Recommendations in a computing advice facility

• US 9,009,096 B2
• Filed: 07/12/2012
• Issued: 04/14/2015
• Est. Priority Date: 07/12/2011
• Status: Active Grant

• Current Assignee: PayPal, Inc. (PayPal Holdings, Inc.)
• Original Assignee: eBay Inc.
• Inventors: Pinckney, Thomas, Dixon, Christopher, Gattis, Matthew Ryan
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): PELLETT, DANIEL T
• Application Number:
  US13/547,893
• Publication Number:
  US 20130124449A1
• Time in Patent Office: 1,006 Days
• Field of Search: None
• US Class Current: 706/52
• CPC Class Codes: G06F 16/2425 Iterative querying; Query formulation based on the results of a preceding query
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/045 Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
• CPC Class Codes: G06N 5/048 Fuzzy inferencing

Abstract

According to various embodiments, a ratings matrix including matrix values is generated, each row of the ratings matrix identifying one of a plurality of users, each column of the ratings matrix identifying one of a plurality of items, and each of the matrix values corresponding to a known affinity rating describing a degree of affinity associated with one of the users and one of the items. The ratings matrix may include a missing entry representing an unknown affinity rating. According to various embodiments, a revised ratings matrix is generated by factoring the ratings matrix into a user matrix and an item matrix, the revised ratings matrix being the product of the user matrix and the item matrix and including at least one entry representing a predicted affinity rating in place of the missing entry.

Claims

1. A method comprising:

generating a ratings matrix including matrix values, each row of the ratings matrix identifying one of a plurality of users, each column of the ratings matrix identifying one of a plurality of items, and each of the matrix values corresponding to a known affinity rating describing a degree of affinity associated with one of the users and one of the items, wherein the ratings matrix includes a missing entry representing an unknown affinity rating;

generating, using one or more processors, a revised ratings matrix by factoring the ratings matrix into a user matrix and an item matrix, the revised ratings matrix being the product of the user matrix and the item matrix and including at least one entry representing a predicted affinity rating in place of the missing entry;

estimating a confidence value associated with at least a portion of the known affinity ratings in the ratings matrix;

generating a confidence matrix that includes the confidence values;

estimating a probability that a specific user had an opportunity to rate a specific item, the specific user and the specific item being associated with a missing affinity rating in the ratings matrix;

replacing the missing affinity rating in the ratings matrix with a low affinity rating; and

inserting a confidence value associated with the specific user and the specific item in the confidence matrix, the confidence value having a value equal to the probability.

2. The method of claim 1, wherein each row of the user matrix includes user parameters describing each user in the ratings matrix, and wherein each column of the item matrix includes item parameters describing each item in the ratings matrix.

3. The method of claim 2, further comprising:

inserting a new row identifying a new user into the ratings matrix, the new row including a known affinity rating describing a degree of affinity associated with the new user; and

estimating user parameters describing the new user, based on the known affinity rating describing the degree of affinity associated with the new user and the item parameters included in the item matrix.

4. The method of claim 3, further comprising:

predicting an affinity rating describing a degree of affinity associated with the new user, based on the estimated user parameters and the item parameters included in the item matrix.

5. The method of claim 2, further comprising:

inserting a new column identifying a new item into the ratings matrix, the new column including a known affinity rating describing a degree of affinity associated with the new item; and

estimating item parameters describing the new item, based on the known affinity rating describing the degree of affinity associated with the new item and the user parameters included in the user matrix.

6. The method of claim 5, further comprising:

predicting an affinity rating describing a degree of affinity associated with the new item, based on the estimated item parameters and the user parameters included in the user matrix.

7. The method of claim 1, further comprising:

receiving a new affinity rating describing a degree of affinity associated with a particular user and a particular item; and

revising user parameters describing the particular user in the user matrix based on the new affinity rating.

8. The method of claim 1, further comprising:

receiving a new affinity rating describing a degree of affinity associated with a particular user and a particular item; and

revising item parameters describing the particular item in the item matrix based on the new affinity rating.

9. The method of claim 1, further comprising:

factoring the ratings matrix into the user matrix and the item matrix to generate the revised ratings matrix, by weighting the low affinity rating in the ratings matrix based on the confidence value in the confidence matrix.

