ABSTRACT
In computational advertising, click-through rate (CTR) prediction is one of the best-known and the most important modeling problems. The purpose of CTR prediction is to rank the retrieved candidate ads. This chapter outlines the principles of logistic regression models and various optimization methods. It introduces some other models based on the evolution of CTR model in recent years, so as to guide readers to choose appropriate model schemes for their own scenarios. For logistic regression (LR) model, the authors usually use the MLS to solve the weighted coefficient. There are many calculation methods for the maximum likelihood solution of LR model, but they focus on its convergence rate and the convenience of distributed computing in the face of massive data in practice. The main challenge in CTR prediction is to enable the model to capture highly dynamic market signals, so as to achieve more accurate forecasting purposes.
