A CTR prediction method based on feature engineering and online learning
Jiehao Chen, Xueyi Li, Ziqian Zhao, Jiyun Shi, Zhang Qiu-Hong · 2017
Computational Advertising aims to advertise to specific group of audience and has been a hotspot issue in the field of emerging internet applications. The key problem is to predict the Click Through Rate(CTR) of an ad and it is usually done by machine learning ways. This essay proposed a method based on feature engineering and online training to predict the CTR of Search Ads. We use the Field-aware Factorization Machine(FFM) to abstract highly sparse feature vectors from the original ones and trained it with Follow-the-Regular-Leader(FTRL). Experiment results show that the method we proposed is 0.65%~6.44% more accurate than common prediction model, LR, and 29.72% more efficient than normal training methods.