Strategies for providing novel recommendations

• US 7,584,159 B1
• Filed: 10/31/2005
• Issued: 09/01/2009
• 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: Smith, Brent R., Kohavi, Ron, Chakrabarti, Kushal
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US11/263,563
• Time in Patent Office: 1,401 Days
• Field of Search: 706/45
• US Class Current: 706/45
• 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 novel 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 at least in part on said at least one source of information;

generating a set of novel recommendations from the set of original recommendations based at least in part 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 novel recommendations to a user;

wherein the reference is known information that identifies at least one known interest of the user, such that the repulsion force analysis ensures that said at least one recommendation is separated from the known information 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 at least in part 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 at least in part 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 known information pertains to the user'"'"'s behavior with respect to one or more items.

7. The method of claim 6, wherein the known information comprises one or more of:

information pertaining to the user'"'"'s selection of an item;

information pertaining to the user'"'"'s purchase of the item;

information pertaining to the user'"'"'ranking of the item;

or

information pertaining to express preference-related information entered by the user with respect to the item.

8. The method of claim 1, wherein the generating of the set of novel recommendations comprises:

identifying obvious original recommendations in the set of original recommendations; and

reducing the concentration of the obvious original recommendations in the set of original recommendations, to yield the set the novel recommendations.

9. The method of claim 8, wherein the identifying of the obvious original recommendations comprises:

determining known items concerning which the user is presumed to have prior knowledge;

for each known item under consideration;

comprising a spacing group containing members that are determined to be related to the known item under consideration;

determining whether any members in the spacing group match any original recommendations in the set of original recommendations, to yield the obvious original recommendations.

10. 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.

11. The method of claim 1, wherein the novel set of recommendations comprises a subset of the set of the original recommendations.

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

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

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

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

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

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

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 a set of original recommendations based at least in part on at least one source of information;

logic configured to generate a set of novel recommendations from the set of original recommendations based at least in part 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 novel recommendations to a user,

wherein the reference is known information that identifies at least one known interest of the user, such that the repulsion force analysis ensures that said at least one recommendation is separated from the known information 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 novel 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 at least in part on at least one source of information;

generating a set of novel recommendations from the set of original recommendations based at least in part 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 novel recommendations to a user;

wherein the reference is known information that identifies at least one known interest of the user, such that the repulsion force analysis ensures that said at least one recommendation is separated from the known information 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 novel 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 at least in part on at least one source of information;

generating a set of novel recommendations which improve novelty in the set of original recommendations by;

determining known items concerning which the user is presumed to have prior knowledge;

expanding the known items to include a collection of items which are determined to be related to the known items, to provide an expanded set of items;

identifying whether any members of the expanded set of items are present in the set of original recommendations to define obvious original recommendations; and

reducing he concentration of the obvious original recommendations in the set of original recommendations, to yield the set the novel recommendations; and

providing the set of novel 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 novel 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 at least in part on at least one source of information;

generating a set of novel recommendations from the set of original recommendations through the use of cluster analysis, wherein the generating ensures that at lest 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 novel recommendations to a user,

wherein the reference is known information that identifies at least one known interest of the user, such that the cluster analysis ensures that said at least one recommendation is separated from the known information by the sufficient amount.

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

27. A method comprising:

under control of one or more computer systems configured with executable instructions,

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

generating a first set of recommendations based at least in part on said at least one source of information;

generating a second set of recommendations from the first set of recommendations based at least in part on repulsion force analysis applied to the first set of 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 second set of recommendations to a user,

wherein the reference is known information that identifies at least one known interest of the user, such that the repulsion force analysis ensures that said at least one recommendation is separated from the known information by the sufficient amount.

28. The method of claim 27, wherein the second set of recommendations comprises a set of novel or unique recommendations.

29. The method of claim 27, wherein the generating of the second set of recommendations comprises:

determining known items of which the user is presumed to have prior knowledge; and

removing at least one determined known item from the first set of recommendations to at least partly define the second set of recommendations.