Reading Notes on "Forecasting High-Dimensional Data"
“Forecasting High-Dimensional Data” is a paper by Yahoo! about traffic forecasting. In guaranteed advertising, it’s necessary to forecast the traffic volume for specific targeting in advance for reasonable selling and allocation. However, since there are many combinations of targeting (due to diverse advertiser needs), and engineering constraints don’t allow forecasting traffic for every possible targeting, this paper proposes to first forecast traffic for some basic targeting, then calculate traffic for various targeting combinations through a correlation model. This approach is highly practical and is also the traffic forecasting method used in the previously mentioned article “Budget Pacing for Targeted Online Advertisements at LinkedIn”.