Systems and methods for a framework for generating predictive models for media planning

1. A method for selecting a predictive model for predicting an average audience size at a future date, based on data quality and time until the future date, the method comprising:

receiving the future date for which to predict a future average audience size;

receiving, from a database, reference data associated with a program to be transmitted on the future date, wherein the reference data includes past schedule data for the program and an associated average audience size;

determining a prediction period between a current date and the future date;

determining, for the reference data, a granularity indicative of how often information of the past schedule data was measured;

retrieving, from memory, a plurality of predictive models;

determining, for a predictive model of the plurality of predictive models, a minimum granularity and maximum prediction period;

determining whether (1) the prediction period is less than the maximum prediction period and (2) the granularity is greater than the minimum granularity;

in response to determining that (1) the prediction period is less than the maximum prediction period and (2) the granularity is greater than the minimum granularity, selecting the predictive model of the plurality of predictive models;

subsequent to the selecting, training the predictive model by:

inputting, into the predictive model, a portion of the past schedule data for the program from the reference data;

generating, based on the predictive model, a predicted average audience size associated with the portion of the past schedule data;

determining a deviation between the predicted average audience size and an average audience size associated with the portion of the past schedule data; and

updating the predictive model based on the deviation; and

subsequent to the training of the predictive model, predicting the future average audience size by:

inputting, into the predictive model, future schedule data associated the program on the future date; and

generating, based on the predictive model, the future average audience size associated with the future schedule data.

2. The method of claim 1 , prior to predicting the future average audience size, comprising:

determining whether the deviation exceeds a threshold deviation;

based on determining that the deviation exceeds the threshold deviation, retraining the predictive model.

3. The method of claim 2 , wherein retraining the predictive model comprises:

inputting, into the predictive model, the portion of the past schedule data for the program from the reference data;

generating, based on the predictive model, a second predicted average audience size associated with the portion of the past schedule data;

determining a second deviation between the second predicted average audience size and the average audience size associated with the portion of the past schedule data;

updating the predictive model based on the second deviation.

4. The method of claim 1 , further comprising initializing one or more parameters of the predictive model to a random value.

5. The method of claim 1 , wherein the plurality of predictive models comprises a first predictive model that is suitable for a short-term prediction indicated by a smaller prediction period.

6. The method of claim 1 , wherein the plurality of predictive models comprises a second predictive model that is suitable for a long-term prediction indicated by a larger prediction period.

7. The method of claim 1 , comprising:

determining whether the prediction period exceeds a second threshold period;

based on determining that the prediction period exceeds the second threshold period, designating historical data type of the reference data to be time-based.

8. The method of claim 7 , comprising:

based on determining that the prediction period does not exceed the second threshold period:

based on receiving no future schedule data for the program, designating a historical data type to be time-based;

based on receiving the future schedule data for the program, designating the historical data type to be program-based.

9. The method of claim 8 , comprising determining a look-back period based on the prediction period and the historical data type.

10. The method of claim 9 , wherein the portion of the past schedule data for the program is selected based on the look-back period.

11. A system for selecting a predictive model for predicting an average audience size at a future date, based on data quality and time until the future date, comprising control circuitry configured to:

receive the future date for which to predict a future average audience size;

receive, from a database, reference data associated with a program to be transmitted on the future date, wherein the reference data includes past schedule data for the program and an associated average audience size;

determine a prediction period between a current date and the future date;

determine, for the reference data, a granularity indicative of how often information of the past schedule data was measured;

retrieve, from memory, a plurality of predictive models;

determine, for a predictive model of the plurality of predictive models, a minimum granularity and maximum prediction period;

determine whether (1) the prediction period is less than the maximum prediction period and (2) the granularity is greater than the minimum granularity;

in response to determining that (1) the prediction period is less than the maximum prediction period and (2) the granularity is greater than the minimum granularity, select the predictive model of the plurality of predictive models;

subsequent to the selecting, train the predictive model by:

inputting, into the predictive model, a portion of the past schedule data for the program from the reference data;

generating, based on the predictive model, a predicted average audience size associated with the portion of the past schedule data;

determining a deviation between the predicted average audience size and an average audience size associated with the portion of the past schedule data; and

updating the predictive model based on the deviation; and

subsequent to the training of the predictive model, predicting the future average audience size by:

inputting, into the predictive model, future schedule data associated with the program on the future date; and

generating, based on the predictive model, the future average audience size associated with the future schedule data.

12. The system of claim 11 , prior to predicting the future average audience size, wherein the control circuitry is configured to:

determine whether the deviation exceeds a threshold deviation;

based on determining that the deviation exceeds the threshold deviation, retrain the predictive model.

13. The system of claim 12 , wherein the control circuitry configured to retrain the predictive model comprises the control circuitry configured to:

input, into the predictive model, the portion of the past schedule data for the program from the reference data;

generate, based on the predictive model, a second predicted average audience size associated with the portion of the past schedule data;

determine a second deviation between the second predicted average audience size and the average audience size associated with the portion of the past schedule data;

update the predictive model based on the second deviation.

