System for obtaining recommendations from multiple recommenders

• US 7,991,757 B2
• Filed: 08/12/2008
• Issued: 08/02/2011
• Est. Priority Date: 08/12/2008
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Kane, Francis J. Jr.
• Primary Examiner(s): Wong; Don
• Assistant Examiner(s): Nguyen; Kim T
• Application Number:
  US12/190,471
• Publication Number:
  US 20100042608A1
• Time in Patent Office: 1,085 Days
• Field of Search: 707/706, 707/748
• US Class Current: 707/706
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation

Abstract

A personalization network service enables developers to develop recommenders that can be made available to content site operators for providing recommendations to end users. The personalization network service may also be capable of optimizing the use and selection of the recommenders for different end users, groups or segments of end users, content sites, and the like.

Claims

1. A system for providing a network service to one or more content sites, the system comprising:

a computer system comprising computer hardware programmed to implement at least;

a network service interface operative to;

receive a first request from a content site to store event data descriptive of user-generated events, the event data comprising identifiers of particular items and particular users to which the events correspond, the first request being received over a computer network; and

receive a second request from the content site for item recommendations, the second request being received over the computer network;

an event data repository operative to store the event data received from the content site, the event data repository comprising physical computer storage;

a registration module configured to provide functionality for registering an external recommender with a recommendations service;

a data interface operative to provide access to the event data to a plurality of external recommenders registered with the recommendations service, the plurality of external recommenders being provided by one or more entities that differ from a provider of the recommendations service; and

the recommendations service comprising a recommendations retriever operative to, in response to the second request;

query the plurality of external recommenders over the computer network for recommendations responsive to the second request, and

receive recommendations and associated scores from the external recommenders over the computer network, at least some of the recommendations being generated based at least in part on the event data stored in the event data repository; and

an optimization module configured to normalize the associated scores;

wherein the recommendations service is further configured to select a subset of the recommendations, based at least in part on the normalized scores, to provide to the content site.

2. The system of claim 1, wherein the optimization module is further configured to optimize the recommendations based at least in part on user-generated events responsive to previous recommendations results and a number of previous content site requests served by the external recommenders.

3. The system of claim 2, wherein the optimization module is further operative to assign weights to the external recommenders.

26. The system of claim 1, wherein the computer system comprises multiple physical computing devices.

27. The system of claim 1, wherein the registration module is further configured to allow a provider of an external recommender associated with the plurality of external recommenders to provide machine readable descriptions of the external recommender.

28. The system of claim 3, wherein the content site selects at least some of the weights to assign to the external recommenders.

4. A computer-implemented method of optimizing recommender results, the method comprising:

by a computer system comprising computer hardware;

receiving a request from a content site for item recommendations, the request being received with a personalization network service;

in response to the request from the content site, requesting recommendations from a plurality of recommenders, wherein the plurality of recommenders comprise external recommenders that are operated by different operators than an operator of the personalization network service;

receiving recommendations and corresponding scores from the recommenders;

applying weights to each of the recommenders;

combining the scores with the weights to produce overall scores;

selecting a subset of the recommendations based on the overall scores to provide to the content site; and

adjusting the weights for each of the recommenders based at least in part on user-generated events responsive to the recommendations.

5. The method of claim 4, wherein applying the weights to each of the recommenders comprises applying the weights based at least in part on user-generated events responsive to previous recommendations results.

6. The method of claim 5, wherein the user-generated events comprise one or more of purchase events, item viewing events, and item rating events.

7. The method of claim 4, wherein applying the weights to each of the recommenders comprises applying site-specific weights to at least some of the recommenders.

8. The method of claim 4, wherein applying the weights to each of the recommenders comprises applying user-specific weights to at least some of the recommenders.

9. The method of claim 4, wherein applying the weights to each of the recommenders comprises applying user segment-specific weights to at least some of the recommenders.

