| E-commerce advertising bidding ranking refers to the process in which e-commerce platform designs different advertising positions,advertisers choose advertising position and bid according to their needs,and e-commerce platform ranks and charges advertising according to advertisers’ bids and other factors.This process also implies that the advertiser adjusts the bid based on user feedback.E-commerce advertising bidding ranking involves multiple transaction subjects such as users,advertisers and advertising platforms,which plays a huge role in the operation of e-commerce.Appropriate advertising bidding ranking strategy can not only improve the revenue of advertising platforms,but also improve user experience and enhance user stickiness.The feedback cycle of traditional advertisement bidding ranking process is long,and users’ feedback on advertisements cannot be obtained in real time,which will affect the embodiment of advertisers’ value and thus affect the revenue of advertising platform.With the continuous in-depth research and application of e-commerce big data technology,e-commerce transaction behavior data continues to accumulate,forming fine granular behavior track big data coordinated by numerous transaction subjects.This provides an opportunity to study the precise decision of e-commerce advertising bidding ranking.Aiming at the decision-making problem of the three parties in the bidding ranking process of e-commerce advertising,this paper takes the advertiser and the advertising platform as the research focus,and studies the pricing decision-making law of the advertiser.Reinforcement learning plays an outstanding role in solving the sequential decision problem.The advertiser price adjustment decision problem studied in this paper is a typical sequential decision problem.So we use Markov decision process for advertisers modeling,pricing decision model is proposed based on depth against the reinforcement learning algorithm,through strategy and real advertisers maximize model decision strategies of the similar degree,finally constructed for ordering strategy based on advertisers decision-making simulation platform Virtual-ADer.Virtual-ADer is based on generative adversarial network,which includes advertiser decision model and discriminant model.The advertiser decision model is based on actor-critic reinforcement learning method.In order to optimize training and improve the robustness of the model,Virtual-ADer introduced an optimization idea,namely WGAN-GP optimization idea with gradient penalty.In addition,a pre-training model of Virtual-ADer is also proposed in this paper,which can not only verify the effectiveness of the above optimization ideas for the construction of simulation platform,but also improve the convergence speed of the deep adversarial reinforcement learning model.For the different bidding ranking strategies of the advertising platform,the "virtual advertiser" analyzes the impact of each bidding ranking strategy on the revenue of the advertising platform by simulating the price adjustment decision-making behavior of the advertisers,and provides the decision-making basis for the advertising platform to update the bidding ranking strategy.Based on the real advertisement exposure data,this paper develops a simulation platform of advertisement bidding ranking strategy based on the advertiser’s price adjustment decision.It can test the influence of different bidding ranking strategies on the advertiser’s price adjustment behavior,so as to analyze the advantages and disadvantages of bidding ranking strategies.This paper designs experiments to prove that the proposed advertiser decision model in Virtual-ADer has a good simulation effect on the price adjustment behavior of real advertisers,and verifies the effectiveness of the simulation platform.Through the simulation of the advertiser’s price adjustment behavior,the model can provide the decision basis for the advertising platform to choose the optimal ranking strategy in the complex bidding ranking strategy. |
