| With the rapid development of Internet applications,the number of Internet users has also expanded dramatically.For most publishers,internet advertising has become an important source of income.Advertisers might have some requirements in the ad-vertising process,such as 1)their ads must be assigned to users who have specified attributes,2)in a certain period,their ads need to achieve a certain amount of user as-signments,and 3)for the same users,one ad should not be showed many times to avoid making them disgusted.Therefore,for every online user's visit,how to choose the ad-vertisements to meet the needs of advertisers,and how to obtain the biggest possible income have become the focus of publishers.This dissertation abstracts the advertising problem into a series of theoretical mod-els for the real delivery system,such as the advertiser demand and data forecasting technology of Tencent Video.And this dissertation put forward a three-stage advertis-ing delivery program to meet the system needs.The concrete work is as follows:1.In view of the unpredictable user frequency in the current delivery system,this dissertation proposes a set of offline schemes that divide the patterns into sub-patterns after a group of patterns' generation,and allocates a sub-pattern online which can be modified dynamically for each arriving user.This dissertation also proposes a lightweight pattern generation algorithm,which satisfies the timeli-ness requirement of only 24h per period of delivery in the delivery system.2.This dissertation collects part of the real visit data of Tencent video and uses it to compare multiple sets of experiments.The experimental results show that the algorithm of this paper has only a small loss of interest compared to the tradi-tional scheme that is applicable to the predictable user frequency.And through the experimental data statistics,the results show that the pattern generation algo-rithm of this dissertation due to time-sensitive requirements caused by up to 10%.Therefore,this dissertation uses real data to prove that the algorithm is effective and time-effective in the delivery system of unpredictable user frequency. |
