US 12,223,554 B2Granted

Recommendations using session relevance and incremental learning

AssigneeMicrosoft Technology Licensing, LLC

InventorsRohan Ramanath, Konstantin Salomatin, Jeffrey Douglas Gee, Onkar Anant Dalal, Gungor Polatkan, Sara Smoot Gerrard, Deepak Kumar, Rupesh Gupta, Jiaqi Ge, Lingjie Weng, Shipeng Yu

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

In some embodiments, a computer system generates a recommendation for a user of an online service based on user actions that have been performed by the user within a threshold amount of time before the generation of the recommendation. For each user action, the computer system determines an intent classification that identifies an activity of the user and that corresponds to different types of user actions, as well as a preference classification that identifies a target of the activity, and then stores these intent and preference classifications as part of indications of the user actions for use in generating different types of recommendations using different types of recommendation models. Additionally, the computer system may use mini-batches of data from an incoming stream of logged data to train an incremental update to one or more recommendation models.

Claims20

Dependent claims

A computer-implemented method performed by a computer system having at least one hardware processor, the computer-implemented method comprising:

receiving a plurality of user actions performed on an online service by a user via a computing device of the user;
generating a plurality of intent classifications corresponding to a plurality of different types of user actions, wherein at least one intent classification corresponds to two or more user actions;
for at least one of the plurality of user actions:
selecting an intent classification from the plurality of intent classifications according to a type of the user action, the selected intent classification identifying an intent of the user in performing the corresponding user action;
and
generating, according to the type of the user action, a preference classification identifying a target of the corresponding user action selected from a plurality of different user action targets;
storing the intent classification and the preference classification for each one of the plurality of user actions;
for each of a plurality of different types of user actions, storing a count of a number of occurrences of the corresponding type of user action found in the intent classifications for the plurality of user actions;
generating, at a first point in time, a first recommendation of a first type based on a first recommendation model, the generating of the first recommendation comprising using the stored intent classifications and preference classifications of the plurality of user actions as feature data for the first recommendation model based on the plurality of user actions having been performed within a threshold amount of time before the first point in time;
causing the first recommendation to be displayed on the computing device;
generating, at a second point in time after the first point in time, a second recommendation of a second type different from the first type based on a second recommendation model different from the first recommendation model, the generating of the second recommendation comprising using the stored intent classifications and preference classifications of the plurality of user actions as feature data for the second recommendation model based on the plurality of user actions having been performed within the threshold amount of time before the second point in time;
and
causing the second recommendation to be displayed on the computing device.

10 more elements

A system comprising:

at least one hardware processor;
and
a non-transitory machine-readable medium embodying a set of instructions that, when executed by the at least one hardware processor, cause the at least one hardware processor to perform operations, the operations comprising:
receiving a plurality of user actions performed on an online service by a user via a computing device of the user;
generating a plurality of intent classifications corresponding to a plurality of different types of user actions, wherein at least one intent classification corresponds to two or more user actions;
for at least one of the plurality of user actions;
selecting an intent classification from the plurality of intent classifications according to a type of the user action, the selected intent classification identifying an intent of the user in performing the corresponding user action;
and
generating, according to the type of the user action, a preference classification identifying a target of the corresponding user action selected from a plurality of different user action targets;
storing the intent classification and the preference classification for each one of the plurality of user actions;
for each of a plurality of different types of user actions, storing a count of a number of occurrences of the corresponding type of user action found in the intent classifications for the plurality of user actions;
generating, at a first point in time, a first recommendation of a first type based on a first recommendation model, the generating of the first recommendation comprising using the stored intent classifications and preference classifications of the plurality of user actions as feature data for the first recommendation model based on the plurality of user actions having been performed within a threshold amount of time before the first point in time;
causing the first recommendation to be displayed on the computing device;
generating, at a second point in time after the first point in time, a second recommendation of a second type different from the first type based on a second recommendation model different from the first recommendation model, the generating of the second recommendation comprising using the stored intent classifications and preference classifications of the plurality of user actions as feature data for the second recommendation model based on the plurality of user actions having been performed within the threshold amount of time before the second point in time;
and
causing the second recommendation to be displayed on the computing device.

13 more elements

A non-transitory machine-readable medium embodying a set of instructions that, when executed by at least one hardware processor, cause the at least one hardware processor to perform operations, the operations comprising:

receiving a plurality of user actions performed on an online service by a user via a computing device of the user;
generating a plurality of intent classifications corresponding to a plurality of different types of user actions, wherein at least one intent classification corresponds to two or more user actions;
for at last one of the plurality of user actions;
selecting an intent classification from the plurality of intent classifications according to a type of the user action, the selected intent classification identifying an intent of the user in performing the corresponding user action;
and
generating, according to the type of the user action, a preference classification identifying a target of the corresponding user action selected from a plurality of different user action targets;
storing the intent classification and the preference classification for each one of the plurality of user actions;
for each of a plurality of different types of user actions, storing a count of a number of occurrences of the corresponding type of user action found in the intent classifications for the plurality of user actions;
generating, at a first point in time, a first recommendation of a first type based on a first recommendation model, the generating of the first recommendation comprising using the stored intent classifications and preference classifications of the plurality of user actions as feature data for the first recommendation model based on the plurality of user actions having been performed within a threshold amount of time before the first point in time;
causing the first recommendation to be displayed on the computing device;
generating, at a second point in time after the first point in time, a second recommendation of a second type different from the first type based on a second recommendation model different from the first recommendation model, the generating of the second recommendation comprising using the stored intent classifications and preference classifications of the plurality of user actions as feature data for the second recommendation model based on the plurality of user actions having been performed within the threshold amount of time before the second point in time;
and
causing the second recommendation to be displayed on the computing device.

10 more elements

Dependencies are inferred from the claim text.

Facts

Application: 16/912,245Filed 25 Jun 2020
Published: US 2021/0406838 A130 Dec 2021
Granted: 11 Feb 2025
Term ends: Not estimatedThe source records don’t support an estimate.

Classification12 codes

G06Q 10/1053
Human resources: employment or hiring
G06F 16/9535
Querying, e.g. by the use of web search engines: search customisation based on user profiles and personalisation
G06F 16/958
Retrieval from the web: organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
G06F 18/214
Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation: generating training…

Family

US 2021/0406838 A1Pre-grant publication · published 30 Dec 2021
US 12,223,554 B2This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key