Inductive matrix completion and graph proximity for content item recommendation

Patent metadata

Field	Value
Publication number	US-11269962-B2
Application number	US-201916724949-A
Country	US
Kind code	B2
Filing date	Dec 23, 2019
Priority date	Apr 9, 2015
Publication date	Mar 8, 2022
Grant date	Mar 8, 2022

Abstract

Users may consume and/or share information through various types of content items. For example, user may post a family photo through a social network, create a running blog through a microblogging service, etc. Because users may be overwhelmed by the amount of available content items, it may be advantageous to recommend content items, such as blogs to follow, to users. Accordingly, inductive matrix completion is used to evaluate user interactions with content items (e.g., a user following a blog), content item features (e.g., text and/or images of a blog is evaluated to identify a topic of the blog), and/or user features (e.g., a user liking or reblogging a blog, user demographics, user interests, etc.) to determine whether to recommend a content item to a user. Additionally, graph proximity is used to recommend content items based upon weights of edges connecting user nodes to content item nodes within a directed graph.

First claim

What is claimed is:

1. A method for recommending a content item to a user, comprising:
generating, by a processor, a user to item interaction matrix indicating whether users have interacted with content items;
evaluating, by the processor, an image associated with a content item to recognize an object within the image;
identifying, by the processor, an image topic feature based upon the object;
generating, by the processor, a content item description matrix for the content item, the content item description matrix populated with one or more content item features, including the image topic feature, about the content item;
generating, by the processor, a user description matrix for a user, the user description matrix populated with one or more user features about the user, the one or more user features determined based upon an evaluation of at least one of geographic locations visited by the user, social network posts of the user, a consumer good purchase history of the user, a music streaming history of the user, calendar entries of the user or emails of the user;
evaluating, by the processor, the user to item interaction matrix, the content item description matrix, and the user description matrix to determine whether to recommend the content item to the user, wherein the evaluating comprises jointly factorizing the user to item interaction matrix, the content item description matrix, and the user description matrix to determine whether to recommend the content item to the user, the evaluating comprising performing inductive matrix completion upon one or more feature vectors to determine one or more latent factors of whether to recommend the content item to the user; and
controlling transmission of one or more recommendations to a computing device comprising sending, by the processor, a recommendation of the content item over a computer network to the computing device for rendering on a display of the computing device based upon a result of the joint factorization and refraining from sending, by the processor, a second recommendation of a second content item over the computer network to the computing device.

2. The method of claim 1 , the generating the user description matrix comprising identifying the social network posts of the user, evaluating the social network posts of the user to identify at least one of a first demographic of the user or a first interest of the user, and populating the user description matrix for the user based upon at least one of the first demographic or the first interest.

3. The method of claim 1 , the generating the user description matrix comprising identifying the consumer good purchase history of the user, evaluating the consumer good purchase history of the user to identify at least one of a first demographic of the user or a first interest of the user, and populating the user description matrix for the user based upon at least one of the first demographic or the first interest.

4. The method of claim 1 , the generating the user description matrix comprising identifying the music streaming history of the user, evaluating the music streaming history of the user to identify at least one of a first demographic of the user or a first interest of the user, and populating the user description matrix for the user based upon at least one of the first demographic or the first interest.

5. The method of claim 1 , the generating the user description matrix comprising identifying the calendar entries of the user, evaluating the calendar entries of the user to identify at least one of a first demographic of the user or a first interest of the user, and populating the user description matrix for the user based upon at least one of the first demographic or the first interest.

6. The method of claim 1 , the generating the user description matrix comprising identifying the emails of the user, evaluating the emails of the user to identify at least one of a first demographic of the user or a first interest of the user, and populating the user description matrix for the user based upon at least one of the first demographic or the first interest.

7. The method of claim 1 , the generating a user description matrix comprising:
identify a user feature for inclusion within the user description matrix based upon at least one user content item share activity.

8. The method of claim 1 , the evaluating the user to item interaction matrix, the content item description matrix, and the user description matrix comprising:
performing inductive matrix completion to jointly factorize the user to item interaction matrix, the content item description matrix, and the user description matrix to determine whether to recommend the content item to the user.

9. The method of claim 1 , the one or more user features determined based upon the social network posts of the user.

10. The method of claim 1 , the one or more user features determined based upon the calendar entries of the user.

11. The method of claim 1 , the one or more user features determined based upon the emails of the user.

12. The method of claim 1 ,
the evaluating the user to item interaction matrix, the content item description matrix, and the user description matrix comprising:
performing matrix completion factorization on the user to item interaction matrix to obtain a residual matrix; and
performing inductive matrix completion to factorize the residual matrix, the content item description matrix, and the user description matrix to determine whether to recommend the content item to the user; and

the performing matrix completion factorization comprising:
performing matrix completion factorization on the user to item interaction matrix to generate an approximate matrix; and
obtaining the residual matrix as a difference between the user to item interaction matrix and the approximate matrix.

13. The method of claim 1 , the content item corresponding to a blog.

14. The method of claim 13 , the blog comprising a new blog with less than a threshold number of followers.

15. The method of claim 12 , the user comprising a new user with less than a threshold number of followed blogs.

16. A system for recommending a content item to a user, comprising:
a content recommendation component, executed by a processor using instructions stored in memory, configured to:
generate a user to item interaction matrix indicating whether users have interacted with content items;
evaluate an image associated with a content item to recognize an object within the image;
identify an image topic feature based upon the object;
generate a content item description matrix for the content item, the content item description matrix populated with one or more content item features, including the image topic feature, about the content item;
generate a user description matrix for a user, the user description matrix populated with one or more user features about the user, the one or more user features determined based upon an evaluation of at least one of:
geographic locations visited by the user; or
content of the user;

evaluate the user to item interaction matrix, the content item description matrix, and the user description matrix by jointly factorizing the user to item interaction matrix, the content item description matrix, and the user description matrix to determine whether to recommend the content item to the user, the evaluating comprising performing inductive matrix completion upon one or more feature vectors to determine one or more latent factors of whether to recommend the content item to the user; and
selectively send or refrain from sending a r

Assignees

• Oath Inc
• Verizon Media Inc

Inventors

• Cetintas Suleyman
• Lee Kuang-Chih

Classifications

• G06F16/951 Primary

  Indexing; Web crawling techniques · CPC title

• G06F16/9024 Primary

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F16/3331

  Query processing · CPC title