Real-time online-learning object recommendation engine

• US 9,773,063 B2
• Filed: 12/07/2011
• Issued: 09/26/2017
• Est. Priority Date: 12/07/2011
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Shi, Yanxin, Goder, Andrey, Ye, David
• Primary Examiner(s): Truong, Dennis
• Application Number:
  US13/313,984
• Publication Number:
  US 20130151539A1
• Time in Patent Office: 2,120 Days
• Field of Search: 707722, 707736, 707999001, 705 1449, 705 144
• 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

Abstract

In one embodiment, a system includes one or more computing systems that implement a social networking environment containing a large number of heterogeneous objects type, each of the plurality of object types having varying features, the system implementing a generic object recommendation engine for scoring objects and recommending the objects to users of the social networking system. In particular embodiments, the user and content object features are fed as inputs into a heuristic model that generates an expected value for the content object and user. In particular embodiments, the object recommendation engine includes an online learner that may log a user'"'"'s actions after the initial impression to determine the relatively degree of interest to the user.

Claims

1. A method comprising:

by one or more computing devices, receiving a request from a first user for a content page;

by one or more computing devices, determining a page identifier for the requested content page;

by one or more computing devices, determining, based on the page identifier, one or more second users, wherein the one or more second users are identified in the requested content page;

by one or more computing devices, determining a plurality of content items based on one or more of the second users, wherein each of the plurality content items is owned by a second user;

by one or more computing devices, determining one or more user features of the first user;

by one or more computing devices, determining for each of the content items one or more content features;

by one or more computing devices, calculating for each of the content items an expected value of the content item with respect to the first user based on the user features of the first user and the content features of the content item;

by one or more computing devices, ordering the content items by decreasing expected value; and

by one or more computing devices, delivering to the first user, with the requested content page, one or more of the content items as recommendations to the first user based on the ordering of the content items.

2. The method of claim 1, wherein the plurality of content items are determined based on the page identifier.

3. The method of claim 1, wherein the plurality of content items are determined based on a list of one or more services for the content page.

4. The method of claim 1, wherein the user features of the first user comprise an age of the first user.

5. The method of claim 1, wherein the user features of the first user comprise a gender of the first user.

6. The method of claim 1, wherein the user features comprise one or more previous requests for one or more content pages received from the first user.

7. The method of claim 1, wherein, for each of the content items, the content features of the comprise aggregate statistics for the content item based a plurality of users of a social-networking system.

8. The method of claim 7, wherein the aggregate statistics comprise a click-through rate for the content item.

9. The method of claim 7, wherein the aggregate statistics comprise a conversion rate for the content item.

10. The method of claim 7, wherein the aggregate statistics comprise a number of impressions of the content item delivered to the plurality of users during a predetermined period of time.

11. The method of claim 1, wherein the expected values are calculated for each of the content items using a respective recommendation model, wherein the recommendation model comprises a set of weights calculated for the first user using one or more statistical models based on the user features of the first user and the object content features of the content item.

12. The method of claim 11, wherein the one or more statistical models comprises logistic regression.

13. The method of claim 11, further comprising:

by one or more computing devices, generating an example data point, wherein the data point comprises one of a positive example or a negative example for the user features of the first user and the content features of the content item; and

by one or more computing devices, updating the set of weights of the respective recommendation model based on the example data point.

14. The method of claim 13, wherein:

the example data point is a positive example when a viewing user having the user features of the first user fails to convert on a viewed content item having the content features of the content item after an impression; and

the example data point is a negative example when the viewing user having the user features of the first user converts on the viewed content item having the content features of the content item after the impression.

15. The method of claim 14, wherein the viewing user converts on the viewed content item by clicking on a link displayed in a landing page corresponding to the viewed content item.

16. A non-transitory, computer-readable media comprising instructions operable, when executed by one or more computing systems, to:

receive a request from a first user for a content page;

determine a page identifier for the requested content page;

determine, based on the page identifier, one or more second users, wherein the one or more second users are identified in the requested content page;

determine a plurality of content items based on one or more of the second users, wherein each of the plurality content items is owned by a second user;

determine user features of the first user;

determine for each of the content items one or more content features;

calculate for each of the content items an expected value of the content item with respect to the first user based on the user features of the first user and the content features of the item;

order the content items by decreasing expected value; and

deliver to the first user, with the requested content page, one or more of the content items as recommendations to the first user based on the ordering of the content items.

17. The media of claim 16, wherein the plurality of content items are determined based on the page identifier.

18. The media of claim 16, the instructions further operable, when executed by one or more computing systems, to:

calculate for each of the content items the expected values using a respective recommendation model, wherein the recommendation model comprises a set of weights calculated for the first user using one or more statistical models based on the user features of the first user and the content features of the content item;

generate an example data point, wherein the data point comprises one of a positive example or a negative example for the user features of the first user and the content features of the content item; and

update the set of weights of the respective recommendation model based on the example data point.

19. The media of claim 18, wherein:

the example date point is a positive example when a viewing user having the user features of the first user fails to convert on a viewed content item having the content features of the content item after an impression; and

the example date point is a negative example when the viewing user having the user features of the first user converts on the viewed content item having the content features of the content item after the impression.