System and method for feature determination and content selection

Patent metadata

Field	Value
Publication number	US-12585718-B2
Application number	US-202418608597-A
Country	US
Kind code	B2
Filing date	Mar 18, 2024
Priority date	Oct 26, 2022
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

In an example, first entities are extracted from user profiles. Second entities are extracted from content information associated with content item. User-associated metrics associated with the first entities are determined based upon the user profiles and/or content events. First vector representations of the first entities and second vector representations of the second entities are processed to generate an attention distribution array. Each value of the attention distribution array represents, for a user interested in an entity of the first entities, a proportion of (i) entity-specific activity, of the user, related to an entity of the second entities relative to (ii) an entirety of activity of the user. An inferred activity distribution array is generated by applying the user-associated metrics to the attention distribution array. A filtered subset of activity distribution values is generated by pruning values from the inferred activity distribution array. Transmission of content is controlled using the filtered subset of activity distribution values.

First claim

What is claimed is:

1 . A method, comprising:
determining a plurality of user-associated metrics associated with first entities in which one or more users have an interest, wherein:
a first user-associated metric is representative of online activity of one or more first users having an interest in a first entity of the first entities; and
a second user-associated metric is representative of online activity of one or more second users having an interest in a second entity of the first entities;
processing, using a neural network model, first vector representations associated with the first entities in which one or more users have an interest and second vector representations associated with second entities associated with one or more content items to generate an attention distribution array, wherein each value of the attention distribution array is based upon, for a user interested in an entity of the first entities in which one or more users have an interest, both (i) entity-specific activity, of the user, related to an entity of the second entities associated with one or more content items and (ii) an entirety of activity of the user; generating an inferred activity distribution array by applying the plurality of user-associated metrics to the attention distribution array; generating a filtered subset of activity distribution values by pruning values from the inferred activity distribution array; training a profile-processing machine learning model using the filtered subset of activity distribution values to generate a trained profile-processing machine learning model; and controlling transmission of content using the trained profile-processing machine learning model.

2 . The method of claim 1 , wherein controlling the transmission of content comprises:
receiving a request for content associated with a client device associated with a user profile; in response to receiving the request for content, determining, using the user profile and the trained profile-processing machine learning model, a plurality of content item scores associated with a second plurality of content items; selecting, based upon the plurality of content item scores, a first content item of the second plurality of content items; and providing the first content item for presentation via the client device.

3 . The method of claim 1 , wherein generating the filtered subset of activity distribution values comprises:
comparing each value of the inferred activity distribution array with a threshold; and pruning the values in response to determining that the values do not meet the threshold.

4 . The method of claim 1 , wherein generating the filtered subset of activity distribution values comprises:
pruning the values in response to determining that the values are n lowest values of the inferred activity distribution array, wherein n is equal to a difference between a total quantity of values of the inferred activity distribution array and a defined quantity of values to be included in the filtered subset of activity distribution values.

5 . The method of claim 1 , wherein the plurality of user-associated metrics are determined based upon one or more content events comprising at least one of:
a presentation event associated with presentation of a first content item; or a click event associated with selection of a second content item.

6 . The method of claim 1 , wherein:
the first user-associated metric is indicative of a measure of presentation events of the one or more first users; and the second user-associated metric is indicative of a measure of presentation events of the one or more second users.

7 . The method of claim 6 , wherein:
each value of the attention distribution array represents an inferred measure of presentation events, of a user interested in an entity of the first entities, related to an entity of the second entities.

8 . The method of claim 1 , wherein:
the first user-associated metric is indicative of a measure of click events of the one or more first users; and the second user-associated metric is indicative of a measure of click events of the one or more second users.

9 . The method of claim 8 , wherein:
each value of the attention distribution array represents an inferred measure of click events, of a user interested in an entity of the first entities, related to an entity of the second entities.

10 . The method of claim 1 , wherein:
the neural network model comprises a multi-head attention model associated with a first head and a second head; and processing the first vector representations and the second vector representations to generate the attention distribution array is performed using the first head of the multi-head attention model.

11 . The method of claim 10 , comprising:
processing, using the second head of the multi-head attention model, the first vector representations and the second vector representations to generate a second attention distribution array; and generating a second inferred activity distribution array by applying the plurality of user-associated metrics to the second attention distribution array, wherein generating the filtered subset of activity distribution values comprises pruning values from the second inferred activity distribution array and including remaining values, from the second inferred activity distribution array, in the filtered subset of activity distribution values.

12 . The method of claim 11 , wherein:
processing the first vector representations and the second vector representations to generate the attention distribution array is performed based upon at least one of:
one or more first parameters of the first head of the multi-head attention model; or
a first vector sub-space associated with the first head of the multi-head attention model; and
processing the first vector representations and the second vector representations to generate the second attention distribution array is performed based upon at least one of:
one or more second parameters of the second head of the multi-head attention model; or
a second vector sub-space associated with the second head of the multi-head attention model.

13 . A computing device comprising:
a processor; and memory comprising processor-executable instructions that when executed by the processor cause performance of operations, the operations comprising:
determining a plurality of user-associated metrics associated with first entities in which one or more users have an interest, wherein:
a first user-associated metric is representative of online activity of one or more first users having an interest in a first entity of the first entities; and
a second user-associated metric is representative of online activity of one or more second users having an interest in a second entity of the first entities;

processing, using a model, first vector representations associated with the first entities in which one or more users have an interest and second vector representations associated with second entities associated with one or more content items to generate an attention distribution array, wherein each value of the attention distribution array is based upon, for a user interested in an entity of the first entities in which one or more users have an interest, both (i) entity-specific activity, of the user, related to an entity of the second entities associated with one or more content items and (ii) an entirety of activity of the user;
generating an inferred activity distribution array by applying the plurality of user-associated metrics to the attention distribution array;
training a machine learning model usin

Assignees

• Yahoo Assets Llc

Inventors

• Ma Yufeng
• Shen Rao
• Tsioutsiouliklis Kostas
• Kim Donghyun
• Li Liuqing

Classifications

• G06F16/9538

  Presentation of query results · CPC title

• G06N3/08

  Learning methods · CPC title

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title

• G06N3/045 Primary

  Combinations of networks · CPC title