The Research And Realization Of Real-time Bidding Algorithm On DSP Of Online Advertising

Real-time Bidding accounts for the bulk of the programmatic buying advertisement on the DSP. Nowadays, more and more advertisers have understood the pattern of RTB and been pleasure to display online advertising with it. The publishers like to sell inventories of long-tail traffic to them. In the environment of the Sponsored Search advertisement and contract advertisement, RTB stands out from inventory and introduces in audience targeting. The right advertisement is delivered to the right person at the right time. It makes the advertising pattern that is provided with public bidding mechanism have a place in advertising market.The advertisers can receive bid request from Ad exchange and extract page information and user cookies as basis for accurate targeting advertisement through technological means. At the meanwhile, because of the pattern of the RTB, many DSPs compete for auction ad at the same time. Each of DSP bid their price with own algorithm. The whole market competition is fierce. The aspect of ROI is full of gaming.The paper introduces the RTB ecosystem and its workflow. It has described the feature of RTB and bidding pattern. The algorithm of the RTB has been studied in depth. The research is real-time bidding algorithm in this thesis. It contains two main topics, CTR prediction and bidding strategy. There are two CTR prediction methods. The one is logistic regression which is applied in sponsored search. Another is gradient boost decision tree which can handle no-linear continuous variables. After that, both of them have been compared with results of prediction performance. This thesis has proposed fixed bidding, segment bidding and synthesize bidding. The mathematical derivation of bidding function has been underlined by synthesize bidding. The parameter of function has been tuned by training datasets. Finally, the results of three bidding strategies has been evaluated.The datasets of this thesis is from i Pin You which has been provided them at global RTB algorithm competition in 2013. They are real datasets and few DSP company likes to provide their private data to public. It will be helpful to global RTB research that these datasets open to public. The result of the research show that key performance indicator of the synthesize bidding is better than the other two bidding methods.