REGULARIZED MODEL ADAPTATION FOR IN-SESSION RECOMMENDATIONS

• US 20170091652A1
• Filed: 09/24/2015
• Published: 03/30/2017
• Est. Priority Date: 09/24/2015
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: LinkedIn Corporation (Microsoft Corporation)
• Inventors: Miao, Xu, Zhou, Yitong, Young, Joel D., Tang, Lijun, Bhasin, Anmol
• Granted Patent: US 10,586,167 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06N 7/00 Computing arrangements based on specific mathematical models
• CPC Class Codes: G06Q 10/00 Administration; Management
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: G06Q 30/0282 Rating or review of business operators or products

Abstract

The disclosed embodiments provide a method and system for performing regularized model adaptation for in-session recommendations. During operation, the system obtains, from a server, a first global version of a statistical model. During a first user session with a user, the system improves a performance of the statistical model by using the first global version to output one or more recommendations to the user and using the first global version and user feedback from the user to create a first personalized version of the statistical model. At an end of the first user session, the system transmits an update containing a difference between the first personalized version and the first global version to the server for use in producing a second global version of the statistical model by the server.

Claims

1. A method, comprising:

obtaining, from a server by a computer system, a first global version of a statistical model; and

during a first user session with a user, improving a performance of the statistical model by performing the following steps on the computer system;

using the first global version to output one or more recommendations to the user; and

using the first global version and user feedback from the user to create a first personalized version of the statistical model; and

at an end of the first user session, transmitting an update comprising a difference between the first personalized version and the first global version to the server for use in producing a second global version of the statistical model by the server.

2. The method of claim 1, further comprising:

obtaining the second global version of the statistical model from the server, wherein the second global version comprises the update and one or more additional updates to the statistical model from one or more additional computer systems;

replacing the first global version with the second global version; and

using the second global version and additional user feedback from the user to create a second personalized version of the statistical model during a second user session with the user.

3. The method of claim 1, further comprising:

using the first personalized version to output one or more additional recommendations to the user.

4. The method of claim 3, wherein using the first personalized version to output the one or more additional recommendations to the user comprises:

selecting the one or more additional recommendations based on the one or more recommendations and the user feedback.

5. The method of claim 4, wherein selecting the one or more additional recommendations based on the one or more recommendations and the user feedback comprises:

selecting the one or more additional recommendations based on a similarity to content associated with the user feedback.

6. The method of claim 4, wherein selecting the one or more additional recommendations based on the one or more recommendations and the user feedback comprises:

using the one or more recommendations and the user feedback to avoid selecting a recommendation that was previously outputted during the first user session.

7. The method of claim 1, wherein using the first global version and user feedback from the user to create the first personalized version of the statistical model comprises:

determining one or more weights for the user feedback based on a quality of the user feedback; and

including the one or more weights and the user feedback in training data for creating the first personalized version from the first global version.

8. The method of claim 1, wherein using the first global version and the user feedback from the user to create the first personalized version of the statistical model comprises:

updating the first personalized version after one or more units of the user feedback are received from the user.

9. The method of claim 1, wherein the one or more recommendations comprise at least one of:

a job recommendation;

a news recommendation;

a user-interface element; and

an action.

10. The method of claim 1, wherein the user feedback comprises at least one of:

a click;

a view;

a conversion; and

a search.

11. A method, comprising:

transmitting, from a server to a set of computer system, a first global version of a statistical model for generating a set of recommendations for a set of users;

obtaining, by the server from a first subset of the computer systems, a first set of updates comprising differences between the first global version and a first set of personalized versions of the statistical model generated from user feedback received during user sessions on the first subset of the computer systems;

merging, by the server, the first set of updates into a second global version of the statistical model; and

transmitting at least a portion of the second global version to the set of computer systems asynchronously from receiving a second set of updates to the first and second global versions from a second subset of the computer systems.

12. The method of claim 11, further comprising:

associating the first set of updates with the second global version;

using the associated first set of updates to merge the second set of updates into a third global version of the statistical model without merging the first set of differences into the third global version; and

transmitting at least a portion of the third global version to the set of computer systems.

13. The method of claim 12, wherein associating the first set of updates with the second global version comprises:

combining the first set of update identifiers into a version identifier for the second global version.

14. The method of claim 13, wherein using the associated first set of updates to merge the second set of updates into the third global version without merging the first set of differences into the third global version comprises:

identifying the first set of update identifiers in the version identifier for the second global version; and

omitting merging of the first set of updates into the second global version based on the identified first set of update identifiers in the version identifier.

15. The method of claim 11, further comprising:

matching the server to the set of client computer systems for distribution of the statistical model based on a first attribute of the server and a second attribute of the set of client computer systems.

16. An apparatus, comprising:

one or more processors; and

memory storing instructions that, when executed by the one or more processors, cause the apparatus to;

obtain, from a server, a first global version of a statistical model; and

during a first user session with a user, improving a performance of the statistical model by;

using the first global version to output one or more recommendations to the user; and

using the first global version and user feedback from the user to create a first personalized version of the statistical model; and

at an end of the first user session, transmitting an update comprising a difference between the first personalized version and the first global version to the server for use in producing a second global version of the statistical model by the server.

17. The apparatus of claim 16, wherein the memory further stores instructions that, when executed by the one or more processors, cause the apparatus to:

obtain the second global version of the statistical model from the server, wherein the second global version comprises the update and one or more additional updates to the statistical model from one or more additional computer systems;

replace the first global version with the second global version; and

use the second global version and additional user feedback from the user to create a second personalized version of the statistical model during a second user session with the user.

18. The apparatus of claim 16, wherein the memory further stores instructions that, when executed by the one or more processors, cause the apparatus to:

use the first personalized version to output one or more additional recommendations to the user.

19. The apparatus of claim 16, wherein using the first global version and user feedback from the user to create the first personalized version of the statistical model comprises:

determining one or more weights for the user feedback based on a quality of the user feedback; and

including the one or more weights and the user feedback in training data for creating the first personalized version from the first global version.

20. The apparatus of claim 16, wherein using the first global version and the user feedback from the user to create the first personalized version of the statistical model comprises:

updating the first personalized version after each unit of the user feedback is received from the user.