Field-weighted Factorization Machines for Click-Through Rate Prediction in Display Advertising Author : Junwei Pan, Jian Xu, Alfonso Lobos Ruiz, Wenliang Zhao, Shengjun Pan, Yu Sun, Quan Lu Source : WWW’ 18 Advisor : Jia-Ling Koh Speaker : Chia-Yi Huang Date : 2018/05/22
Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion � 2
Introduction ▸ Goal • Display ad CTR prediction use Field- weighted Factorization Machines � 3
Introduction ▸ Goal • Display ad CTR prediction use Field- weighted Factorization Machines � 4
Introduction ▸ Field & Feature � 5
Challenge ▸ Feature interactions are prevalent and need to be specifically modeled. ▸ Features from one field often interact differently with features from different other fields. ▸ Potentially high model complexity needs to be taken care of. � 6
Background ▸ Factorization Machine(FM, 因子分解機 ) ▸ Field-aware Factorization Machine(FFM) ▸ Field-weighted Factorization Machine(FwFM) � 7
Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion � 8
Evolution ▸ Logistic Regression model ▸ Degree-2 Polynomial model � 9
Evolution ▸ Factorization Machine ▸ Field-aware Factorization Machine(FFM) � 10
Evolution ▸ Field-weighted Factorization Machine(FwFM) � 11
Mutual Information � 12
Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion � 13
Experiment ▸ Data sets � 14
Experiment ▸ Comparison of FwFMs with Existing Models. � 15
Experiment ▸ Comparison of FwFMs and FFMs using the same number of parameters. � 16
Experiment ▸ L 2 Regularization � 17
Experiment ▸ Learning Rate � 18
Experiment ▸ Embedding Vector Dimension � 19
Experiment ▸ Learned field interaction strengths • For • For • For � 20
Experiment P1356 � 21
Experiment � 22
Outline ▸ Introduction ▸ Method ▸ Experiment ▸ Conclusion � 23
Conclusion ▸ FwFMs are competitive to FFMs with significantly less parameters. ▸ FwFMs can indeed learn different feature interaction strengths from different field pairs. � 24
Recommend
More recommend
