Predicting future viewership
Approaches provide for predictive viewership associated with a device. Information associated with viewership by the device may be received over an interval. The received viewership information is merged with panel information to further generate merged information. The merged information is then aggregated at a predetermined increment to form aggregated date. The aggregated data can then be used as input training data to a model to generate probability of viewership by the device. One or more metrics associated with the predicted viewership can be tracked to evaluate model performance.
1. A computer implemented method of predicting a probability of viewership by a device, the method comprising:
receiving, over an interval, information associated with the viewership by the device;
merging the received information with source of content displayed on the device and information associated with one or more partnering devices to track content viewership and generate merged information;
aggregating the merged information at an increment to generate aggregated data;
providing the aggregated data as training data to a first model by selecting features of the aggregated data as the training data to train the first model, wherein the first model is trained to perform probability prediction of content viewership on the device;
generating one or more metrics associated with first model output;
comparing the one or more metrics against a baseline model generated from the aggregated data to track performance of the first model output;
providing the selected features as input training data to train a second model as a network viewership model, wherein the second model is trained to perform probability prediction of network viewership on the device; and
performing evaluation of the second model for feature correlations and generating a second model score based on the evaluation of the second model,
wherein the probability prediction of the network viewership by the second model is performed only after the device is predicted to be viewed under the content viewership prediction as performed by the first model,
wherein the one or more partnering devices are identified from a device list that is updated on a periodic basis, wherein the device list comprises a panel of devices that provide accurate content information, and the one or more partnering devices are categorized by make and model, and
wherein the first model performs the probability prediction of the content viewership on the device.
2. The method of claim 1 , wherein the receiving, over an interval, information associated with the viewership by the device comprises:
receiving content viewership information by the device over the interval; and
receiving additional datasets associated with viewership by the device over the interval.
3. The method of claim 2 , wherein the content viewership information is generated by matching instances of device content associated with instances of fingerprint data using an automatic content recognition (ACR) process.
4. The method of claim 2 , wherein the increment is a predetermined increment that is adjustable between a range of a minute and an hour.
5. The method of claim 1 , wherein the device list is updated on a weekly basis.
6. The method of claim 1 , further comprising:
performing evaluation of the first model for feature correlations and generating a first model score based on the evaluation of the first model, wherein the first model score is represented by one or more Shapley Additive Explanation (SHAP) values.
7. The method of claim 1 , wherein the one or more metrics comprises one or more network viewership metrics and content viewership metrics, and
wherein the one or more metrics track prediction performance of the first model by comparing the first model output against actual device activity.
8. The method of claim 1 , wherein the interval can be adjusted to between a minute and an hour.
9. The method of claim 1 , further comprising:
performing probability prediction of program viewership on the device by overlaying network viewership predictions as generated by the second model with program scheduling information.
10. A computer device, comprising:
a processor; and
memory including instructions that, when executed by the processor, cause the computing device to:
receiving, over an interval, information associated with viewership by the device;
merging the received information with information associated with one or more partnering devices to track content viewership and generate merged information;
aggregating the merged information at an increment to generate aggregated data;
providing the aggregated data as training data to a first model by selecting features of the aggregated data as the training data to train the first model, wherein the first model is trained to perform probability prediction of content viewership on the device;
generating one or more metrics associated with first model output; and
comparing the one or more metrics against a baseline model to track performance of the first model output,
providing the selected features as input training data to train a second model as a network viewership model, wherein the second model is trained to perform probability prediction of network viewership on the device;
performing evaluation of the second model for feature correlations and generating a second model score based on the evaluation of the second model,
wherein the probability prediction of the network viewership by the second model is performed only after the device is predicted to be viewed under the content viewership prediction as performed by the first model,
wherein the one or more partnering devices are identified from a device list that is updated on a periodic basis, wherein the device list comprises a panel of devices that provide accurate content information, and the one or more partnering devices are categorized by make and model, and
wherein the first model performs the probability prediction of the content viewership on the device.
11. The computer device of claim 10 , wherein the receiving, over an interval, information associated with the viewership by the device comprises:
receiving content viewership information by the device over the interval; and
receiving additional datasets associated with viewership by the device over the interval.
12. The computer device of claim 11 , wherein the content viewership information is generated by matching instances of device content associated with instances of fingerprint data using an automatic content recognition (ACR) process.
13. The computer device of claim 11 , wherein the increment is a predetermined increment that is adjustable between a range of a minute and an hour.
14. The computer device of claim 10 , wherein the device list is updated on a weekly basis.
15. The computer device of claim 10 , wherein the processor is further configured to:
perform evaluation of the first model for feature correlations and generating a first model score based on the evaluation of the first model, wherein the first model score is represented by one or more Shapley Additive Explanation (SHAP) values.
16. The computer device of claim 10 , wherein the one or more metrics comprises one or more network viewership metrics and content viewership metrics, and
wherein the one or more metrics track prediction performance of the first model by comparing the first model output against actual device activity.
17. The computer device of claim 10 , wherein the interval can be adjusted to between a minute and an hour.
18. The computer device of claim 10 , further comprising:
perform probability prediction of program viewership on the device by overlaying network viewership predictions as generated by the second model with program scheduling information.