Method and apparatus for generating a recommended content list

• US 10,387,513 B2
• Filed: 08/15/2016
• Issued: 08/20/2019
• Est. Priority Date: 08/28/2015
• Status: Active Grant

• Current Assignee: Direct Cursus Technology LLC
• Original Assignee: Yandex Europe AG (Yandex N.V.)
• Inventors: Royzner, Mikhail Aleksandrovich
• Primary Examiner(s): Bullock, Joshua
• Application Number:
  US15/236,538
• Publication Number:
  US 20170061021A1
• Time in Patent Office: 1,100 Days
• Field of Search: 707667
• 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 5/04 Inference or reasoning models

Abstract

There is disclosed a method for generating a content recommendation for a given user of a recommendation system. The method comprises: receiving a request for the content recommendation; responsive to the request generating a set of content recommendations for the given user, the generating being executed by a prediction module of the recommendation server, the prediction module having been trained using a training set of training events, such that for each given training event from the training set of training events: at least one user-nonspecific feature is used as a first input parameter for the prediction module training, the at least one user-nonspecific feature having been retrieved from a latest version of a snapshot archive available at a time of the given training event occurring; and at least one user-specific feature is used as a second input parameter for the prediction module training.

Claims

1. A method for generating a content recommendation for a given user of a recommendation system, the method executable at a recommendation server, the method comprising:

receiving, by the recommendation server, from an electronic device associated with the given user a request for the content recommendation;

responsive to the request generating, by the recommendation server, a set of content recommendations for the given user, the generating being executed by a prediction module of the recommendation server, the prediction module having been trained using a training set of training events, such that for each given training event from the training set of training events;

at least one user-nonspecific feature is used as a first input parameter for the prediction module training, the at least one user-nonspecific feature having been retrieved from the latest version of a snapshot archive available at a time of the given training event occurring, the latest version of the snapshot archive having been generated prior to the time of the given training event occurring;

at least one user-specific feature is used as a second input parameter for the prediction module training the at least one user-specific feature available at the time of the given training event occurring;

the generating comprising;

acquiring at least one in-use user non-specific feature from a then latest version of the snapshot archive, the then latest version of the snapshot archive having been generated prior to the generating the set of content recommendations;

generating an in-use user-specific feature at a moment of time of generating the set of content recommendations;

using the at least one in-use user non-specific feature and the in-use user-specific feature for generating the set of content recommendations;

transmitting at least a sub-set of the set of content recommendations to the electronic device.

2. The method of claim 1, further comprising acquiring the at least one user-nonspecific feature.

3. The method of claim 1, wherein the generating the at least one user-nonspecific feature is executed off-line.

4. The method of claim 1, further comprising generating the at least user-specific feature.

5. The method of claim 4, wherein the generating the at least one user-specific feature is executed in real time at the time of training.

6. The method of claim 1, wherein the prediction module training is based on an indication of the training event and the associated at least one user-nonspecific feature and at least one user-specific feature.

7. The method of claim 1, wherein at least one user-nonspecific feature comprises a plurality of user-nonspecific features and at least one user-specific feature comprises a plurality of user-specific features and wherein none of the plurality of user-nonspecific features is the same as any of the plurality of user-specific features.

8. The method of claim 1, further comprising generating the latest version of the snapshot archive and storing the latest version of the snapshot archive in a memory accessible by the recommendations server, and wherein the generating the latest version of the snapshot archive is executed at a point of time after generating a previous version of the snapshot archive and wherein once the latest version of the snapshot archive is generated, its content is used instead of a content of the previous version of the snapshot archive.

9. The method of claim 1, wherein the at least one user-specific feature was non-available at the time the latest version of the snapshot archive was generated.

10. A method of training a prediction module, the prediction module being part of a recommendation server, the method comprising:

generating a training set of training events, such that for each given training event from the training set of training events;

at least one user-nonspecific feature is used as a first input parameter for the prediction module training, the at least one user-nonspecific feature having been generated and stored prior to generating the training set, and retrieved from a latest version of a snapshot archive available at a time of the given training event occurring, the latest version of the snapshot archive having been generated prior to the time of the given training event occurring;

at least one user-specific feature is used as a second input parameter for the prediction module training, at least one user-specific feature generated at the time of the given training event occurring;

using the training set to train the prediction module to generate an indication of at least one recommendation item.

11. The method of claim 10, wherein the at least one user-specific feature was not non-available at the time the latest version of the snapshot archive was generated.

12. The method of claim 10, wherein the at least one user-specific feature is generated at the time of the using the training set.

13. The method of claim 10, wherein the at least one user-specific feature is generated by a second prediction module of the recommendation server.

14. A server, the server comprising:

a processing module configured to;

receive from an electronic device associated with the given user a request for the content recommendation;

responsive to the request, generate a set of content recommendations for the given user, the generating being executed by a prediction module of the recommendation server, the prediction module having been trained using a training set of training events, such that for each given training event from the training set of training events;

at least one user-nonspecific feature is used as a first input parameter for the prediction module training, the at least one user-nonspecific feature having been retrieved from a latest version of a snapshot archive available at a time of the given training event occurring, the latest version of the snapshot archive having been generated prior to the time of the given training event occurring;

at least one user-specific feature is used as a second input parameter for the prediction module training, at least one user-specific feature available at the time of the given training event occurring, the at least one user-specific feature being non-available at the time the latest version of the snapshot archive was generated;

the generating comprising;

acquiring at least one in-use user non-specific feature from a then latest version of the snapshot archive, the then latest version of the snapshot archive having been generated prior to the generating the set of content recommendations;

generating an in-use user-specific feature at a moment of time of generating the set of content recommendations;

using the at least one in-use user non-specific feature and the in-use user-specific feature for generating the set of content recommendations;

transmit at least a sub-set of the set of content recommendations to the electronic device.