Systems and methods for online content recommendation

Patent metadata

Field	Value
Publication number	US-11494666-B2
Application number	US-201514748333-A
Country	US
Kind code	B2
Filing date	Jun 24, 2015
Priority date	Jun 17, 2015
Publication date	Nov 8, 2022
Grant date	Nov 8, 2022

Abstract

The present disclosure relates to computer systems implementing methods for online content recommendation. The computer systems may be configured to receive a training sample from a first client device corresponding to a predefined feedback interacting with online content displayed on the first client device; update a preexisting training database in real-time based on the received training sample to generate an updated training sample, wherein prior to being updated based on the training sample received from the first client, the training database includes a set of historical training samples; conduct a regression training to a computer learning model in real-time, using the updated training sample, to produce a set of trained parameters for an online content recommendation model; call the set of trained parameters in real-time to determine recommend online content for a second user with the online content recommendation model; and send the recommended online content to a second client device of the second user.

First claim

The invention claimed is:

1. A computer system, comprising:
a storage medium comprising a set of instructions for online content recommendation; and
a processor in communication with the storage medium, wherein when executing the set of instructions, the processor is directed to:
receive in real-time a user request for online content from a first client device;
identify request features, comprising context related features comprising at least one of a time of day associated with a first type of activity or a day of week associated with a second type of activity, from the user request;
determine an amount of time of a time interval based on available memory space, wherein a difference between a start of the time interval and an end of the time interval is the amount of time;
cache, in a buffer, the request features for the amount of time of the time interval;
after the request features are cached in the buffer, receive in real-time a predefined feedback over the online content requested by the user request from the first client device;
determine, based on whether the predefined feedback over the online content requested by the user request was received within the amount of time of the time interval, whether to discard the request features identified from the user request for the online content and cached in the buffer or use the request features identified from the user request for the online content and cached in the buffer for training sample generation;
responsive to the predefined feedback being received within the amount of time of the time interval, that was determined based on the available memory space, from the receipt of the user request:
generate a training sample by combining in real-time:
(i) the request features identified from the user request and comprising the context related features comprising at least one of the time of day associated with the first type of activity or the day of week associated with the second type of activity, and
(ii) the predefined feedback,

the training sample corresponding to a predefined feedback action taken by a first user interacting with the online content displayed on the first client device;

receive in real-time a second user request for second online content;
identify second request features, comprising second context related features comprising at least one of a second time of day associated with a third type of activity or a second day of week associated with a fourth type of activity, from the second user request;
determine a second amount of time of a second time interval based on second available memory space, wherein a difference between a start of the second time interval and an end of the second time interval is the second amount of time;
cache, in the buffer, the second request features for the second amount of time of the second time interval;
after the second request features are cached in the buffer, determine, based on whether second predefined feedback over the second online content requested by the second user request was received within the second amount of time of the second time interval, whether to discard the second request features identified from the second user request for the second online content and cached in the buffer or use the second request features identified from the second user request for the second online content and cached in the buffer for training sample generation;
responsive to determining that the second predefined feedback over the second online content has not been received within the second amount of time of the second time interval determined based on the second available memory space, discard the second request features identified from the second user request for the second online content and cached in the buffer;
update a preexisting training database in real-time based on the training sample generated using the user request for which the predefined feedback was received within the amount of time of the time interval, but not based on the discarded second request features identified from the second user request for which the second predefined feedback was not received within the second amount of time of the second time interval, to generate an updated training sample, wherein prior to being updated based on the training sample received from the first client device, the preexisting training database includes a set of historical training samples;
conduct a regression training to a computer learning model in real-time, using the updated training sample generated based on the training sample generated using the user request but not based on the discarded second request features identified from the second user request, to produce a set of trained parameters for an online content recommendation model and save the set of trained parameters in a model parameter database;
call the set of trained parameters in the model parameter database in real-time to determine recommended online content for a second user with the online content recommendation model; and
send the recommended online content to a second client device of the second user.

2. The computer system of claim 1 , wherein the amount of time is determined based on the available memory space, an amount of query traffic and a feature size.

3. The computer system of claim 1 , wherein the user request is received from the first client device when the first client device opens a user interface to load the online content.

4. The computer system of claim 1 , wherein the request features comprise at least one of demographic information of a user, device parameters of a client device, and a timestamp showing a time that the user request occurs.

5. The computer system of claim 1 , wherein the predefined feedback comprises at least one of a click of the online content, a “like” or “dislike” action associated with the online content, and at least one of sharing the online content or forwarding to a website.

6. The computer system of claim 1 , wherein at least one of the day of week of the context related features corresponds to weekends, or the time of day of the context related features corresponds to evenings.

7. The computer system of claim 1 , wherein to determine the recommended online content, the processor is further directed to:
call the set of trained parameters of the online content recommendation model;
based on the request features and for each item of a plurality of online items, determine a probability value that a user will perform at least one predefined feedback action over an online item;
select a predetermined number of online items from the plurality of online items that have highest probability values; and
identify the selected online items as the recommended online content to display on the second client device.

8. A method for online content recommendation, comprising:
receiving in real-time a user request for online content from a first client device;
identifying request features from the user request;
determining an amount of time of a time interval based on at least one of available memory space, an amount of query traffic or a feature size, wherein a difference between a start of the time interval and an end of the time interval is the amount of time;
caching, in a buffer, the request features for the amount of time of the time interval;
after the request features are cached in the buffer, receiving in real-time a predefined feedback over the online content requested by the user request from the first client device;
determining, based on whether the predefined feedback over the online content requested by the user request was received within the amount of time of the time interval, whether to discard the request features identified from the user request for the on

Assignees

• Yahoo Holdings Inc
• Yahoo Assets Llc

Inventors

• Golbandi Nadav
• Wang Chao

Classifications

• G06N5/04 Primary

  Inference or reasoning models · CPC title

• G06Q30/0269

  based on user profile or attribute · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06Q30/0241

  Advertisements · CPC title

• G06F16/9535

  Search customisation based on user profiles and personalisation · CPC title