]
Li, Jiancheng [3 ]
Zhu, Tanchao [3 ]
Sun, Lifeng [4 ,5 ]
机构
[1] Tsinghua Univ, Dept Comp Sci & Technol, Beijing Key Lab Networked Multimedia, Beijing, Peoples R China
[2] Univ Wisconsin, Dept Elect & Comp Engn, Madison, WI 53706 USA
[3] Alibaba Grp, Hangzhou, Peoples R China
[4] Tsinghua Univ, Key Lab Pervas Comp, Minist Educ, Beijing, Peoples R China
[5] Tsinghua Univ, Dept Comp Sci & Technol, BNRist, Beijing, Peoples R China
来源
SIGIR '21 - PROCEEDINGS OF THE 44TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL | 2021年
基金
美国国家卫生研究院;
关键词
Click-through Rate Prediction; Gradient-based Neural Architecture; Search; Feature Interaction; Interaction Ensemble;
D O I
10.1145/3404835.3462842
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Modeling powerful interactions is a critical challenge in Click-through rate (CTR) prediction, which is one of the most typical machine learning tasks in personalized advertising and recommender systems. Although developing hand-crafted interactions is effective for a small number of datasets, it generally requires laborious and tedious architecture engineering for extensive scenarios. In recent years, several neural architecture search (NAS) methods have been proposed for designing interactions automatically. However, existing methods only explore limited types and connections of operators for interaction generation, leading to low generalization ability. To address these problems, we propose a more general automated method for building powerful interactions named AutoPI. The main contributions of this paper are as follows: AutoPI adopts a more general search space in which the computational graph is generalized from existing network connections, and the interactive operators in the edges of the graph are extracted from representative hand-crafted works. It allows searching for various powerful feature interactions to produce higher AUC and lower Logloss in a wide variety of applications. Besides, AutoPI utilizes a gradient-based search strategy for exploration with a significantly low computational cost. Experimentally, we evaluate AutoPI on a diverse suite of benchmark datasets, demonstrating the generalizability and efficiency of AutoPI over hand-crafted architectures and state-of-the-art NAS algorithms.
引用
收藏
页码:1298 / 1307
页数:10
相关论文
共 32 条
[1]  
[Anonymous], 2017, ATTENTIONAL FACTORIZ
[2]  
Blondel M, 2016, ADV NEUR IN, V29
[3]  
Cheng WY, 2020, AAAI CONF ARTIF INTE, V34, P3609
[4]  
Chu X., 2020, ARXIV200503566
[5]  
Elsken T, 2019, J MACH LEARN RES, V20
[6]  
Glen Anderson, 2016, P 1 WORKSH DEEP LEAR, P7
[7]  
Guo HF, 2017, PROCEEDINGS OF THE TWENTY-SIXTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, P1725
[8]   Neural Factorization Machines for Sparse Predictive Analytics [J].
He, Xiangnan ;
Chua, Tat-Seng .
SIGIR'17: PROCEEDINGS OF THE 40TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL, 2017, :355-364
[9]   MULTILAYER FEEDFORWARD NETWORKS ARE UNIVERSAL APPROXIMATORS [J].
HORNIK, K ;
STINCHCOMBE, M ;
WHITE, H .
NEURAL NETWORKS, 1989, 2 (05) :359-366
[10]   FiBiNET: Combining Feature Importance and Bilinear Feature Interaction for Click-Through Rate Prediction [J].
Huang, Tongwen ;
Zhang, Zhiqi ;
Zhang, Junlin .
RECSYS 2019: 13TH ACM CONFERENCE ON RECOMMENDER SYSTEMS, 2019, :169-177