Dynamic Bidding Strategy Based on Probabilistic Feedback in Display Advertising

Bidding strategy is an issue of fundamental importance to Demand Side Platform (DSP) in real-time bidding (RTB). Bidding strategies employed by the Demand Siders may have significant impacts on their own benefits. In this paper, we design a dynamic bidding strategy based on probabilistic feedback, called PFDBS, which is different from previous work that is mainly focused on fixed strategies or continuous feedback strategies. Our dynamic bidding strategy is more in accordance with environment of Internet advertising to solve the instability problem. If evaluated valid, we will retain the current strategy, otherwise, we present an approach to amend strategy combined with previous feedback. The experiments on real-world RTB dataset demonstrate that our method has the best performance on Key Performance Indicator (KPI) compared to other popular strategies, meanwhile, the consumption trend of overall budget is the most consistent with real market situation.