CTR prediction in terms of multi-banner widgets.

The modern online advertising market is a fast-growing entity that brings together many advertisers who want to increase their business performance and media resources, who want to monetize their traffic. Between these two sides are located billions of users of browsers, mobile applications, games. The availability of accurate predictive models and optimal strategies is a competitive advantage for a company that provides services in this market. This paper examines and compares different approaches to the problem of predicting the CTR (click-through rate), and also compares the quality of different banner rotation strategies in the online advertising system using simplest reinforcement learning abstractions. The results indicate that it is possible to extrapolate existing experience to more complex situations, such as multi-banner rotation. The evaluation of the best approaches in terms of the possibility of their use in real production systems was also conducted.