Dynamic selection of advertisements using deep learning models on client devices

Patent metadata

Field	Value
Publication number	US-2020410553-A1
Application number	US-201916454375-A
Country	US
Kind code	A1
Filing date	Jun 27, 2019
Priority date	Jun 27, 2019
Publication date	Dec 31, 2020
Grant date	—

Abstract

Systems, methods, and computer program products to transmit, by a web browser to a web server, a hypertext transfer protocol request for a web page at a first uniform resource identifier (URI). The web browser may receive, from the web server via, the web page and metadata of a plurality of candidate advertisements, the plurality of candidate advertisements determined based on a master machine learning (ML) model. A client ML model executing in the web browser may process the received metadata, the client ML model trained based on prior interactions between one or more users of the web browser and a plurality of previously displayed advertisements. The client ML model may determine based on the processing, a first candidate advertisement of the plurality of candidate advertisements to display in the web browser with the received web page. The web browser may receive, from a second URI, the first candidate advertisement of the plurality of candidate advertisements and output the web page and the first candidate advertisement on a display device.

First claim

1 . A system, comprising:
a plurality of processors; and a memory storing instructions which when executed by one or more of the processors, cause the one or more of the processors to:
transmit, by a web browser executing on a first processor of the plurality of processors to a web server executing on a second processor of the plurality of processors, a hypertext transfer protocol (HTTP) request for a web page at a first uniform resource identifier (URI);
receive, by the web browser from the web server via a network, the web page at the first URI and metadata of a plurality of candidate advertisements to display in the web browser with the web page, the plurality of candidate advertisements determined by the web server based on a master machine learning (ML) model trained based on a plurality of users;
process, by a client ML model executing in the web browser, the received metadata of the plurality of candidate advertisements, the client ML model trained based on the master ML model and data stored by the web browser describing prior interactions between one or more users of the web browser and a plurality of previously displayed advertisements;
determine, by the client ML model based on the processing, a first candidate advertisement of the plurality of candidate advertisements to display in the web browser with the received web page;
receive, by the web browser from a second URI, the first candidate advertisement of the plurality of candidate advertisements; and
output, by the web browser, the received web page and the first candidate advertisement for display on a display device.

2 . The system of claim 1 , the memory storing instructions which when executed by one or more of the processors, cause the one or more of the processors to:
receive, by the web browser, the master ML model from the web server; determine, by the web browser for each previously displayed advertisement, whether the one or more users clicked on the previously displayed advertisement; and store, by the web browser in a user profile in the memory, an indication of whether the one or more users clicked on each previously displayed advertisement, wherein the master ML model is associated with a domain comprising a plurality of web pages including the web page at the first URI, wherein the client ML model is associated with the user profile in the memory.

3 . The system of claim 2 , the memory storing instructions which when executed by one or more of the processors, cause the one or more of the processors to:
iteratively train the master ML model based on a ML algorithm and training data to generate the client ML model, the training data comprising: (i) the indications of whether the one or more users clicked on each previously displayed advertisement stored in the user profile, (ii) IP addresses of a plurality of client devices displaying the previously displayed advertisements, (iii) user profile data, (iv) unique identifiers for each previously displayed advertisement, (v) a category of each previously displayed advertisement, and (vi) an entity associated with each previously displayed advertisement; remove a first attribute from the client ML model based on a determination that the first attribute is not relevant in determining whether an advertisement will be clicked; and convert a second attribute of the client ML model from a first data type to a second data type based on a determination that a precision of the first data type is not required to determine whether an advertisement will be clicked, the precision of the first data type greater than a precision of the second data type.

4 . The system of claim 1 , the memory storing instructions which when executed by one or more of the processors, cause the one or more of the processors to:
receive, by the web browser, the metadata of the plurality of candidate advertisements, the metadata comprising an advertisement type, an advertiser type, and an advertisement identifier of each of the plurality of candidate advertisements; and invoke, by the web browser based on processing the received web page, the client ML model to process the metadata of the plurality of candidate advertisements.

5 . The system of claim 4 , the memory storing instructions which when executed by one or more of the processors, cause the one or more of the processors to:
compute, by the client ML model based at least in part on the metadata of the plurality of candidate advertisements and one or more interests of the one or more users of the web browser, a respective score reflecting a probability that the one or more users of the web browser will click on the plurality of candidate advertisements; select, by the client ML model, the first candidate advertisement and a second candidate advertisement of the plurality of candidate advertisements based on the scores for the first and second candidate advertisements; receive, by the web browser from a third URI, the second candidate advertisement of the plurality of candidate advertisements; and output, by the web browser, the received web page and the first and second candidate advertisements for display on the display device.

6 . The system of claim 1 , the memory storing instructions which when executed by one or more of the processors, cause the one or more of the processors to:
determine, by the web server for each of a plurality of advertisements displayed on a plurality of client devices, whether a respective user clicked on each advertisement of the plurality of advertisements displayed on the respective client device; store, by the web server, an indication of whether the user clicked on each advertisement of the plurality of advertisements displayed on the respective client device; and train the master ML model based on a ML algorithm and training data, the training data comprising: (i) the stored indication of whether the user clicked on each advertisement of the plurality of advertisements displayed on the respective client device, (ii) IP addresses of the plurality of client devices, (iii) user profile data, (iv) unique identifiers for each previously displayed advertisement, (v) a category of each previously displayed advertisement, and (vi) an entity associated with each previously displayed advertisement.

7 . The system of claim 1 , wherein the plurality of candidate advertisements are stored by an advertisement server executing on a third processor of the plurality of processors, wherein the web browser receives the first advertisement from the advertisement server.

8 . A non-transitory computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more processors of a plurality of processors to cause the one or more processors to:
receive, by a web server executing on a first processor of the plurality of processors, a hypertext transfer protocol (HTTP) request from a web browser executing on a second processor of the plurality of processors to access a web page associated with a first uniform resource identifier (URI); determine, by the web server based on a master machine learning (ML) model trained based on a plurality of users, a plurality of candidate advertisements to display in the web browser with the web page; transmit, by the web server to the web browser via a network, the web page at the first URI and metadata of the determined plurality of candidate advertisements; receive, by the web server from the web browser, an identifier of a first candidate advertisement of the plurality of candidate advertisements to display in the web browser with the received web page, a client ML model executing in the web browser to process the metadata of the plurality of candidate advertisements

Assignees

• Capital One Services Llc

Inventors

• Watson Mark Louis
• Rafferty Galen
• Walters Austin Grant
• Goodsitt Jeremy Edward
• Truong Anh
• Pham Vincent

Classifications

• G06N20/00

  Machine learning · CPC title

• G06Q30/0277 Primary

  Online advertisement · CPC title