Research On Real Time Bidding Mechanism Design In Online Advertising

Online advertising is to advertise on the Internet through the online advertising platform. As a carrier conveying information to users, it has been proved that online advertising plays an important significance in the era of digital economy. It becomes an important commercial profit pattern for many Internet companies and the major source of website owners’revenue. With the development and the wide application of network technology, it has always been a concern that how to make an accurate matching for advertisers and ad slot under the background of big data, which means the right browser with the right message at the right moment in academia and industry.Click-through rate is an important indicator for online advertising, which helps advertisers optimize advertising performance and budget. However, due to its data sparsity, the estimation of click-through can hardly achieve high accuracy. Traditionally, we transfer this into a classification problem then use machine learning methods to solve it. This paper proposes a dynamic click-through rate estimation method based on data hierarchy, which improves the performance not only from the perspective of decision tree classifier by introducing the gradient boosting tree but also from the perspective of advertisers data itself. We map data both from space dimension and time dimension, which alleviates the data sparsity effectively and improves the reliability of the estimation greatly.Real-time bidding is the trend of the online advertising, and its algorithms will directly affect the advertisers’return on investment (ROI). At present, linear bidding strategy is commonly used, when the bid is proportional to click-through rate. This algorithm can meet the demand of most advertisers, but not apply for small advertisers whose budgets are limited. In this paper, we analyze advertisers’bid success probability model and draw the conclusion:when bidding is low, improving bidding can significantly increases the probability of winning the bidding. While the bidding is high, reducing bidding has few effects on the winning probability. Moreover, we put forward a budget reallocation model. In this Model, we divide bidding space into several intervals and cost the budget in the optimal price range. Combining the above tow points with the dynamic click-through rate estimation, this paper designs a piecewise bidding algorithm, which well protects the interests of small advertisers.Finally, we implement two algorithms on the Internet the company’s advertising data set. Compared with the traditional method, the results show that both two algorithms have great improvement in performance, which is significant to improve the ROI for the budget-limited advertisers.