14. The system of claim 11 , wherein the control circuitry is further configured to initialize one or more parameters of the predictive model to a random value.

15. The system of claim 11 , wherein the plurality of predictive models comprises a first predictive model that is suitable for a short-term prediction indicated by a smaller prediction period.

16. The system of claim 11 , wherein the plurality of predictive models comprises a second predictive model that is suitable for a long-term prediction indicated by a larger prediction period.

17. The system of claim 11 , wherein the control circuitry is configured to:

determine whether the prediction period exceeds a second threshold period;

based on determining that the prediction period exceeds the second threshold period, designate a historical data type of the reference data to be time-based.

18. The system of claim 17 , wherein the control circuitry is configured to:

based on determining that the prediction period does not exceed the second threshold period:

based on receiving no future schedule data for the program, designate a historical data type to be time-based;

based on receiving the future schedule data for the program, designate the historical data type to be program-based.

19. The system of claim 18 , wherein the control circuitry is configured to determine a look-back period based on the prediction period and the historical data type.

20. The system of claim 19 , wherein the portion of the past schedule data for the program is selected based on the look-back period.

21. A method for predicting an average audience that can inform a future advertisement rate for insertion of an advertisement during transmission of a program on a future date, the method comprising:

receiving the future date for which to predict a future average audience size;

receiving, from a database, reference data associated with a program to be transmitted on the future date, wherein the reference data includes past schedule data for the program and an associated average audience size;

determining a prediction period between a current date and the future date;

determining whether the prediction period exceeds a threshold period;

based on determining that the prediction period does not exceed the threshold period, selecting a first type for a predictive model;

based on determining that the prediction period exceeds the threshold period, selecting a second type for the predictive model different than the first type;

determining whether the prediction period exceeds a second threshold period;

based on determining that the prediction period exceeds the second threshold period, designating a historical data type of the reference data to be time-based;

subsequent to the selecting one of the first type and the second type, training the predictive model by:

inputting, into the predictive model, a portion of the past schedule data for the program from the reference data;

generating, based on the predictive model, a predicted average audience size associated with the portion of the past schedule data;

determining a deviation between the predicted average audience size and an average audience size associated with the portion of the past schedule data; and

updating the predictive model based on the deviation; and

subsequent to the training of the predictive model, predicting the future average audience size by:

inputting, into the predictive model, future schedule data associated the program on the future date; and

generating, based on the predictive model, the future average audience size associated with the future schedule data.

22. The method of claim 21 , further comprising:

in response to determining that the prediction period does not exceed the second threshold period:

based on receiving no future schedule data for the program, designating a historical data type to be time-based;

based on receiving the future schedule data for the program, designating the historical data type to be program-based.

23. The method of claim 22 , comprising determining a look-back period based on the prediction period and the historical data type.

24. The method of claim 23 , wherein the portion of the past schedule data for the program is selected based on the look-back period.

25. A system for predicting an average audience size that can inform a future advertisement rate for insertion of an advertisement during transmission of a program on a future date, the system comprising control circuitry configured to:

receive the future date for which to predict a future average audience size;

receive, from a database, reference data associated with a program to be transmitted on the future date, wherein the reference data includes past schedule data for the program and an associated average audience size;

determine a prediction period between a current date and the future date;

determine whether the prediction period exceeds a threshold period;

based on determining that the prediction period does not exceed the threshold period, select a first type for a predictive model;

based on determining that the prediction period exceeds the threshold period, select a second type for the predictive model different than the first type;

determine whether the prediction period exceeds a second threshold period;

based on determining that the prediction period exceeds the second threshold period, designate a historical data type of the reference data to be time-based;

subsequent to the selecting one of the first type and the second type, train the predictive model by:

inputting, into the predictive model, a portion of the past schedule data for the program from the reference data;

generating, based on the predictive model, a predicted average audience size associated with the portion of the past schedule data;

determining a deviation between the predicted average audience size and an average audience size associated with the portion of the past schedule data; and

updating the predictive model based on the deviation; and

subsequent to the training of the predictive model, predict the future average audience size by:

inputting, into the predictive model, future schedule data associated the program on the future date; and

generating, based on the predictive model, the future average audience size associated with the future schedule data.

26. The system of claim 25 , wherein the control circuitry is configured to:

based on determining that the prediction period does not exceed the second threshold period:

based on receiving no future schedule data for the program, designate a historical data type to be time-based;

based on receiving the future schedule data for the program, designate the historical data type to be program-based.

27. The system of claim 26 , wherein the control circuitry is configured to determine a look-back period based on the prediction period and the historical data type.

28. The system of claim 27 , wherein the portion of the past schedule data for the program is selected based on the look-back period.