Design And Implementation Of Information Flow Oriented AD Delivery System

With the popularity of mobile devices and the development of the Internet,online advertising has become an important means of commercial realisation for Internet companies.Under the traditional ad delivery model,ads are reached by users browsing the page to perform the operation of mandatory display of ads.Under this delivery method,the display of advertisements is independent of the context of the content,and users are frequently disturbed by uninteresting advertisements,which reduces the user experience.At the same time,due to the single delivery strategy,the traditional ad delivery cannot meet the increasing demand of advertisers for accurate crowd tagging.To solve these problems,this thesis proposes an ad delivery system for infomercials.The system combines the business scenario of infomercial with the CTR(Click Through Rate)model and CVR(Conversion Rate)model to recall the ads.In order to ensure smooth ad loading,this thesis optimises the ad delivery process.In the ad retrieval stage,the ad lists are regularly written to the Redis cache after the recall and sorting operations to speed up the efficiency and response of client requests.In order to deliver candidate ads to the target audience more accurately,the ad filtering rules are divided into five parts: user rule filtering,frequency control filtering,flow control filtering,style filtering and wind control filtering,and the candidate ads are constrained in the form of funnel filtering according to the above five types of delivery rules to achieve accurate delivery of ads by the ad delivery system.The ad placement system adopts MVC architecture and is functionally divided into four modules: ad placement management module,session management module,placement rules management module and ad retrieval module.The ad placement management module is responsible for generating ad candidate lists and returning ad styles,while the session management module is responsible for recording and processing user ad session events based on user behaviour modelling.The placement rules management module is responsible for generating filtering rules for the ads.The Ad Retrieval module is responsible for ad recall and ranking based on the ad recall model and ad ranking model.The ad placement system is now live and the system provides ad placement features that greatly improve ad placement efficiency and ad content matching,as well as positive revenue from commercialisation and user feedback.