Actual exposure depends on the ad size and ad traffic coverage levels are relatively competitive in competing ads. Directional flow of the crowd which depends on advertising to cover ad (matching the number of users corresponding features), creative size (matching ad slot) and put in time, budget and other settings [ ad triggered ]. The impact of advertising competitive bid mainly, ad quality, and other factors (such as pctr / pcvr, etc.), as well as control strategies of the user experience [ ad auction ]. Generally speaking, the basic competitiveness can ecpm = 1000 * cpc_bid * pctr = 1000 * cpa_bid * pctr * pcvr (cpc, cpa represent pay-per-click and pay-per-conversion mode mode). In summary, the former determines the number of times the ad can compete and competitors, which determines the probability of winning each competition. Both the final decision of the exposure of advertisements each day.
Question : based on exposure history log, a new estimate of the amount of daily advertising exposure.
Evaluation criteria :
1) the accuracy of the index;
2) bid monotonically related indicators;
Difficulty :
1, how to develop model optimization goals [how to combine the two evaluation criteria]
2, how to build a training and test sets, label is what?
3, how to manipulate data using advertising [operating time, delivery time, directed the crowd]
https://maimai.cn/article/detail?fid=1317017525&efid=S4zwF094BbaOA3EYLRBlpg