10. The method of claim 4, wherein applying the weights to each of the recommenders comprises applying a default weight to one or more of the recommenders.

11. The method of claim 4, further comprising normalizing the scores provided by each of the recommenders to produce normalized scores.

12. The method of claim 11, wherein combining the scores with the weights to produce overall scores comprises combining the normalized scores with the weights to produce the overall scores.

13. The method of claim 4, wherein at least some of the recommenders are operated by recommender providers separate from an operator of the personalization network service.

14. A computer-implemented method of optimizing recommender results, the method comprising:

by a computer system comprising computer hardware;

receiving a request from a content site for item recommendations over a network;

requesting recommendations from a plurality of recommenders in response to the request from the content site, wherein the plurality of recommenders comprise external recommenders that are operated by different operators than an operator of the personalization network service;

receiving recommendations results from the recommenders;

adjusting the recommendations results based at least in part on one or more of user-generated events responsive to previous recommendations results and a number of previous content site requests served by the recommenders; and

selecting a subset of the adjusted recommendations results to provide to the content site.

15. The method of claim 14, wherein the user-generated events comprise one or more of purchase events, item viewing events, and item rating events.

16. The method of claim 14, wherein adjusting the recommendations results comprises applying weights to the recommendations results.

17. A computer-implemented method of providing a network service to content sites, the method comprising:

by a computer system comprising computer hardware;

receiving registration submissions for recommenders from a plurality of recommender providers, the registration submissions operative to request registration of the recommenders with a recommendations system, wherein the recommender providers are different entities than a provider of the recommendations system;

storing, in computer storage, access information for accessing the recommenders with the recommendations system;

receiving requests from one or more content sites for item recommendations, the requests being received over a computer network with a network service interface;

in response to the requests, using the recommendations system to cause the recommenders to provide recommendations for the one or more content sites;

normalizing scores provided by the recommenders to produce normalized scores;

applying weights to each of the recommenders; and

combining the normalized scores with the weights to generate overall scores to be used in selecting recommendations to provide to the one or more content sites.

18. The method of claim 17, wherein the information for accessing the recommenders with the recommendations system comprises uniform resource indicators.

19. The method of claim 17, further comprising receiving recommender selections from the one or more content sites.

20. The method of claim 17, wherein applying weights to each of the recommenders further comprises weighting the recommenders based on the performance of the recommenders.

21. The method of claim 20, further comprising receiving weighting selections from the one or more content sites.

22. A system for enabling site operators to obtain item recommendations for presentation to users, the system comprising:

a computer data repository that stores information regarding a plurality of recommenders, including network addresses for accessing the recommenders on a network, each recommender provided by a respective recommender provider, at least some of the recommenders provided by different recommendation providers than others, the recommenders capable of using collected event data regarding user-generated events to select particular items to recommend to users; and

a server system comprising computer hardware operative to provide a service through which content sites obtain recommendations from the plurality of recommenders for presentation to users, the server system responsive to a request from a content site for recommendations by using the network addresses to request recommendations from the recommenders, and by combining resulting recommendation sets returned by the recommenders to generate a combined set of recommendations to provide to the content site;

wherein said combining comprises;

normalizing scores provided by the recommenders to produce normalized scores,

applying weights to each of the recommenders, and

combining the normalized scores with the weights to generate overall scores to be used in selecting the combined set of recommendations.

23. The system of claim 22, wherein the server system is operative to monitor context-specific performance of each recommender in each of a plurality of contexts, and to use collected data regarding the context-specific performance to determine, for each of the contexts, how much weight to give to particular recommenders in combining the recommendation sets returned by the recommenders.

24. The system of claim 23, wherein the context-specific performance of each recommender comprises one or more of user-specific performance, segment-specific performance, and site-specific performance.

25. The system of claim 22, wherein the server system provides an interface that enables an operator of a content site to specify which of the plurality of recommenders are to be used to generate recommendations for the content site.