Research On Cross-platform Measurement Method Of Online Advertising

Online advertising is playing an increasingly important role in the advertising market.In 2016,the online advertising market in China reached 290.27 billion yuan,accounting for nearly 70%of the revenue of the ads market.At present,launching large-scale advertising through ad exchange platform(ADX)is the main way of online advertising.ADX matches users and ads that need to be displayed.In order to achieve high precision of advertising,ADX usually tracks the information of users,explores their interest from the information,and delivers relevant ads based on their interests.Currently,there are a large number of competing ADXs at home and abroad.How to answer the following questions,as for advertisers,are crucial:Which ADX platform to choose can increase the effectiveness of their ads campaigns?Therefore,it is the primary demand to identify and compare the advertising performance of ADX.Traditional method relies on training artificial online personas to represent behavioral traits.Then it uncovers existing correlation between users each exhibiting a certain behavioral trait and the display ads shown to them.This approach only measures and evaluates the performance of a single ADX.Due to without common measurement basis,this method does not able to apply to the comparative study of the performance of multiple ADXs.Therefore,in this dissertation,a synchronous cross-platform measurement method is proposed and implemented.This method can realize the comparison of the performance of different ADXs,and help advertisers select the appropriate ADX.The main work and contributions of this dissertation are as follows:(1)We proposed a benchmark for performance comparison of multiple ADX.Measuring multiple platforms independently by traditional method,these artificial personas inevitably have interesting feature deviations due to the inconsistent training pages on different platforms.In order to provide a unified measurement benchmark for multiple ADX,this dissertation proposes a method of training personas by overlapping web pages monitored by multiple platforms.This method can avoid the measurement deviation of single platform measurement method.(2)We proposed a method to access ad links in parallel.In traditional method of measurement,it is necessary to visit both the web page and the advertising content in the web page.Once you read the content of the advertisement,it will cause the platform's understanding of the persona.In order to solve this problem,the traditional method access the set of training web pages and then get the advertising content.However,this lag may lead to changes in advertising content,which can affect the measurement accuracy.We address this issue by getting access to these ad links concurrently with another full-time persona reading ads.In this way,the content of the advertisement link can be obtained in real time without causing any deviation from the knowledge of the currently measured persona.In order to evaluate the effectiveness of the proposed synchronous cross-platform measurement method proposed in this dissertation,we designed and developed a system to measure and compare the actual ADX platform.The experimental results reach the expected results.Under the same benchmark,the system can distinguish the advertising differences between different ADX for the same persona.It also finds that advertising on different ADX platforms will change with the change of personas,and the sensitivity of this change is different.