Customized website predictions for machine-learning systems

Patent metadata

Field	Value
Publication number	US-11100421-B2
Application number	US-201615332370-A
Country	US
Kind code	B2
Filing date	Oct 24, 2016
Priority date	Oct 24, 2016
Publication date	Aug 24, 2021
Grant date	Aug 24, 2021

Abstract

In one aspect, a request for web content is received from a user device communicatively coupled to the processing device via the network. In response to receiving the request, user information associated with the user is determined. Predicted responses of the user to each variation of a plurality of variations of the web content are determined using prediction models and the user information. The prediction models include one or more decision trees generated using a splitting criterion requiring a minimum number of positive responses to a variation and a minimum number of negative responses to the variation as a condition of considering the possible split. The variation determined to have a threshold likelihood of yielding a predicted positive response of the predicted responses is selected based on the user information. The variation is transmitted to the user device via the network.

First claim

What is claimed is:

1. A computer-implemented method usable by a machine-learning system to transmit customized content via a network, the computer-implemented method comprising:
determining, by a processing device and in response to receiving a request for web content from a user device that is communicatively coupled to the processing device via the network, user information associated with a user of the user device, wherein the user information is used to determine variables associated with the user;
determining, by the processing device, likelihoods of predicted responses of the user to each variation of a plurality of variations of the web content using prediction models and the user information, wherein each prediction model is configured to predict a response of the user to a variation of the web content corresponding to the prediction model and includes one or more decision trees, wherein a decision tree among the one or more decision trees comprises a set of nodes, a node of the set of nodes corresponds to at least one of the variables, and the node is split into multiple nodes according to a value of the at least one variable and using a splitting criterion requiring: (1) training records falling on one side of a split point contain a minimum number of positive responses to a content corresponding to the variation of the web content and a minimum number of negative responses to the content corresponding to the variation of the web content and (2) training records falling on the other side of the split point contain the minimum number of the positive responses and the minimum number of the negative responses;
selecting, by the processing device, a variation of the plurality of variations determined to have a threshold likelihood of yielding a positive predicted response; and
transmitting, by the processing device, the selected variation to the user device via the network.

2. The computer-implemented method of claim 1 , further including:
receiving a training dataset including historical user information corresponding to a plurality of users and historical content data corresponding to one or more variations of the plurality of variations, wherein the training records are obtained from the training dataset; and
training the prediction models using the training dataset and the splitting criterion.

3. The computer-implemented method of claim 1 , wherein determining the user information associated with the user includes extracting the user information from the request, the user information including at least one of an internet protocol address associated with the user, a browser type associated with the user device, or a previous universal resource locator accessed by the user device.

4. The computer-implemented method of claim 1 , wherein determining the user information associated with the user includes retrieving historical user information from a database that is accessible to the processing device, the historical user information corresponding to one or more previous visits to a webpage hosted by a server including the processing device.

5. The computer-implemented method of claim 4 , wherein the historical user information is stored in the database in a manner that associates the historical user information with the user using a user identifier, the user identifier being extractable from the request.

6. The computer-implemented method of claim 5 , wherein determining the likelihoods of the predicted responses includes evaluating the user information against each of the prediction models, wherein the positive predicted response corresponds to a statistical likelihood, based on the evaluation, that the user will perform a desired action when presented the respective variation.

7. The computer-implemented method of claim 1 , wherein selecting the variation includes:
comparing respective statistical likelihoods, for each of the prediction models, that the user will perform a desired action when presented each variation of the plurality of variations; and
selecting the variation of the plurality of variations corresponding to one of the highest statistical likelihoods of the respective statistical likelihoods.

8. The computer-implemented method of claim 7 , wherein the desired action corresponds to a user purchase subsequent to viewing the variation of the web content.

9. The computer-implemented method of claim 1 , wherein the prediction models correspond to an expectation of a monetary value for one or more of: a purchase order, an item within the purchase order, or a service request.

10. A method comprising:
a step for determining, by a processing device and in response to receiving a request for web content from a user device, user information associated with a user of the user device wherein the user information is used to determine variables associated with the user;
a step for determining, by the processing device, likelihoods of predicted responses of the user to each variation of a plurality of variations of the web content using prediction models and the user information, wherein each prediction model is configured to predict a response of the user to a variation of the web content corresponding to the prediction model and includes one or more decision trees, wherein a decision tree among the one or more decision trees comprises a set of nodes, a node of the set of nodes corresponds to at least one of the variables, and the node is split into multiple nodes according to a value of the at least one variable and using a splitting criterion requiring: (1) training records falling on one side of a split point contain a minimum number of positive responses to a content corresponding to the variation of the web content and a minimum number of negative responses to the content corresponding to the variation of the web content and (2) training records falling on the other side of the split point contain the minimum number of the positive responses and the minimum number of the negative responses;
a step for selecting, by the processing device, a variation of the plurality of variations determined to have a threshold likelihood of yielding a positive predicted response; and

a step for transmitting, by the processing device, the selected variation to the user device via a network.

11. The method of claim 10 , further comprising:
a step for training the prediction models using a training dataset and the splitting criterion, the training dataset including historical user information corresponding to a plurality of users and historical content data corresponding to one or more variations of the plurality of variations, wherein the training records are obtained from the training dataset.

12. The method of claim 10 , further comprising a step for extracting a user identifier from the request,
wherein the step for determining the user information associated with the user of the user device includes a step for retrieving historical user information from a database using the user identifier.

13. The method of claim 10 , wherein the step for determining the likelihoods of the predicted responses includes a step for evaluating the user information against each of the prediction models, wherein the positive predicted response corresponds to a statistical likelihood, based on the evaluation, that the user will perform a desired action when presented a variation of the plurality of variations, wherein each of the prediction models corresponds to a different variation of the plurality of variations.

14. The method of claim 10 , wherein the step for selecting the variation comprises:
a step for comparing respective statistical likelihoods, f

Assignees

• Adobe Inc

Inventors

• Kucera John T

Classifications

• H04L67/535

  Tracking the activity of the user (network monitoring arrangements H04L43/00; recording of computer activity G06F11/34) · CPC title

• G06F16/95

  Retrieval from the web · CPC title

• G06Q30/0243

  Comparative campaigns · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06Q30/0269

  based on user profile or attribute · CPC title