10. The method of claim 1, further comprising:

estimating a probability that a missing affinity rating in the ratings matrix that describes an affinity associated with a specific user and a specific item corresponds to a negative affinity rating;

replacing the missing affinity rating in the ratings matrix with a low affinity rating; and

inserting a confidence value associated with the specific user and the specific item in the confidence matrix, where the confidence value has a value equal to the probability.

11. The method of claim 10, further comprising:

factoring the ratings matrix into the user matrix and the item matrix to generate the revised ratings matrix, by weighting the low affinity rating in the ratings matrix based on the confidence value in the confidence matrix.

12. The method of claim 2, further comprising:

determining that a first user is similar to a second user, based on the user parameters in the user matrix describing the first and second users; and

generating a recommendation of a particular item for the second user, based on a recommendation of the particular item for the first user.

13. The method of claim 2, further comprising

determining that a first item is similar to a second item, based on the item parameters in the item matrix describing the first and second items;

and

generating a recommendation of the second item for a particular user, based on a recommendation of the first item for the particular user.

14. The method of claim 1, further comprising

counting a number of common affinity ratings between a first user and a second user, wherein each of the common affinity ratings describe a degree of affinity associated with a first item or a second item that is determined to be similar to the first item;

determining that the first user is similar to the second user, based on the number of common affinity ratings.

15. A non-transitory machine-readable storage medium having embodied thereon instructions executable by one or more machines to perform operations comprising:

generating a ratings matrix including matrix values, each row of the ratings matrix identifying one of a plurality of users, each column of the ratings matrix identifying one of a plurality of items, and each of the matrix values corresponding to a known affinity rating describing a degree of affinity associated with one of the users and one of the items, wherein the ratings matrix includes a missing entry representing an unknown affinity rating;

generating, using one or more processors, a revised ratings matrix by factoring the ratings matrix into a user matrix and an item matrix, the revised ratings matrix being the product of the user matrix and the item matrix and including at least one entry representing a predicted affinity rating in place of the missing entry;

estimating a confidence value associated with at least a portion of the known affinity ratings in the ratings matrix;

generating a confidence matrix that includes the confidence values;

estimating a probability that a specific user had an opportunity to rate a specific item, the specific user and the specific item being associated with a missing affinity rating in the ratings matrix;

replacing the missing affinity rating in the ratings matrix with a low affinity rating; and

inserting a confidence value associated with the specific user and the specific item in the confidence matrix, the confidence value having a value equal to the probability.

16. An apparatus comprising:

a matrix generation module configured to generate a ratings matrix including matrix values, each row of the ratings matrix identifying one of a plurality of users, each column of the ratings matrix identifying one of a plurality of items, and each of the matrix values corresponding to a known affinity rating describing a degree of affinity associated with one of the users and one of the items, wherein the ratings matrix includes a missing entry representing an unknown affinity rating; and

a prediction module, implemented by one or more processors, configured to;

generate a revised ratings matrix by factoring the ratings matrix into a user matrix and an item matrix, the revised ratings matrix being the product of the user matrix and the item matrix and including at least one entry representing a predicted affinity rating in place of the missing entry;

estimate a confidence value associated with at least a portion of the known affinity ratings in the ratings matrix;

generate a confidence matrix that includes the confidence values;

estimate a probability that a specific user had an opportunity to rate a specific item, the specific user and the specific item being associated with a missing affinity rating in the ratings matrix;

replace the missing affinity rating in the ratings matrix with a low affinity rating; and

insert a confidence value associated with the specific user and the specific item in the confidence matrix, the confidence value having a value equal to the probability.

17. The apparatus of claim 16, wherein the prediction module is further configured to:

insert a new row identifying a new user into the ratings matrix, the new row including a known affinity rating describing a degree of affinity associated with the new user; and

estimate user parameters describing the new user, based on the known affinity rating describing the degree of affinity associated with the new user and item parameters included in the item matrix.

18. The apparatus of claim 16, wherein the prediction module is further configured to:

insert a new column identifying a new item into the ratings matrix, the new column including a known affinity rating describing a degree of affinity associated with the new item; and

estimate item parameters describing the new item, based on the known affinity rating describing the degree of affinity associated with the new item and user parameters included in the user matrix.