Real-time dayparting management

1. A system comprising:

one or more processors; and

one or more non-transitory computer-readable media storing computing instructions that, when executed on the one or more processors, cause the one or more processors to perform:

obtaining historical revenue per click (RPC) data;

generating hourly RPC prediction data for a predetermined time period based on the historical RPC data;

determining (i) time intervals from within the predetermined time period and (ii) a respective modifier for each of the time intervals, based on the hourly RPC prediction data;

uploading the time intervals and the respective modifiers for the time intervals to a dayparting system of a search engine;

obtaining real-time observed orders per minute (OPM) data;

training an autoregressive integrated moving average (ARIMA) time-series prediction model to make a real-time OPM prediction for a current time period, based on the real-time observed OPM data over past time steps based on lagged time steps in a moving average;

determining, in real-time, whether a demand surge exists based on the real-time observed OPM data and the real-time OPM prediction, to generate a first surge modifier, comprising:

comparing whether the real-time observed OPM data for the current time period exceeds the real-time OPM prediction for the current time period; and

when the real-time observed OPM data for the current time period exceeds the real-time OPM prediction for the current time period:

determining whether the real-time observed OPM data is a statistical outlier for the real-time OPM prediction; and

when the real-time observed OPM data is the statistical outlier for the real-time OPM prediction, determining that the demand surge exists;

when the demand surge exists, generating, in real-time, a sub-hour RPC prediction for a first sub-hour time interval;

determining, in real-time, the first surge modifier for the first sub-hour time interval; and

uploading, in real-time, the first surge modifier to the dayparting system of the search engine to bypass the time intervals and the respective modifiers.

2. The system of claim 1 , wherein generating the hourly RPC prediction data for the predetermined time period further comprises:

using a time-series forecasting model trained using rolling-horizon training and prediction.

3. The system of claim 1 , wherein determining the time intervals from within the predetermined time period further comprises:

using a decision tree classification model to determine the time intervals.

4. The system of claim 3 , wherein using the decision tree classification model to determine the time intervals further comprises:

using a loss function to minimize a difference between (i) hourly RPCs for a time interval of the time intervals and (ii) an average RPC for the time interval.

5. The system of claim 4 , wherein using the loss function further comprises:

weighting each hour of the time interval in the loss function based on a respective magnitude of each of the hourly RPCs.

6. The system of claim 1 , wherein determining whether the real-time observed OPM data is the statistical outlier for the real-time OPM prediction further comprises:

calculating a P value for the real-time observed OPM data based on the real-time OPM prediction.

7. The system of claim 6 , wherein determining whether the real-time observed OPM data is the statistical outlier for the real-time OPM prediction further comprises:

determining whether the P value is less than a predetermined threshold.

8. The system of claim 1 , wherein the computing instructions, when executed on the one or more processors, cause the one or more processors to further perform, while the demand surge exists:

determining a second surge modifier for a second sub-hour time interval, wherein the first sub-hour time interval and the second sub-hour time interval are within a single hour; and

uploading the second surge modifier to the dayparting system of the search engine to bypass the first surge modifier.

9. A method implemented via execution of computing instructions configured to run at one or more processors, the method comprising:

obtaining historical revenue per click (RPC) data;

generating hourly RPC prediction data for a predetermined time period based on the historical RPC data;

determining (i) time intervals from within the predetermined time period and (ii) a respective modifier for each of the time intervals, based on the hourly RPC prediction data;

uploading the time intervals and the respective modifiers for the time intervals to a dayparting system of a search engine;

obtaining real-time observed orders per minute (OPM) data;

training an autoregressive integrated moving average (ARIMA) time-series prediction model to make a real-time OPM prediction for a current time period, based on the real-time observed OPM data over past time steps based on lagged time steps in a moving average;

determining, in real-time, whether a demand surge exists based on the real-time observed OPM data and the real-time OPM prediction, to generate a first surge modifier, comprising:

comparing whether the real-time observed OPM data for the current time period exceeds the real-time OPM prediction for the current time period; and

when the real-time observed OPM data for the current time period exceeds the real-time OPM prediction for the current time period:

determining whether the real-time observed OPM data is a statistical outlier for the real-time OPM prediction; and

when the real-time observed OPM data is the statistical outlier for the real-time OPM prediction, determining that the demand surge exists;

when the demand surge exists, generating, in real-time, a sub-hour RPC prediction for a first sub-hour time interval;

determining, in real-time, the first surge modifier for the first sub-hour time interval; and

uploading, in real-time, the first surge modifier to the dayparting system of the search engine to bypass the time intervals and the respective modifiers.

10. The method of claim 9 , wherein generating the hourly RPC prediction data for the predetermined time period further comprises:

using a time-series forecasting model trained using rolling-horizon training and prediction.

11. The method of claim 9 , wherein determining the time intervals from within the predetermined time period further comprises:

using a decision tree classification model to determine the time intervals.

12. The method of claim 11 , wherein using the decision tree classification model to determine the time intervals further comprises:

using a loss function to minimize a difference between (i) hourly RPCs for a time interval of the time intervals and (ii) an average RPC for the time interval.

13. The method of claim 12 , wherein using the loss function further comprises:

weighting each hour of the time interval in the loss function based on a respective magnitude of each of the hourly RPCs.

14. The method of claim 9 , wherein determining whether the real-time observed OPM data is the statistical outlier for the real-time OPM prediction further comprises:

calculating a P value for the real-time observed OPM data based on the real-time OPM prediction.

15. The method of claim 14 , wherein determining whether the real-time observed OPM data is the statistical outlier for the real-time OPM prediction further comprises:

determining whether the P value is less than a predetermined threshold.

16. The method of claim 9 further comprising, while the demand surge exists:

determining a second surge modifier for a second sub-hour time interval, wherein the first sub-hour time interval and the second sub-hour time interval are within a single hour; and

uploading the second surge modifier to the dayparting system of the search engine to bypass the first surge modifier.