Embedding calendar knowledge in event-driven inventory forecasting

Abstract

Systems and methods for automatically forecasting the future availability of one or more resources, such as Internet advertising opportunities, are described herein. In accordance with various embodiments, a forecasting model that accounts for event-driven resource availability is trained based both on historical supply data and calendar information specifying events and event duration. The trained forecasting model is then used to forecast the availability of resources at one or more specified future time periods. In accordance with certain embodiments, the forecasting model comprises a Gaussian process model that has an event-driven kernel as a covariance function.

1 - 20 . (canceled) 
     
     
         21 . A system stored in a non-transitory medium executable by a processor, comprising:
 a forecasting device configured to use a forecasting model to generate a forecasted number of resources available for a future time period,   wherein the forecasting model is based on historical supply data and calendar information,   wherein the historical supply data includes a past time period and a number of resources that were available during the past time period,   wherein the calendar information identifies an event and a corresponding time period during which the event occurs, and   wherein the forecasting model includes a covariance that is based on a distance function that defines a de-correlation level between a calendar feature vector associated with the historical supply data and a calendar feature vector associated with an event type.   
     
     
         22 . The system of  claim 21 , wherein the resources comprise online display advertising opportunities. 
     
     
         23 . The system of  claim 21 , wherein the forecasting model includes Gaussian process model. 
     
     
         24 . The system of  claim 21 , wherein the forecasting model is a non-parametric forecasting model. 
     
     
         25 . The system of  claim 21 , wherein the forecasting model includes an event-driven kernel with a definition, and wherein the definition of the event-driven kernel includes the distance function that defines a de-correlation level between a calendar feature vector associated with the historical supply data and a calendar feature vector associated with an event type. 
     
     
         26 . The system of  claim 21 , wherein the medium further comprises a monetizing device configured to monetize the resources. 
     
     
         27 . The system of  claim 21 , wherein the medium further comprises an ad booking server configured to book a new ad campaign according to the forecasted number of resources available for a future time period. 
     
     
         28 . The system of  claim 27 , wherein the booking of a new ad campaign is also according to an ad allocation plan. 
     
     
         29 . A method, comprising:
 identifying historical supply data, by a processor, the historical supply data including a past time period and a number of resources that were available during the past time period;   identifying calendar information, the calendar information identifying an event and a corresponding time period during which the event occurs;   determining a covariance according to a distance function that defines a de-correlation level between a calendar feature vector associated with the historical supply data and a calendar feature vector associated with an event type;   generating a forecasting model according to the historical supply data, the calendar information, and the covariance; and   using the forecasting model to generate a forecasted number of resources available for a future time period.   
     
     
         30 . The method of  claim 29 , wherein the resources comprise online display advertising opportunities. 
     
     
         31 . The method of  claim 29 , wherein the generating of the forecasting model is also according to a Gaussian process model. 
     
     
         32 . The method of  claim 29 , wherein the generating of the forecasting model is also according to a non-parametric forecasting model. 
     
     
         33 . The method of  claim 29 , wherein the covariance is implemented through an event-driven kernel. 
     
     
         34 . The method of  claim 29 , further comprising monetizing the resources. 
     
     
         35 . The method of  claim 29 , further comprising booking a new ad campaign according to the forecasted number of resources available for a future time period. 
     
     
         36 . The method of  claim 35 , wherein the booking of a new ad campaign is also according to an ad allocation plan. 
     
     
         37 . A non-transitory computer readable medium, comprising:
 instructions executable by a processor to identify historical supply data, the historical supply data including a past time period and a number of resources that were available during the past time period;   instructions executable by a processor to identify calendar information, the calendar information identifying an event and a corresponding time period during which the event occurs;   instructions executable by a processor to determine a covariance according to a distance function that defines a de-correlation level between a calendar feature vector associated with the historical supply data and a calendar feature vector associated with an event type;   instructions executable by a processor to generate a forecasting model according to the historical supply data, the calendar information, and the covariance;   instructions executable by a processor to use the forecasting model to generate a forecasted number of resources available for a future time period; and   instructions executable by a processor to book a new ad campaign according to the forecasted number of resources available for a future time period.   
     
     
         38 . The medium of  claim 37 , wherein the generating of the forecasting model is also according to a non-parametric Gaussian process model. 
     
     
         39 . The medium of  claim 37 , further comprising instructions executable by a processor to monetize the resources. 
     
     
         40 . The medium of  claim 37 , wherein the booking of a new ad campaign is also according to an ad allocation plan.