Recommender system

• US 10,438,268 B2
• Filed: 02/09/2012
• Issued: 10/08/2019
• Est. Priority Date: 02/09/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Nice, Nir, Koenigstein, Noam, Paquet, Ulrich
• Primary Examiner(s): Palavecino, Kathleen
• Assistant Examiner(s): Cai, Lance Y
• Application Number:
  US13/369,318
• Publication Number:
  US 20130211950A1
• Time in Patent Office: 2,798 Days
• Field of Search: None
• CPC Class Codes: G06Q 30/0631 Item recommendations

Abstract

Embodiments of the invention provide methods and apparatus for recommending items from a catalog of items to a user by parsing the catalog of items into a plurality of catalog clusters of related items and recommending catalog items to the user from catalog clusters to which items previously preferred by the user belong.

Claims

1. A method of recommending items to a user, the method comprising:

clustering, by one or more processors, catalog items in a catalog into a plurality of catalog clusters of catalog items in a coordinate system responsive to preferences exhibited by a population of users, wherein the coordinate system comprises an origin point;

representing, by the one or more processors, each catalog item by a trait vector from the origin point;

determining, by the one or more processors, for each catalog cluster a catalog cluster trait vector, from the origin point, based on trait vectors of catalog items in the cluster;

determining, by the one or more processors, user legacy items belonging in each catalog cluster;

representing, by the one or more processors, each user legacy item by a trait vector from the origin point;

determining, by the one or more processors, for each catalog cluster a representative legacy trait vector, from the origin point, that is a function of and representative of the trait vectors of the user legacy items determined to belong in the catalog cluster;

recommending, by the one or more processors, a catalog item in a catalog cluster when an inner product of the trait vector, from the origin point, of the catalog item and the representative legacy trait vector, from the origin point, is greater than the inner product between the representative legacy trait vector, from the origin point, and the cluster trait vector, from the origin point; and

transmitting, by the one or more processors over a network, the recommended catalog item to a computing device for presentation of the recommended catalog item to a user.

2. A method according to claim 1, further comprising using a metric that provides a measure of distance between any two catalog clusters of the plurality of clusters used to recommend a catalog item to the user.

3. A method according to claim 2, wherein using the metric comprises determining distances between catalog clusters of the plurality of catalog clusters and identifying a catalog cluster of the plurality of catalog clusters to which a legacy item does not belong that has a determined distance less than a desired distance to at least one catalog cluster of the plurality of catalog clusters to which a legacy item does belong.

4. A method according to claim 3, further comprising recommending an item from the identified catalog cluster that does not comprise a legacy item.

5. A method according to claim 1, wherein clustering catalog items comprises acquiring rankings of the catalog items for a population of people and using the rankings to cluster the catalog items.

6. A method according to claim 5, wherein using the rankings comprises factoring a ranking matrix comprising the rankings to determine a trait vector to represent each catalog item.

7. A method according to claim 6, further comprising using the trait vectors to cluster the catalog items into the plurality of clusters.

8. A method according to claim 7, wherein using the trait vectors to cluster catalog items comprises clustering the trait vectors responsive to inner products of the trait vectors.

9. A method according to claim 8, wherein clustering responsive to the inner products comprises using a k-means algorithm.

10. A method according to claim 1, wherein clustering catalog items comprises using a collaborative filtering algorithm to cluster the catalog items.

11. A system, comprising:

a database having data responsive to preferences exhibited by a population of users;

one or more processors;

memory embodied with instructions executable by the one or more processors for;

generating a clustered database comprising a plurality of catalog clusters of the catalog items responsive to the data in the database in a coordinate system comprising an origin point;

identifying user legacy items previously preferred by a user;

determining identified user legacy items belonging in each catalog cluster;

representing each legacy item by a trait vector from the origin point;

determining for each catalog cluster a representative legacy trait vector, from the origin point, that is a function and representative of the trait vectors of the legacy item trait vectors in the catalog cluster;

determining for each catalog cluster a cluster trait vector, from the origin point, based on the trait vectors of catalog items in the cluster;

recommending a catalog item in a catalog cluster when an inner product of the trait vector of the catalog item and the representative legacy trait vector is greater than the inner product between the representative legacy trait vector and the cluster trait vector; and

transmitting, over a network, the recommended catalog item to a computing device for presentation of the recommended catalog item to a user.

12. The system according to claim 11, wherein the memory is further embodied with executable instructions for providing a representation for each catalog item that the cluster engine uses to generate the clustered database.

13. The system according to claim 12 wherein the memory is further embodied with executable instructions for modeling the catalog items as trait vectors in a latent space and generates a representation for each catalog item that comprises a trait vector.

14. The system according to claim 13, the trait vectors are used to generate the clustered database.

15. The system according to claim 11, wherein the catalog items are clustered using a collaborative filtering algorithm.

16. The system according to claim 11 wherein the recommender engine:

determines distances between catalog clusters of the plurality of clusters responsive to a metric;

identifies a catalog cluster of the plurality of catalog clusters to which a legacy item does not belong and which has a determined distance less than a desired distance to at least one catalog cluster of the plurality of catalog clusters to which a legacy item does belong; and

recommends a catalog item from the identified catalog cluster to which a legacy item does not belong.

17. A method, comprising:

clustering, by one or more processors, catalog items in a catalog into a plurality of catalog clusters of catalog items in a coordinate system responsive to preferences exhibited by a population of users;

representing, by the one or more processors, catalog items by trait vectors;

determining, by the one or more processors, for at least a subset of the catalog clusters a catalog cluster trait vector based on trait vectors of catalog items in the subset of the catalog of clusters;

determining, by the one or more processors, user legacy items belonging for the subset of the catalog of clusters, wherein the user legacy items represent items users have expressed preferences in previous periods of time;

representing, by the one or more processors, the user legacy items by trait vectors in the coordinate system;

determining, by the one or more processors, for each catalog cluster a representative legacy trait vector that is a function of and representative of the trait vectors of the user legacy items determined to belong in the catalog cluster;

recommending, by the one or more processors, a catalog item in a catalog cluster when an inner product of the trait vector of the catalog item and the representative legacy trait vector, from the origin point, is greater than the inner product between the representative legacy trait vector, from the origin point, and the cluster trait vector; and

transmitting, by the one or more processors over a network, the recommended catalog item to a computing device for presentation of the recommended catalog item to a user.

18. A method according to claim 17 wherein said clustering catalog items comprises acquiring rankings of the catalog items for a population of people and using the rankings to cluster the catalog items.

19. A method according to claim 18 wherein use of the rankings comprises factoring a ranking matrix comprising the rankings to determine a trait vector to represent catalog items.

20. A method according to claim 19, further comprising using the trait vectors to cluster the catalog items into the plurality of clusters.