Granted Patent B2

US 12,455,985 · App. 17/156,517 · Granted Oct 28, 2025

Inventors: Jonathon C. Peruski (Vienna, VA); Bonnie E. Harvey (Arlington, VA); Xuyao Jiang (Rockville, MD); Frank E. Pecjak (Kirkland, WA)

Assignee: Comscore, Inc.

G06F21/6263G06F21/6254G06N3/04G06N3/088

Abstract

Claims (46)

1 . A method for securing online data privacy, the method comprising:

training a machine learning model using panel data, wherein the panel data comprises measurements of interactions actually performed by panelists during a predetermined time period;

generating a plurality of personas that are initially empty;

receiving census data associated with first online interactions actually performed during the predetermined time period, wherein the census data is different from the panel data used to train the machine learning model;

generating, via the machine learning model for each of the plurality of personas, data associated with second online interactions that simulate Internet traffic, wherein input for the generation of the data associated with the second online interactions comprises event data that is sampled from the census data on a periodic basis, and wherein each of the plurality of personas does not include more than a threshold percentage of events associated with a user associated with the persona;

populating the plurality of personas with the generated data associated with the second online interactions;

selecting a plurality of the received census data associated with the first online interactions that match the generated data associated with the second online interactions for each of the plurality of personas;

replacing the generated data associated with the second online interactions populated in each of the plurality of personas with the selected data; and

outputting the plurality of personas with the replaced data.

2 . The method of claim 1 , wherein none of the received census data is associated with a persistent identifier of a device or a user that performed the first online interactions.

3 . The method of claim 1 , further comprising:

initiating the generation of data associated with second online interactions for each of the plurality of personas based on one or more random numbers.

4 . The method of claim 3 , wherein the Internet traffic is simulated for different types of panelists that are individually selected based on at least one of a demographic and device attribute.

5 . The method of claim 1 , wherein the first online interactions are performed by at least one of a plurality of census network members or a plurality of subscribers to a third party that tracks the subscribers without using trackers.

6 . The method of claim 1 , wherein each of the first online interactions comprises at least one of a single web browsing interaction, online application interaction, or a media consumption interaction.

7 . The method of claim 1 , wherein the machine learning model is an adversarial or variational autoencoder.

8 . The method of claim 1 , further comprising:

arbitrarily generating an identifier for each of the plurality of personas.

9 . The method of claim 1 , wherein matching the generated data associated with the second online interactions for each of the plurality of personas is behaviorally based on a measure of statistical distance to the received census data associated with the first online interactions.

10 . The method of claim 1 , wherein matching the generated data associated with the second online interactions for each of the plurality of personas is randomly assigned.

11 . The method of claim 10 , wherein the matching is performed such that at least a same number of actual events is recorded into each of the plurality of personas.

12 . The method of claim 1 , further comprising:

clustering the plurality of personas across time or across comparable activity data.

13 . The method of claim 1 , wherein the machine learning model is trained using the panel data generated by a first set of panelists using a first device characteristic, and wherein a different machine learning model is trained using other panel data generated by a different set of panelists using another device characteristic.

14 . The method of claim 1 , wherein a sum of the second online interactions equals a sum of the first online interactions.

15 . A system comprising:

at least one processor; and

at least one memory storing instructions that, when executed, cause the at least one processor to:

train a machine learning model using panel data, wherein the panel data comprises measurements of interactions actually performed by panelists during a predetermined time period;

generate a plurality of personas that are initially empty;

receive census data associated with first online interactions actually performed during the predetermined time period, wherein the census data is different from the panel data used to train the machine learning model;

generate, via the machine learning model for each of the plurality of personas, data associated with second online interactions that simulate Internet traffic, wherein input for the generation of the data associated with the second online interactions comprises event data that is sampled from the census data on a periodic basis, and wherein each of the plurality of personas does not include more than a threshold percentage of events associated with a user associated with the persona;

populate the plurality of personas with the generated data associated with the second online interactions;

select a plurality of the received census data associated with the first online interactions that match the generated data associated with the second online interactions for each of the plurality of personas;

replace the generated data associated with the second online interactions populated in each of the plurality of personas with the selected data; and

output the plurality of personas with the replaced data.

16 . The system of claim 15 , further comprising a storage device, wherein a storage space of the storage device is reduced by an extent that satisfies a criterion due to an encoding function of the machine learning model.

17 . A non-transitory, computer-readable medium storing instructions that, when executed, cause:

training a machine learning model using panel data, wherein the panel data comprises measurements of interactions actually performed by panelists during a predetermined time period;

generating a plurality of personas that are initially empty;

receiving census data associated with first online interactions actually performed during the predetermined time period, wherein the census data is different from the panel data used to train the machine learning model;

generating, via the machine learning model for each of the plurality of personas, data associated with second online interactions that simulate Internet traffic, wherein input for the generation of the data associated with the second online interactions comprises event data that is sampled from the census data on a periodic basis, and wherein each of the plurality of personas does not include more than a threshold percentage of events associated with a user associated with the persona;

populating the plurality of personas with the generated data associated with the second online interactions;

selecting a plurality of the received census data associated with the first online interactions that match the generated data associated with the second online interactions for each of the plurality of personas;

replacing the generated data associated with the second online interactions populated in each of the plurality of personas with the selected data; and

outputting the plurality of personas with the replaced data.

Assignments (5)

Continuity (2)

References Cited (10)