Systems and methods for generating a geo-level hierarchical Bayesian model

Patent metadata

Field	Value
Publication number	US-10706191-B2
Application number	US-201715693326-A
Country	US
Kind code	B2
Filing date	Aug 31, 2017
Priority date	Aug 31, 2017
Publication date	Jul 7, 2020
Grant date	Jul 7, 2020

Abstract

Systems, methods, and computer-readable storage media that may be used to generate a Bayesian hierarchical model. One method includes generating a plurality of geographic regions by grouping one or more geographic sub-regions into each of the plurality of geographic regions. The method further includes receiving data for the geographic sub-regions, the data including responses, content inputs, content types, and location identifiers. The method further includes generating geo-level data from the received data by grouping the responses and content inputs of the received data based on a correlation of the location identifiers of the received data to the plurality of geographic regions. The method includes fitting a Bayesian hierarchical model based on at least the geo-level data, the content types, and the geographic regions and determining a content input mix for the content types for each geographic region based on the Bayesian hierarchical model and a content input constraint.

First claim

What is claimed is:

1. A method comprising:
generating a plurality of geographic regions by grouping one or more geographic sub-regions into each of the plurality of geographic regions, generating the plurality of geographic regions comprises:
generating a first geographic region by grouping a first set of the geographic sub-regions; and
generating a second geographic region by grouping a second set of the geographic sub-regions;

receiving data for the geographic sub-regions, the data comprising responses, content inputs, content types, and location identifiers;
generating geo-level data from the received data by grouping the responses and content inputs of the received data based on a correlation of the location identifiers of the received data to the plurality of geographic regions;
fitting a Bayesian hierarchical model based on at least the geo-level data, the content types, and the geographic regions, wherein the Bayesian hierarchical model comprises a posterior distribution for response for each content type of the plurality of content types in each geographic region of the plurality of geographic regions; and
determining a content input mix for the content types for each geographic region based on the Bayesian hierarchical model and a content input constraint, the content input mix indicating a particular content input for each of the plurality of content types in each of the plurality of geographic regions, the content input constraint indicating a total content input for each of the geographic regions, determining the content input mix for the content types for each geographic region comprises:
determining a first content input mix for the content types for the first geographic region; and
determining a second content input mix different than the first content input mix for the content types for the second geographic region.

2. The method of claim 1 , wherein determining the content input mix for the content types
for each geographic region comprises maximizing a posterior mean of a predicted response of the Bayesian hierarchical model; and
wherein the content input mix is determined for a particular period of time.

3. The method of claim 1 , wherein fitting the Bayesian hierarchical model comprises generating posteriors for one or more parameters of the Bayesian hierarchical model based on at least the geo-level data.

4. The method of claim 3 , wherein the posteriors of the one or more parameters of the Bayesian hierarchical model are generated based on prior distributions for each of the parameters; and
wherein the prior distributions are each based on one or more hyper-parameters.

5. The method of claim 4 , wherein fitting the Bayesian hierarchical model further comprises fitting the Bayesian hierarchical model based on the plurality of hyper-parameters and a hyper-prior for each of the plurality of hyper-parameters.

6. The method of claim 1 , the method further comprising:
standardizing the geo-level data for the generated geographic regions by standardizing the responses and the content inputs of the geo-level data based on each population of the generated geographic regions; and
normalizing the standardized geo-level data to be between zero and one.

7. The method of claim 1 , wherein fitting the Bayesian hierarchical model with the geo-level data further comprises fitting the Bayesian hierarchical model based on a plurality of control variables and control variable types, the plurality of control variables and control variable types indicating conditions for each of the geographic regions at one or more points in time;
wherein the method further comprising standardizing the plurality of control variables based on a population of each geographic region.

8. The method of claim 1 , wherein the Bayesian hierarchical model comprises one or more functions that models carryover, lag, and saturation effects, wherein the one or more functions comprise an adstock function and a Hill function.

9. A system comprising:
at least one computing device operably coupled to at least one memory and configured to:
generate a plurality of geographic regions by grouping one or more geographic sub-regions into each of the plurality of geographic regions by:
generating a first geographic region by grouping a first set of the geographic sub-regions; and
generating a second geographic region by grouping a second set of the geographic sub-regions;

receive data for the geographic sub-regions, the data comprising responses, content inputs, content types, and location identifiers;
generate geo-level data from the received data by grouping the responses and content inputs of the received data based on a correlation of the location identifiers of the received data to the plurality of geographic regions;
fit a Bayesian hierarchical model based on at least the geo-level data, the content types, and the geographic regions wherein the Bayesian hierarchical model comprises a posterior distribution for response for each content type of the plurality of content types in each geographic region of the plurality of geographic regions; and
determine a content input mix for the content types for each geographic region based on the Bayesian hierarchical model and a content input constraint, the content input mix indicating a particular content input for each of the plurality of content types in each of the plurality of geographic regions, the content input constraint indicating a total content input for each of the geographic regions, determining the content input mix for the content types for each geographic region comprises:
determining a first content input mix for the content types for the first geographic region; and
determining a second content input mix different than the first content input mix for the content types for the second geographic region.

10. The system of claim 9 , wherein the computing device is configured to determine the content input mix for the content types for each geographic region comprises maximizing a posterior mean of a predicted response of the Bayesian hierarchical model; and
wherein the content input mix is determined for a particular period of time.

11. The system of claim 10 , wherein the computing device is configured to standardize the geo-level data for the generated geographic regions by standardizing the responses and the content inputs of the geo-level data based on each population of the generated geographic regions.

12. The system of claim 10 , wherein the computing device is configured to:
generate the Bayesian hierarchical model with the geo-level data further comprises fitting the Bayesian hierarchical model based on a plurality of control variables, the plurality of control variables indicating conditions for each of the geographic regions; and standardize the plurality of control variables based on a population of each geographic region.

13. The system of claim 9 , wherein the computing device is configured to fit the Bayesian hierarchical model to generate posteriors for one or more parameters of the Bayesian hierarchical model.

14. The system of claim 13 , wherein the posteriors of the one or more parameters of the Bayesian hierarchical model are generated based on prior distributions for each of the parameters; and
wherein the prior distributions are each based on one or more hyper-parameters.

15. The system of claim 14 , wherein the computing device is configured to fit the Bayesian hierarchical model by fitting the Bayesian hierarchical model based on the one or more hyper-parameters and a hyper-prior for each of the hyper-parameters.

16.

Assignees

• Google Inc
• Google Llc

Inventors

• Sun Yunting
• Jin Yuxue
• Koehler James
• Huang Xiaojing
• Chan David
• Wang Yueqing
• Sontag Conor
• Zhong Shi
• Perez Luis Gonzalez

Classifications

• G06F17/18 Primary

  for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)} · CPC title

• G06F2111/10

  Numerical modelling · CPC title

• G06T7/11

  Region-based segmentation · CPC title

• G06Q30/02

  Marketing; Price estimation or determination; Fundraising · CPC title

• H04N21/25

  Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies {(maintenance or administration in data networks H04L41/00)} · CPC title