Adaptive learning methods for selecting page components to include on dynamically generated pages

• US 9,390,186 B2
• Filed: 07/01/2015
• Issued: 07/12/2016
• Est. Priority Date: 03/19/2002
• Status: Expired due to Term

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Dicker, Russell A., Cottrell, Robert G., York, Jeremy C.
• Primary Examiner(s): Bilgrami, Asghar
• Application Number:
  US14/789,895
• Publication Number:
  US 20150302105A1
• Time in Patent Office: 377 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/955 using information identifiers, e.g. uniform resource locators [URL]
• CPC Class Codes: G06F 16/958 Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 30/0633 Lists, e.g. purchase orders, compilation or processing

Abstract

A subset of a set of components is selected for inclusion on a dynamically-generated page within a particular user context. The subset of components is selected based on scores associated with the components in the set. The score of a component is preferably determined based on measurements of user activity resulting from exposures of the components to users in the particular context.

Claims

1. A computer process for adaptively selecting components to include on dynamically generated electronic pages, the process comprising:

collecting exposure data for exposure events in which a component is included on dynamically generated electronic pages served to requesting devices, the exposure data including positional values reflecting display positions at which the component is included on the dynamically generated pages;

collecting activity data reflective of user interactions with the component as included on the dynamically generated electronic pages, the activity data based on monitored interactions of a plurality of users;

calculating a score for the component based on at least (1) the collected exposure data, including the positional values, and (2) the collected activity data, the score representing an effectiveness of the component;

receiving a page request from a user device; and

generating an electronic page to provide to the user device in response to the page request, wherein generating the electronic page comprises determining, based at least partly on the score, whether to include the component on the electronic page;

said process performed by a computing system under control of executable program instructions.

2. The process of claim 1, wherein the positional values are selected such that more exposure credit is given to exposure events in which the component is positioned near the top of a page than to exposure events in which the component is positioned near the bottom of a page.

3. The process of claim 1, wherein the process comprises:

collecting the exposure data and activity data separately for each of a plurality of defined contexts in which the component is capable of being displayed; and

calculating a respective context-specific score for each of the plurality of contexts based on the exposure and activity data corresponding to the respective context.

4. The process of claim 3, wherein generating the electronic page comprises determining a context of the page request, said context being dependent upon a user attribute associated with the user device, and looking up the context-specific score corresponding to the determined context.

5. The process of claim 4, wherein the context is additionally dependent upon a past browsing history of a user of the user device.

6. The process of claim 1, wherein the component is a dynamically generated component that displays content that depends upon a context in which the component is displayed.

7. The process of claim 1, wherein generating the electronic page comprises using a random selection algorithm in which a likelihood that the page component is selected for inclusion on the electronic page is dependent on the score.

8. A system for selecting components to include on dynamically generated pages, the system comprising:

a plurality of page components represented in computer storage;

a data repository of exposure data reflective of exposure events in which the page components are selected for inclusion on dynamically generated pages, the exposure data including positional values reflective of positions at which the components are included on the dynamically generated pages;

a data repository of activity data reflective of user interactions with the page components as included on the pages, the activity data based on monitored interactions of a plurality of users; and

a computer system programmed with executable instructions to calculate, for each page component, a score representing an effectiveness of the respective page component, the scores based on the exposure data, including the positional values, and based additionally on the activity data.

9. The system of claim 8, further comprising a component selection module that uses the scores to dynamically select the page components to include on pages in response to requests from user devices.

10. The system of claim 9, wherein the component selection module uses a random selection algorithm in which a likelihood that a page component will be selected for inclusion on a requested page depends on the score corresponding to the component.

11. The system of claim 8, wherein the exposure data and activity data are stored separately for each of a plurality of defined contexts in which page components can be displayed, and wherein the computer system is programmed to calculate, for a page component, a plurality of context-specific scores, each of which represents an effectiveness of the page component within a respective context.

12. The system of claim 8, further comprising a plurality of context definitions stored in computer storage, each context definition defining a different respective one of the plurality of contexts, wherein at least some of the context definitions include user attribute values.

13. The system of claim 8, wherein at least one of the page components displays content that is dependent upon a context in which the component is displayed.

14. A computer-implemented method of selecting components to include on dynamically generated pages, the method comprising:

receiving, from a user computing device of a user, a request for a dynamically generated page;

mapping the request to a context defined in computer storage, said context being one of a plurality of defined contexts in which the page is capable of being requested;

for each of a plurality of candidate components, determining a respective context-specific score that reflects an effectiveness of the candidate component in the context, wherein each score is dependent upon responses of prior users to presentation of the respective component in the context;

selecting, based at least in part on the scores, a subset of said plurality of candidate components to present to the user on the page; and

dynamically generating the page in response to the request, wherein dynamically generating the page comprises incorporating the selected subset of components into the page;

the method performed programmatically by a computing system that comprises a server.

15. The method of claim 14, wherein mapping the request to a context comprises selecting the context based at least partly on an attribute of the user.

16. The method of claim 15, wherein mapping the request to a context comprises selecting the context based additionally on a state of a browsing session of the user.

17. The method of claim 14, wherein the scores dependent upon display positions at which the components were presented to the prior users on pages.

18. The method of claim 14, wherein the component displays dynamically selected content that depends on the context, and the method further comprises dynamically selecting the content in response to the request.

19. The method of claim 14, wherein the subset of candidate components is selected based additionally on a randomization function.