Displaying Ad with Optimal Real-Time Bidding

Abstract

Real-time bidding is a new paradigm for displaying an ad. Through our research work, we have tried to find a bid optimization solution for displaying an ad in RTB. Advertisers are able to bid per impression through RTB to display their ads in publisher sites. The internal mechanism is quite complex and it correlates different parameters like user data, demographic location, culture and so on to determine the winning bid. In addition, it must be mentioned that this is different from the sponsored search auction where the bid price is related to keywords. Considering the budget, the objective of the predefined campaign and miscellaneous information collection in run time and from history is the key challenge for DSP. In our project, optimizing the bid in a programmatic manner is the desired problem. We have tried to develop a simple optimization bidding function which will be used to calculate in real-time within certain limitations. Finding non-linearity was the sole purpose of our work and it simply proves that CTR and CVR rate have that relationship with each and every estimated impression with different level of features. All the earlier works are basically focused on budget capping or reducing campaign period or prioritizing key features which are all falling in bidding with linearity. Bidding optimally which is our mathematical derivation indicates that conventional bidding strategy should be changed from high-value low set of impression to low value set of a huge impression because firstly it is much more cost-effective and secondly and definitely increases the winning rate. Moreover, effectiveness and outperformance of our optimization framework and optimal bidding strategy have been shown by offline and online evaluation using a real dataset and production RTB system.