Recommending items to users utilizing a bi-linear collaborative filtering model

• US 8,781,915 B2
• Filed: 10/17/2008
• Issued: 07/15/2014
• Est. Priority Date: 10/17/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Herbrich, Ralf, Graepel, Thore, Stern, David
• Primary Examiner(s): Jarrett, Scott L
• Application Number:
  US12/253,854
• Publication Number:
  US 20100100416A1
• Time in Patent Office: 2,097 Days
• Field of Search: 705/26.7, 705/347
• US Class Current: 705/26.7
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0203 Market surveys; Market polls
• CPC Class Codes: G06Q 30/0633 Lists, e.g. purchase orders, compilation or processing

Abstract

A recommender system may be used to predict a user behavior that a user will give in relation to an item. In an embodiment such predictions are used to enable items to be recommended to users. For example, products may be recommended to customers, potential friends may be recommended to users of a social networking tool, organizations may be recommended to automated users or other items may be recommended to users. In an embodiment a memory stores a data structure specifying a bi-linear collaborative filtering model of user behaviors. In the embodiment an automated inference process may be applied to the data structure in order to predict a user behavior given information about a user and information about an item. For example, the user information comprises user features as well as a unique user identifier.

Claims

1. A computer-implemented method of predicting a user behavior with respect to an item, the method comprising:

arranging a memory to store a factor graph specifying a bi-linear collaborative filtering model, wherein the factor graph is updated based on;

one or more latent user traits, the latent user traits including one or more demographic traits;

one or more latent item traits, the latent item traits including one or more product feature descriptions or service feature descriptions; and

a determination of an inner product of at least one latent user trait and at least one latent item trait, the factor graph comprising a plurality of probability distributions representing belief about the one or more latent user traits and the one or more latent item traits of the bi-linear collaborative filtering model;

predicting the user behavior with respect to a plurality of different user and item pairs by arranging a first processor to apply an inference process to the factor graph;

recommending, via an output, at least one of the plurality of items to the user based at least in part on the predicted user behavior; and

updating a variance of the plurality of probability distributions based at least in part on actual user behavior.

2. A method as claimed in claim 1, wherein the inference process comprises an assumed density filtering process.

3. A method as claimed in claim 1, further comprising:

receiving information that includes a first plurality of observed user behaviors from a first specified set of items and a second plurality of observed user behaviors from a second specified set of items, each observed user behavior having associated user features and associated item features;

arranging the first processor to update the factor graph based at least in part on the first plurality of observed user behaviors from the first specified set of items;

arranging a second processor to concurrently update the factor graph based at least in part on the second plurality of observed user behaviors from the second specified set of items; and

arranging a monitor to prevent a conflict in the concurrent updates.

4. A method as claimed in claim 1, wherein the first processor is arranged to update a variance of the plurality of probability distributions at specified time intervals.

9. A method as claimed in claim 1, further comprising providing one or more item recommendations to the user based at least in part on the predicted user behavior.

10. A method as claimed in claim 9, further comprising:

learning at least two predicted rating thresholds for the user; and

determining the one or more item recommendations using the at least two predicted rating thresholds.

5. An apparatus for recommending one or more items to a user, the apparatus comprising:

a memory arranged to store a factor graph specifying a bi-linear collaborative filtering model of user behavior, the factor graph being updated based on;

one or more latent user traits, the latent item traits including one or more product feature descriptions or service feature descriptions;

one or more latent item traits, the latent item traits including one or more product feature descriptions or service feature descriptions; and

a determination of an inner product of at least one latent user trait and at least one latent item trait;

a first processor configured to apply an inference process to the factor graph in order to predict, for each of a plurality of items, a user behavior based at least in part on the inner product, the factor graph comprising a plurality of probability distributions representing belief about the one or more latent user traits and the one or more latent item traits-of the bi-linear collaborative filtering model; and

an output configured to recommend at least one of the plurality of items to the user based at least in part on the predicted user behavior and to update a variance of the plurality of probability distributions based at least in part on actual user behavior.

6. An apparatus as claimed in claim 5 wherein the inference process comprises an assumed density filtering process.

7. An apparatus as claimed in claim 5, further comprising an input arranged to receive information that includes an observed user behavior together with associated user features and associated item features, and wherein the first processor is arranged to update the factor graph based at least in part on the received information.

8. An apparatus as claimed in claim 5, further comprising:

a first input arranged to receive information that includes a first plurality of observed user behaviors from a first specified set of items each having associated user features and associated item features, and wherein the first processor is arranged to update the factor graph based at least in part on the first plurality of observed user behaviors;

a second input arranged to receive information that includes a second plurality of observed user behaviors from a second specified set of items each having associated user features and associated item features;

a second processor arranged to concurrently update the factor graph based at least in part on the second plurality of observed user behaviors; and

a monitor arranged to prevent conflict in the concurrent updates.

11. An apparatus as claimed in claim 5, wherein the output is further arranged to learn at least two predicted rating thresholds for a particular user and determine the at least one recommended item using the at least two predicted rating thresholds.

12. One or more storage media, the one or more storage media being hardware, storing computer-readable instructions that when executed by one or more processors perform actions comprising:

arranging a memory to store a factor graph specifying a bi-linear collaborative filtering model, wherein the factor graph is updated based on;

one or more latent user traits, the latent user traits including one or more demographic traits;

one or more latent item traits, the latent item traits including one or more product feature descriptions or service feature descriptions; and

a determination of an inner product of at least one latent user trait and at least one latent item trait, the factor graph comprising a plurality of probability distributions representing belief about the one or more latent user traits and the one or more latent item traits-of the bi-linear collaborative filtering model;

predicting the user behavior with respect to a plurality of different user and item pairs by arranging a first processor to apply an inference process to the factor graph;

recommending, via an output, at least one of the plurality of items to the user based at least in part on the predicted user behavior; and

updating a variance of the plurality of probability distributions based at least in part on actual user behavior.

13. One or more storage media as claimed in claim 12, wherein the inference process comprises an assumed density filtering process.

14. One or more storage media as claimed in claim 12, the actions further comprising:

receiving information that includes a first plurality of observed user behaviors from a first specified set of items and a second plurality of observed user behaviors from a second specified set of items, each observed user behavior having associated user features and associated item features;

arranging the first processor to update the factor graph based at least in part on the first plurality of observed user behaviors from the first specified set of items;

arranging a second processor to concurrently update the factor graph based at least in part on the second plurality of observed user behaviors from the second specified set of items; and

arranging a monitor to prevent a conflict in the concurrent updates.

15. One or more storage media as claimed in claim 12, wherein the first processor is arranged to update a variance of the plurality of probability distributions at specified time intervals.