Dynamic Hierarchical Empirical Bayes and digital content control

Patent metadata

Field	Value
Publication number	US-10956930-B2
Application number	US-201816034232-A
Country	US
Kind code	B2
Filing date	Jul 12, 2018
Priority date	Jul 12, 2018
Publication date	Mar 23, 2021
Grant date	Mar 23, 2021

Abstract

Dynamic Hierarchical Empirical Bayes techniques and systems are described that are implemented to control output of digital content. In one example, a system identifies splitting variables included in data. An amount of loss is then determined for each of the identified splitting variables by the system using a loss function. Based on the determined amounts of loss, the system selects at least one splitting variable from the plurality of splitting variables that are to be used to partition data in a respective node, e.g., a parent node to form a plurality of child nodes. The system, for instance, may select the splitting variable that minimizes the cost, i.e., has the lowest amount of cost. The selected splitting variable is then employed by the system to generate at least one hierarchical level of the hierarchical structure of the statistical model by partitioning data from the parent node into respective child nodes.

First claim

What is claimed is:

1. In a digital medium environment configured to control output of digital content, a method implemented by at least one computing device, the method comprising:
collecting, by the at least one computing device, a first set of historical data that describes user interaction with digital content;
generating, by the at least one computing device, a hierarchical structure of a statistical model from the first set of historical data, the generating including:
identifying a plurality of splitting variables from the first set of data;
determining an amount of loss for each of the plurality of splitting variables using a hierarchical shrinkage loss function that includes a first term that measures weighted information loss within a respective child node within the hierarchical structure and a second term that incorporates data within the respective child node and a respective parent node within the hierarchical structure;
selecting at least one splitting variable from the plurality of splitting variables based on the determining; and
generating at least one hierarchical level of the hierarchical structure having a plurality of nodes that are partitioned based on the selected splitting variable;

collecting, by the at least one computing device, a second set of data that describes user interaction with digital content, the second set of data collected subsequent to the first set of historical data;
generating, by the at least one computing device, a prediction based on a performance metric by applying the generated statistical model to the second set of data; and
controlling, by the at least one computing device, digital content output based on the prediction.

2. The method as described in claim 1 , wherein the selecting is based on which of the plurality of splitting variables exhibits a minimum amount of loss based on the determining.

3. The method as described in claim 1 , wherein the controlling includes generating a prediction of subsequent user interaction with the digital content through use of the generated hierarchical structure and the received second set of data, the prediction is used to control which item of digital content of a plurality of digital content is to be output.

4. The method as described in claim 3 , wherein the second set of data describes user interaction with the digital content over a pervious timeframe and the prediction is generated for a subsequent timeframe.

5. The method as described in claim 1 , wherein the statistical model is a Hierarchical Bayes statistical model including a hierarchical structure having parent and child nodes.

6. The method as described in claim 1 , wherein the at least one hierarchical level is an intermediate hierarchical level between a root node and a decision unit hierarchical level.

7. The method as described in claim 1 , wherein the hierarchical shrinkage loss function includes a regularization term that incorporates a prior distribution from the respective parent node in the hierarchical structure.

8. The method as described in claim 1 , wherein the controlling includes controlling computational resources of the at least one computing device as part of controlling the digital content output.

9. In a digital medium environment, a system comprising:
a data collection module implemented at least partially in hardware of a computing device to collect a first set of historical data describing user interaction with advertisements and a second set of data collected subsequent to the first set of historical data, the second set of data describing user interaction with advertisements;
a model training module implemented at least partially in hardware of the computing device to generate a hierarchical structure of a statistical model from the first set of historical data, the model training module including:
a splitting variable identification module to identify a plurality of splitting variables from the first set of historical data;
a loss determination module to determine an amount of loss for each of the plurality of splitting variables using a hierarchical shrinkage loss function that includes a first term that measures weighted information loss within a respective child node within the hierarchical structure and a second term that incorporates data within the respective child node and a respective parent node in the hierarchical structure;
a variable selection module to select at least one splitting variable from the plurality of splitting variables based on the determined amount of loss; and
a partition module to generate at least one hierarchical level of the hierarchical structure having a plurality of nodes that are partitioned based on the selected splitting variable; and

a prediction generation module to control advertisement output based on the statistical model as applied to the second set of data.

10. The system as described in claim 9 , wherein the variable selection module is configured to select the at least one splitting variable based on which of the plurality of splitting variables exhibits a minimum amount of loss based on the determining.

11. The system as described in claim 9 , wherein the prediction generation module is configured to control advertisement output by:
receiving the second set of data as describing user interaction with the advertisements; and
generating a prediction of subsequent user interaction with the advertisements through use of the generated hierarchical structure and the received second set of data, and wherein the prediction is used to control which advertisement of the a plurality of advertisements is to be output.

12. The system as described in claim 11 , wherein the second set of data describes user interaction with the advertisements over a pervious timeframe and the prediction is generated for a subsequent timeframe.

13. The system as described in claim 9 , wherein the statistical model is a Hierarchical Bayes statistical model.

14. The system as described in claim 9 , wherein the at least one hierarchical level is an intermediate hierarchical level between a root node and a decision unit hierarchical level.

15. The system as described in claim 9 , wherein the hierarchical shrinkage loss function includes a regularization term that incorporates a prior distribution from the respective parent node in the hierarchical structure.

16. The system as described in claim 9 , wherein the hierarchical shrinkage loss function measures within-node loss regarding the respective child node and loss between the respective child node and the respective parent node.

17. The system as described in claim 9 , wherein the prediction generation module is configured to control computational resources as part of controlling the advertisement output.

18. In a digital medium environment, a system comprising:
means for collecting first and second sets of data;
means for generating a hierarchical structure of a hierarchical Bayes statistical model from the first set of data, the generating means including:
means for identifying a plurality of splitting variables from the data;
means for determining an amount of loss for each of the plurality of splitting variables using a hierarchical shrinkage loss function that includes a first term that measures weighted information loss within a respective child node within the hierarchical structure and a second term that incorporates data within the respective child node and a respective parent node within the hierarchical structure;
means for selecting at least one splitting variable from the plurality of sp

Assignees

• Adobe Inc

Inventors

• Yuan Yuan
• Yan Zhenyu
• Sun Yiwen
• Dong Xiaojing
• Dong Chen
• Pani Abhishek

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• H04L67/535

  Tracking the activity of the user (network monitoring arrangements H04L43/00; recording of computer activity G06F11/34) · CPC title

• G06N20/00

  Machine learning · CPC title

• G06Q30/0247 Primary

  Calculate past, present or future revenues · CPC title