Video view estimation for shows delivered using a video delivery service

1. A method comprising:

sending, by a computing device, videos to users that use a video delivery service, wherein at least a portion of the videos include shows that have episodes released sequentially that are requested on demand when released;

recording, by the computing device, historical records of video views for the video based on the sending of the at least a portion of the videos to the users that were requested on demand;

for a show, determining, by the computing device, a show-specific model to predict future video views by performing:

determining, by the computing device, historical records of video views for different episodes of the show that were requested on demand;

training, by the computing device, the show-specific model with the historical records, wherein the show-specific model models a decay curve with a slow growing term that regularizes a decay speed of a decay of the decay curve;

using, by the computing device, the show-specific model to predict future video views for requesting the show on demand for a future time range for episodes of the show, wherein using the show-specific model comprises:

predicting, by the computing device, future video views for existing episodes of the show by setting a first day of the show-specific model to a first day in the future time range;

predicting, by the computing device, future release dates of future episodes of the show; and

predicting, by the computing device, future video views in the future time range for future episodes of the show by aligning a first day of the show-specific model to the respective future release dates of the future episodes; and

outputting, by the computing device, the future video views to an ad system configured to sell ads for the show.

2. The method of claim 1 , wherein the show-specific model comprises a log linear model for the decay curve with the slow growing term.

3. The method of claim 1 , wherein the slow growing term follows a normal distribution.

4. The method of claim 3 , wherein the normal distribution is substantially near −1 with a variance of σ.

5. The method of claim 1 , wherein predicting future release dates of future episodes of the show comprises:

generating a seasonal first day views model that models first day views for different episodes in a season for the show.

6. The method of claim 5 , wherein the seasonal first day views model changes the first day views based on different factors for the show over time.

7. The method of claim 5 , wherein predicting future video views comprises:

using the first day views model to determine future video views for a future episode.

8. The method of claim 1 , wherein predicting future release dates of future episodes of the show comprises:

generating a seasonal week release and weekday model that models weekly releases of episodes of the show and weekday releases of episodes.

9. The method of claim 8 , wherein predicting future video views comprises:

using the seasonal week release and weekday model to determine future video views for a future episode in a week and on a day of the week determined by the seasonal week release and weekday model.

10. The method of claim 8 , wherein predicting future video views comprises:

estimating a hiatus for the show in which episodes for the show are not released.

11. The method of claim 1 , further comprising:

reviewing the historical records to determine any points that are considered outliers by a threshold; and

removing the outlier points from the historical records.

12. The method of claim 1 , wherein generating the show-specific model comprises:

analyzing the historical records of a latest N number of episodes for the show to determine the show-specific model.

13. The method of claim 1 , wherein generating the show-specific model comprises:

aggregating the historical records of other movie and shows into a virtual show to determine the show-specific model.

14. The method of claim 1 , wherein the show-specific model is a long-form show-specific model, the method further comprising:

determining a short-form show specific model for a shorter form of the video than that used for the long-form show-specific model; and

using the long-form show-specific model and the short-form show-specific model to estimate the video views for the shorter form of the video.

15. A non-transitory computer-readable storage medium containing instructions, that when executed, control a computer system to be configured for:

sending videos to users that use a video delivery service, wherein at least a portion of the videos include shows that have episodes released sequentially that are requested on demand when released;

recording historical records of video views for the video based on the sending of the at least a portion of the videos to the users that were requested on demand;

for a show, determining a show-specific model to predict future video views by performing:

determining historical records of video views for different episodes of the show that were requested on demand;

training the show-specific model with the historical records, wherein the show-specific model models a decay curve with a slow growing term that regularizes a decay speed of a decay of the decay curve;

using the show-specific model to predict future video views for requesting the show on demand for a future time range for episodes of the show, wherein using the show-specific model comprises:

predicting future video views for existing episodes of the show by setting a first day of the show-specific model to a first day in the future time range;

predicting future release dates of future episodes of the show; and

predicting future video views in the future time range for future episodes of the show by aligning a first day of the show-specific model to the respective future release dates of the future episodes; and

outputting the future video views to an ad system configured to sell ads for the show.

16. The non-transitory computer-readable storage medium of claim 15 , wherein the show-specific model comprises a log linear model for the decay curve with the slow growing term.

17. The non-transitory computer-readable storage medium of claim 15 , wherein the slow growing term follows a normal distribution substantially near −1 with a variance of σ.

18. An apparatus comprising:

one or more computer processors; and

a non-transitory computer-readable storage medium comprising instructions, that when executed, control the one or more computer processors to be configured for:

sending videos to users that use a video delivery service, wherein at least a portion of the videos include shows that have episodes released sequentially that are requested on demand when released;

recording historical records of video views for the video based on the sending of the at least a portion of the videos to the users that were requested on demand;

for a show, determining a show-specific model to predict future video views by performing:

determining historical records of video views for different episodes of the show that were requested on demand;

training the show-specific model with the historical records, wherein the show-specific model models a decay curve with a slow growing term that regularizes a decay speed of a decay of the decay curve;

using the show-specific model to predict future video views for requesting the show on demand for a future time range for episodes of the show, wherein using the show-specific model comprises:

predicting future video views for existing episodes of the show by setting a first day of the show-specific model to a first day in the future time range;

predicting future release dates of future episodes of the show; and

predicting future video views in the future time range for future episodes of the show by aligning a first day of the show-specific model to the respective future release dates of the future episodes; and

outputting the future video views to an ad system configured to sell ads for the show.

19. The non-transitory computer-readable storage medium of claim 15 , wherein the slow growing term follows a normal distribution.

20. The non-transitory computer-readable storage medium of claim 15 , wherein the show-specific model is a long-form show-specific model, further configured for:

determining a short-form show specific model for a shorter form of the video than that used for the long-form show-specific model; and

using the long-form show-specific model and the short-form show-specific model to estimate the video views for the shorter form of the video.