BROWSER-BASED SELECTION OF CONTENT REQUEST MODES

• US 20150154506A1
• Filed: 12/02/2013
• Published: 06/04/2015
• Est. Priority Date: 12/02/2013
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Jain, Saral, Vaswani, Ameet Nirmal, Trahan, Matthew Lloyd
• Granted Patent: US 10,242,322 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: H04L 67/01 Protocols
• CPC Class Codes: H04L 67/564 Enhancement of application control based on intercepted application data
• CPC Class Codes: H04L 67/566 Grouping or aggregating service requests, e.g. for unified processing

Abstract

Features are disclosed for generating request decision models for use by client computing devices to determine request paths or modes for content requests. The request modes may correspond to direct requests (e.g., requests made from a client device directly to a content server hosting requested content) or to indirect requests (e.g., requests made from the client device to the content server via an intermediary system). The request decision models may be trained by a machine learning algorithm using performance data (e.g., prior content load times), contextual information (e.g., state information associated with devices at times content requests are executed), and the like.

Claims

1. A system comprising:

one or more computing devices programmed to at least;

obtain performance data reflecting load times for a plurality of content requests, wherein at least a first portion of the content requests were made from a first group of client devices directly to content servers, and wherein at least a second portion of the content requests were made from a second group of client devices to an intermediary system;

obtain contextual information regarding a state of the intermediary system and at least a portion of the first or second groups of client devices at times corresponding to individual requests of the plurality of content requests; and

generate a request decision model by using a machine learning algorithm to process the performance data and the contextual information, wherein the request decision model is configured to determine whether a content request is to be made directly to a content server or indirectly to an intermediary system based at least partly on a current state of a client device making the particular content request.

2. The system of claim 1, wherein the one or more computing devices are further programmed to provide the request decision model to a plurality of client devices.

3. The system of claim 1, wherein the request decision model is further configured to determine the preferred request mode based at least partly on a current state of the intermediary system.

4. The system of claim 1, wherein the request decision model comprises a classification model or a regression model.

5. The system of claim 1, wherein the one or more server computing devices are further programmed to infer labels for at least a portion of the contextual information based at least partly on the corresponding content load times.

6. A computer-implemented method comprising:

as implemented by a server system comprising one or more computing devices, generating model training data based at least partly on;

request performance data regarding content load times for a plurality of content requests; and

contextual data regarding computing device states at times corresponding to individual requests of the plurality of content requests; and

generating a request decision model using the model training data, wherein the request decision model is configured to determine a preferred request mode for a particular content request.

7. The computer-implemented method of claim 6, wherein generating the request decision model comprises using a machine learning algorithm.

8. The computer-implemented method of claim 6, wherein the request decision model comprises a classification model or a regression model.

9. The computer-implemented method of claim 6, further comprising providing the request decision model to a plurality of client devices.

10. The computer-implemented method of claim 6, further comprising determining labels for at least a portion of the contextual data based at least partly on the corresponding content load times.

11. The computer-implemented method of claim 6, further comprising determining labels for at least a portion of the contextual data based at least party on a comparison of content load times associated with two or more requests made by a single client device.

12. The computer-implemented method of claim 6, wherein the preferred request mode corresponds to one of a direct request to a content server, or an indirect request via an intermediary system.

13. The computer-implemented method of claim 12, wherein the request decision model is further configured to determine whether the content request is to be made indirectly to the intermediary system based at least partly on a current state of the intermediary system.

14. The computer-implemented method of claim 12, wherein the request decision model is further configured to determine whether the content request is to be made indirectly to the intermediary system based at least partly on a current state of a client device.

15. The computer-implemented method of claim 6, further comprising:

receiving, from a client device, additional request performance data regarding content load times; and

generating an updated request decision model based at least partly on the additional request performance data.

16. Non-transitory computer storage having stored thereon a browser module configured to execute on a client computing device, the browser module configured to implement at least:

a first content request mode in which the client computing device retrieves content directly from an origin content server; and

a second content request mode in which the user device retrieves the content indirectly from an intermediary system that operates as an intermediary between client computing devices and content servers;

wherein said browser module is configured to use a decision model to select between the first and second content request modes for retrieving particular content items, said decision model based on aggregated content-loading performance data collected from a plurality of client computing devices for each of said content request modes.

17. The non-transitory computer storage of claim 16, wherein the browser module comprises one of:

an application;

a browser toolbar;

or a browser add-in.

18. The non-transitory computer storage of claim 16, wherein the browser module is further configured to select from the first content request mode and the second content request mode based at least party on a current state of the client computing device.

19. The non-transitory computer storage of claim 16, wherein the browser module is further configured to select from the first content request and mode the second content request mode based at least party on a current state of the intermediary system.

20. The non-transitory computer storage of claim 16, the process further comprising transmitting, to the intermediary system, request performance data regarding a content load time associated with one or more content requests.

21. The non-transitory computer storage of claim 16, wherein the decision model comprises a classification model or a regression model.

22. The non-transitory computer storage of claim 16, the process further comprising:

receiving an updated decision model from the intermediary system; and

using the updated decision module to select one of the first or second content request modes for a content request.

23. A system comprising:

an intermediary system comprising at least one server machine, said intermediary system configured to operate as an intermediary between user devices and origin content servers and to provide content pre-rendering services that offload content processing tasks from the user devices;

a browser module that runs on the user devices, said browser module configured to use a decision model to determine whether to retrieve a content item directly from an origin content server versus indirectly via the intermediary system, said decision model capable of predicting whether use of the intermediary system to retrieve the content item will improve content loading performance from an end user perspective; and

a decision model generator configured to use aggregated content-loading performance data to generate the decision model.