Research On The Related Algorithm Of Online Advertising Real - Time Bidding System

With the fast development of the Internet, it becomes more efficient and significant for advertisers to conduct online advertisements. The technology and methodology of online advertisement has been developed rapidly in recent years and this field is emerging as a new applied discipline,the Computational Advertising. In this paper,we provide an overview of online advertising industry and describe some state-of-art technologies and algorithms in this field,The Real-Time Bidding(RTB), aka programmatic buying, has recently become the fastest growing area in online advertising. Instead of bulking buying and inventory-centric buying, RTB mimics stock exchanges and utilizes computer algorithms to automatically buy and sell ad in real time. It uses per impression context and targets the ad to specific people based on data about them, and hence dramatically increases the effectiveness of display advertising. In this paper, we explain the basic concept and the architecture of RTB, and illustrate the algorithms of some important bidding strategies.However, with more complexity and diversity features in display-related advertising, it is a more challenging prediction task to extract the ad creative and user intension. The commonly used evaluation metrics in online advertising is CTR(Click-Through-Rate). And it is also the necessary procedure in many estimating modules of online advertising. To maximize ROI and user satisfaction,the ad system must predict the expected user behavior for each displayed advertisement and must maximize the expectation that a user will act(click) on it. It’s especially significant for advertisers to bid a price according to the estimated CTR of an impression in Real-Time Bidding. We explain how to predict the CTR by introducing some models and mathematical methods, such as Logical Regression and Factorization Machine method.In this paper, we focus on the research of bidding strategy, or named DSP Bid Optimization problem. This problem task for a given advertiser refers to optimizing a predefined KPI given a cost budget and the coming bid requests during the lifetime of the budget. KPI can be a straightforward CTR or conversion offered by advertisers. Apparently, it’s a constrained optimization problem that maximize KPI subjected to the limited budget. Here we came up with our specific perspectives and suggestions of RTB algorithm optimization. First a new budget pacing algorithm: Make statistical analysis from history bidding and feedback logs and scheduled a bidding plan of one ad campaign. Then a combined bidding strategy with our proposed budget pacing method: Evaluate every ad requests and make decision based on this scheme to participate the final bidding in ad exchange for better performance or ROI. From mathematical derivation and experiments in RTB datasets, we proved it superior to some existing methods and feasible for industrial practice.