Research On Profit Optimization Algorithms Of Online Advertising Demand Side Platform

In recent years,the rapid development of Internet technology has brought great convenience to numerous Internet users,causing the number of them soared and user groups diversified.At present,people's life is increasingly inseparable from the Internet.Many traditional offline activities have been gradually transformed into online activities and it is the same with advertising campaign.It is rapidly transformed to give full play to the advantages of online advertisement.As the main way of delivery in the fastest growing online display advertising,real-time bidding has attracted great attention in the industrial and academic circles.How to make a more accurate decision on the display of each advertisement requires the use of data analysis,regression prediction and a variety of optimization strategy algorithms to maximize revenue,which provides challenges and a huge research space for the majority of researchers.From the perspective of demand side platform,this thesis aims to improve its profit by optimizing the key algorithms in real-time bidding,i.e.click-rate prediction algorithm and bidding strategy algorithm.Click-through rate prediction is a key part in recommendation system and online advertisement.This thesis first investigates several typical click-through rate prediction models,particularly the fusion-structure based deep learning methods.A click rate prediction model based on the fusion structure is further advanced.In this model,the deep neural networks of different structures can be flexibly compromised to study the original high-dimensional sparse characteristics of high order,which makes more advanced feature information available in the click-through rate prediction model.In this research,real data sets are used to evaluate the prediction performance of the model.The experimental results show that the deep learning prediction model based on fusion structure can achieve better performance than both the traditional click-through rate prediction model and the latest prediction model based on deep learning.In real-time bidding,the demand side platform will make a bidding decision for each advertising request that it receives,which is setting a bid for the advertising request after meeting the demand of advertisers.This bid determines whether or not the advertising opportunity will be available to display advertisement of the advertiser.Therefore,a good bidding algorithm can bring huge profits increase to the demand side platform.In this thesis,the reinforcement learning technology is used to model the winning bid price,so as to give a more reasonable one.Most of the traditional bidding algorithms need to assume the winning price distribution in advance and cannot adapt to the dynamic data,while the bidding algorithm based on reinforcement learning can automatically learn from the data,which is more adaptable to the dynamic data and has stronger generalization ability.The experimental results show that this algorithm has better performance than the traditional bidding algorithm.