The Design And Implementation Of Demand Side Platform For Online Advertising

With the rapid development of Internet technology,online advertising has become the most important techniques of advertising marketing.In recent years,with the further development of technologies such as machine learning and big data processing,computing advertising is evolving in the direction of accurate delivery.The programmatic transaction advertisement with real time bidding as the core has been widely concerned in the industry since its appearance.It has become a pivotal force in the display advertising market,and its development has made the entire online advertising market move toward a data-driven and computation-oriented way.As the most critical product in the real time bidding ecosystem,the demand side platform guides the precise placement of advertisements with "customized"labels,and reaches the extreme in traffic selection and control.Therefore,a lot of technical and algorithmic challenges are faced in its implementation.Click tthrough rate estimation is a very key technology in the demand side platform.Its accuracy will directly affect the effectiveness of advertising campaign and the revenue of advertisers.It is a research hotspot in both industry and academia field at present.To satisfy the requirements mentioned above,this thesis designs and implements an online advertising demand side platform,and conducts research on the click through rate estimation algorithms.The contributions of this thesis are summarized as follows:(1)This thesis analyzes the functional and non-functional requirements of the online advertising demand side platform.Combined with demand analysis and key technologies,the overall architecture of the system is designed.This thesis divides the system into seven modules,including:Web service module,bidding monitoring module,advertising management module,user orientation module,advertising recall module,advertising selection module and log collection and processing module.Then the design and implementation of each module are elaborated in detail.Finally,we performed both functional and performance test in this thesis to verify the availability of the system.(2)Click-through rate estimation is the most critical technology in demand side platform.To improve the accuracy of the click-through rate estimation,this thesis proposes a scheme based on a CTR predication algorithm:Factorization-Machine based Neural Network for CTR Prediction.The scheme better mines the interaction between features in different fields,but it introduces a large time cost.Then,this thesis proposes a further optimization scheme to reduce the time cost by pre-calculating the interaction weights between the features.Experiments shows that the proposed scheme can effectively improve the click through rate estimation without introducing too much additional time cost.