Probabilistic modeling for anonymized data integration and bayesian survey measurement of sparse and weakly-labeled datasets

1. An audience measurement computing system comprising:

a network interface;

a processor; and

at least one memory storing computer-readable instructions that, when executed by the processor, cause the audience measurement computing system to perform operations comprising:

accessing input data via the network interface and from meters, the meters configured to monitor media devices associated with a plurality of panelists of an audience measurement entity, the input data including media source data and panel data;

training a model based on training data to produce a trained model configured to reduce a dimensionality of the input data and output a probability model parameter for a probability model, the probability model having dimensions corresponding to the input data;

executing the model to reduce the dimensionality of the input data and output the probability model parameter for the probability model;

evaluating the probability model based on census data, the input data, and the probability model parameter; and

based on the evaluating, updating the probability model by integrating the probability model over a domain of a feature of the input data.

2. The audience measurement computing system of claim 1 , the operations further comprising:

using the updated probability model to generate one or more audience ratings metrics; and

based on a determination that the one or more audience ratings metrics satisfy a ratings threshold:

selecting an advertisement, and

causing a media monitoring application to display the selected advertisement on a media presentation device.

3. The audience measurement computing system of claim 1 , wherein the model comprises a neural network.

4. The audience measurement computing system of claim 1 , wherein the model comprises two different neural networks.

5. The audience measurement computing system of claim 1 , wherein the dimensionality of the input data is reduced based on (i) a prior probability of an audience rating associated with the plurality of panelists and (ii) an approximation of a dependency of the audience rating on at least one of the media source data and the panel data.

6. The audience measurement computing system of claim 1 , wherein:

the census data includes partially labeled data associated with a media exposure,

the partially labeled data identifies at least one of a time of day associated with the media exposure, weather data associated with the media exposure, or demographic data associated with the media exposure, and

evaluating the probability model based on the census data comprises evaluating the probability model based on the partially labeled data.

7. The audience measurement computing system of claim 1 , wherein the updated probability model is configured to label the census data.

8. A non-transitory computer readable medium comprising computer-readable instructions that, when executed, cause a processor to perform operations comprising:

accessing input data via a network interface and from meters, the meters configured to monitor media devices associated with a plurality of panelists of an audience measurement entity, the input data including media source data and panel data;

training a model based on training data to produce a trained model configured to reduce a dimensionality of the input data and output a probability model parameter for a probability model, the probability model having dimensions corresponding to the input data;

executing the model to reduce the dimensionality of the input data and output the probability model parameter for the probability model;

evaluating the probability model based on census data, the input data, and the probability model parameter; and

based on the evaluating, updating the probability model by integrating the probability model over a domain of a feature of the input data.

9. The non-transitory computer readable medium of claim 8 , the operations further comprising:

using the updated probability model to generate one or more audience ratings metrics; and

based on a determination that the one or more audience ratings metrics satisfy a ratings threshold:

selecting an advertisement, and

causing a media monitoring application to display the selected advertisement on a media presentation device.

10. The non-transitory computer readable medium of claim 8 , wherein the model comprises a neural network.

11. The non-transitory computer readable medium of claim 8 , wherein the model comprises two different neural networks.

12. The non-transitory computer readable medium of claim 8 , wherein the dimensionality of the input data is reduced based on (i) a prior probability of an audience rating associated with the plurality of panelists and (ii) an approximation of a dependency of the audience rating on at least one of the media source data and the panel data.

13. The non-transitory computer readable medium of claim 8 , wherein:

the census data includes partially labeled data associated with a media exposure,

the partially labeled data identifies at least one of a time of day associated with the media exposure, weather data associated with the media exposure, or demographic data associated with the media exposure, and

evaluating the probability model based on the census data comprises evaluating the probability model based on the partially labeled data.

14. The non-transitory computer readable medium of claim 8 , wherein the updated probability model is configured to label the census data.

15. A method performed by a computing system comprising a processor, the method comprising:

accessing input data via a network interface and from meters, the meters configured to monitor media devices associated with a plurality of panelists of an audience measurement entity, the input data including media source data and panel data;

training a model based on training data to produce a trained model configured to reduce a dimensionality of the input data and output a probability model parameter for a probability model, the probability model having dimensions corresponding to the input data;

executing the model to reduce the dimensionality of the input data and output the probability model parameter for the probability model;

evaluating the probability model based on census data, the input data, and the probability model parameter; and

based on the evaluating, updating the probability model by integrating the probability model over a domain of a feature of the input data.

16. The method of claim 15 , further comprising:

using the updated probability model to generate one or more audience ratings metrics; and

based on a determination that the one or more audience ratings metrics satisfy a ratings threshold:

selecting an advertisement, and

causing a media monitoring application to display the selected advertisement on a media presentation device.

17. The method of claim 15 , wherein the model comprises two different neural networks.

18. The method of claim 15 , wherein the dimensionality of the input data is reduced based on (i) a prior probability of an audience rating associated with the plurality of panelists and (ii) an approximation of a dependency of the audience rating on at least one of the media source data and the panel data.

19. The method of claim 15 , wherein:

the census data includes partially labeled data associated with a media exposure,

the partially labeled data identifies at least one of a time of day associated with the media exposure, weather data associated with the media exposure, or demographic data associated with the media exposure, and

evaluating the probability model based on the census data comprises evaluating the probability model based on the partially labeled data.

20. The method of claim 15 , wherein the updated probability model is configured to label the census data.