Method and apparatus for generating a content recommendation in a recommendation system

• US 10,430,481 B2
• Filed: 05/26/2017
• Issued: 10/01/2019
• Est. Priority Date: 07/07/2016
• Status: Active Grant

• Current Assignee: Direct Cursus Technology LLC
• Original Assignee: Yandex Europe AG (Yandex N.V.)
• Inventors: Tikhonov, Aleksey Viktorovich
• Primary Examiner(s): Spieler, William
• Application Number:
  US15/606,658
• Publication Number:
  US 20180011937A1
• Time in Patent Office: 858 Days
• CPC Class Codes: G06F 16/335 Filtering based on additional data, e.g. user or group profiles filtering in web context G06F16/9535, G06F16/9536
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9536 Search customisation based on social or collaborative filtering
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 30/0255 based on user history
• CPC Class Codes: G06Q 30/0631 Item recommendations
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

There is disclosed a computer-implemented method of generating a content recommendation for a user of an electronic device, the method executable by a recommendation, the content recommendation being associated with a content item available at one of a plurality of network resources accessible via the communication network. The method comprises: executing a first machine learning algorithm module in order to determine a sub-set of recommended content sources from a plurality of possible content sources that is based on at least some of a first sub-set of user-specific content sources and a generated second sub-set of user-non-specific content sources; analyzing the sub-set of recommended content sources to select a plurality of potentially-recommendable content items; executing a second machine learning algorithm module in order to select, from the plurality of potentially-recommendable content items, at least one recommended content item; the selection being made on the basis of a user-profile-vector.

Claims

1. A computer-implemented method of generating a content recommendation for a user of an electronic device, the method executable by a recommendation server accessible by the electronic device via a communication network, the content recommendation being associated with a content item available at one of a plurality of network resources accessible via the communication network, the method comprising:

receiving, from the electronic device, a request for the content recommendation, the content recommendation including at least one recommended content item;

executing a first machine learning algorithm module in order to determine a sub-set of recommended content sources from a plurality of possible content sources, the determining the sub-set of recommended content sources including;

acquiring an indication of user-past-interactions with at least one of;

(i) the recommendation system and (ii) at least some of the plurality of network resources;

based on the user-past-interactions, determining a first sub-set of user-specific content sources;

based on

(i) a machine learning algorithm trained formula of other user interactions with at least some others of the plurality of network resources

and at least one of;

(ii) the first sub-set of user-specific content sources; and

(iii) a user-profile-vector generated based on the user-past-interactions,

generating a second sub-set of user-non-specific content sources;

processing the first sub-set of user specific content sources and the second sub-set of user-non-specific content sources in order to generate the sub-set of recommended content sources;

analyzing the sub-set of recommended content sources to select a plurality of potentially-recommendable content items;

executing a second machine learning algorithm module in order to select, from the plurality of potentially-recommendable content items, at least one recommended content item;

the selection being made on the basis of the user-profile-vector.

2. The method of claim 1, wherein the determining the first sub-set of user-specific content sources comprises applying an SVD algorithm.

3. The method of claim 2, wherein the determining the first sub-set of user-specific content sources is based on:

(i) the machine learning algorithm trained formula of other user interactions with at least some others of the plurality of network resources; and

(ii) the user-profile-vector generated based on the user-past-interactions.

4. The method of claim 3, the method further comprises, prior to the receiving the request for the content recommendation, generating by the SVD algorithm a matrix of network resources events and users.

5. The method of claim 1, wherein the determining the first sub-set of user-specific content sources comprises applying a PMI algorithm.

6. The method of claim 5, wherein the determining the first sub-set of user-specific content sources is based on:

(i) the machine learning algorithm trained formula of other user interactions with at least some others of the plurality of network resources; and

(ii) the first sub-set of user-specific content sources.

7. The method of claim 1, wherein the selection from the sub-set of recommended content sources the plurality of potentially-recommendable content items comprises determining a pre-defined number of most recent content items.

8. The method of claim 1, wherein the selection from the sub-set of recommended content sources the plurality of potentially-recommendable content items comprises determining a pre-defined number of most popular content items.

9. The method of claim 1, wherein the selection from the sub-set of recommended content sources is specific to a geographic region where the request for content recommendation was executed from.

10. The method of claim 1, further comprising filtering the selected potentially-recommendable content items.

11. The method of claim 10, wherein the filtering is based on a language setting of the electronic device and a language setting of the content item.

12. A server comprising:

a data storage medium;

a network interface configured for communication over a communication network;

a processor operationally coupled to the data storage medium and the network interface, the processor configured to;

receive, from an electronic device, a request for the content recommendation, the content recommendation including at least one recommended content item;

the content recommendation being associated with a content item available at one of a plurality of network resources accessible via the communication network;

execute a first machine learning algorithm module in order to determine a sub-set of recommended content sources from a plurality of possible content sources, the determining the sub-set of recommended content sources including;

acquiring an indication of user-past-interactions with at least one of;

(i) the recommendation system and (ii) at least some of the plurality of network resources;

based on the user-past-interactions, determining a first sub-set of user-specific content sources;

based on

(i) a machine learning algorithm trained formula of other user interactions with at least some others of the plurality of network resources

and at least one of;

(ii) the first sub-set of user-specific content sources; and

(iii) a user-profile-vector generated based on the user-past-interactions,

generating a second sub-set of user-non-specific content sources;

processing the first sub-set of user specific content sources and the second sub-set of user-non-specific content sources in order to generate the sub-set of recommended content sources;

analyze the sub-set of recommended content sources to select a plurality of potentially-recommendable content items;

execute a second machine learning algorithm module in order to select, from the plurality of potentially-recommendable content items, at least one recommended content item;

the selection being made on the basis of the user-profile-vector.

13. The server of claim 12, wherein to determine the first sub-set of user-specific content sources, the processor is configured to apply an SVD algorithm.

14. The server of claim 13, wherein the determining the first sub-set of user-specific content sources is based on:

(i) the machine learning algorithm trained formula of other user interactions with at least some others of the plurality of network resources; and

(ii) the user-profile-vector generated based on the user-past-interactions.

15. The server of claim 14, wherein the server is further configured, prior to the receiving the request for the content recommendation, to generate, suing the SVD algorithm, a matrix of network resources events and users.

16. The server of claim 12, wherein to determine the first sub-set of user-specific content sources, the processor is configured to apply a PMI algorithm.

17. The server of claim 16, wherein the determining the first sub-set of user-specific content sources is based on:

(i) the machine learning algorithm trained formula of other user interactions with at least some others of the plurality of network resources; and

(ii) the first sub-set of user-specific content sources.

18. The server of claim 12, wherein the selection from the sub-set of recommended content sources the plurality of potentially-recommendable content items comprises determining a pre-defined number of most recent content items.

19. The server of claim 12, wherein the selection from the sub-set of recommended content sources the plurality of potentially-recommendable content items comprises determining a pre-defined number of most popular content items.

20. The server of claim 12, wherein the selection from the sub-set of recommended content sources is specific to a geographic region where the request for content recommendation was executed from.