Strategies for providing diverse recommendations

• US 7,542,951 B1
• Filed: 10/31/2005
• Issued: 06/02/2009
• Est. Priority Date: 10/31/2005
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Chakrabarti, Kushal, Smith, Brent R., Kohavi, Ron
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US11/263,068
• Time in Patent Office: 1,310 Days
• Field of Search: 706/45
• US Class Current: 706/45
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

Strategies are described for generating recommendations. The strategies generate a set of original recommendations based on a source of information. The strategies then transform the set of original recommendations into a set of similarity-spaced recommendations based on “repulsion force” analysis applied to the set of original recommendations. In a first implementation, the set of spaced recommendations represent a diverse sampling of items in the set of original recommendations. In a second implementation, the set of spaced recommendations represent a sampling of items in the set of original recommendations which omits or excludes recommendations assessed as obvious. A third implementation can combine the first and second implementations.

Claims

1. A method for generating diverse recommendations, the method implemented at least in part by a computing device and comprising:

providing at least one source of information from which to generate recommendations;

generating a set of original recommendations based on said at least one source of information;

generating a set of diverse recommendations from the set of original recommendations based on repulsion force analysis applied to the set of original recommendations, wherein the repulsion force analysis ensures that at least one recommendation is separated from a reference by at least a sufficient amount to provide a desired degree of variation between said at least one recommendation and the reference; and

providing the set of diverse recommendations to a user,

wherein the reference is another recommendation, such that the repulsion force analysis ensures that said at least one recommendation is separated from the other recommendation by the sufficient amount.

2. The method of claim 1, wherein said at least one source of information comprises information regarding the past behavior of one or more users.

3. The method of claim 1, wherein the generating of the set of original recommendations comprises generating original recommendations which exhibit, based on attraction force analysis, at least a sufficient degree of relatedness to said at least one source of information.

4. The method of claim 1, wherein the separation of said at least one recommendation from the reference is based on a distance metric that defines the similarity between said at least one recommendation and the reference.

5. The method of claim 4, wherein the distance metric is a measure of relatedness within a subject matter hierarchy.

6. The method of claim 1, wherein the generating of the set of diverse recommendations comprises:

identifying clusters of recommendations based on the set of original recommendations, to provide identified clusters; and

selecting representative members from each of the identified clusters, the representative members being used to form the set of diverse recommendations.

7. The method of claim 6, wherein the identifying of the clusters comprises:

ranking the original recommendations in the set of original recommendations;

for each original recommendation under consideration;

computing a spacing group containing members that are determined to be related to the original recommendation under consideration;

determining whether any members in the spacing group match lower-ranking original recommendations in the set of original recommendations, to provide identified lower-ranking original recommendations; and

eliminating any identified lower-ranking original recommendations in the set of original recommendations.

8. The method of claim 1, further comprising applying spacing analysis to said at least one source of information to generate modified source information, and wherein the generating of the recommendations is applied to the modified source information.

9. The method of claim 1, wherein the diverse set of recommendations comprises a subset of the set of the original recommendations.

10. The method of claim 1, wherein the diverse set of recommendations comprises a set of recommendations including at least one recommendation that is not present in the set of original recommendations.

11. The method of claim 1, wherein the diverse set of recommendations comprises a reordering of items in the set of original recommendations.

12. The method of claim 1, wherein the set of diverse recommendations pertain to items that can be acquired by the user.

13. The method of claim 1, wherein the set of diverse recommendations pertain to search results generated in response to a search query made by the user.

14. The method of claim 1, wherein the generating of the set of original recommendations and the generating of the set of diverse recommendations comprise two operations performed in temporal succession.

15. The method of claim 1, wherein the generating of the set of original recommendations and the generating of the set of diverse recommendations comprise a single integrated operation.

16. The method of claim 1, wherein the repulsion force analysis reduces an amount of configuration energy in the set of original recommendations.

17. The method of claim 16, wherein the repulsion force analysis approximately minimizes the configuration energy.

18. One or more machine-readable media containing machine readable instructions for implementing the method of claim 1.

19. A system comprising:

one or more processors;

memory; and

a recommendation-generating module, stored in the memory and executable on the one or more processors, the recommendation-generating module comprising;

logic configured to generate or receive a set of original recommendations, the original recommendations based on at least one source of information;

logic configured to generate a set of diverse recommendations from the set of original recommendations based on repulsion force analysis applied to the set of original recommendations, wherein the repulsion force analysis ensures that at least one recommendation is separated from a reference by at least a sufficient amount to provide a desired degree of variation between said at least one recommendation and the reference; and

logic configured to provide the set of diverse recommendations to a user,

wherein the reference is another recommendation, such that the repulsion force analysis ensures that said at least one recommendation is separated from the other recommendation by the sufficient amount.

20. One or more machine-readable media containing machine readable instructions for implementing the recommendation-generating module of claim 19.

21. A method for generating diverse recommendations, the method implemented at least in part by a computing device and comprising:

receiving a set of original recommendations which have been generated based on at least one source of information;

generating a set of diverse recommendations from the set of original recommendations based on repulsion force analysis applied to the set of original recommendations, wherein the repulsion force analysis ensures that at least one recommendation is separated from a reference by at least a sufficient amount to provide a desired degree of variation between said at least one recommendation and the reference; and

providing the set of diverse recommendations to a user,

wherein the reference is another recommendation, such that the repulsion force analysis ensures that said at least one recommendation is separated from the other recommendation by the sufficient amount.

22. One or more machine-readable media containing machine readable instructions for implementing the method of claim 21.

23. A method for generating diverse recommendations, the method implemented at least in part by a computing device and comprising:

receiving a set of original recommendations which have been generated based on at least one source of information;

generating a set of diverse recommendations which reduce a configuration energy in the set of original recommendations by;

identifying clusters of original recommendations in the set of original recommendations, to provide identified clusters; and

selecting representative members from each of the identified clusters, the representative members being used to form the set of diverse recommendations; and

providing the set of diverse recommendations to a user.

24. One or more machine-readable media containing machine readable instructions for implementing the method of claim 23.

25. A method for generating diverse recommendations, the method implemented at least in part by a computing device and comprising:

receiving a set of original recommendations which have been generated based on at least one source of information;

generating a set of diverse recommendations from the set of original recommendations through the use of cluster analysis, wherein the generating ensures that at least one recommendation is separated from a reference by at least a sufficient amount to provide a desired degree of variation between said at least one recommendation and the reference; and

providing the set of diverse recommendations to a user,

wherein the reference is another recommendation, such that the cluster analysis ensures that said at least one recommendation is separated from the other recommendation by the sufficient amount.

26. One or more machine-readable media containing machine readable instructions for implementing the method of claim 25.