Estimating conversion rate in display advertising from past performance data

• US 10,037,543 B2
• Filed: 08/13/2012
• Issued: 07/31/2018
• Est. Priority Date: 08/13/2012
• Status: Active Grant

• Current Assignee: Amobee, Inc. (Nexxen International Ltd.)
• Original Assignee: Amobee, Inc. (Nexxen International Ltd.)
• Inventors: Lee, Kuang-Chih, Orten, Burkay Birant, Dasdan, Ali, Li, Wentong
• Primary Examiner(s): Brown, Luis A
• Application Number:
  US13/584,545
• Publication Number:
  US 20140046754A1
• Time in Patent Office: 2,178 Days
• Field of Search: 705 1471, 705 1441, 705 1473, 705 144, 705 1445, 705 1446, 705 1454, 705 1469, 706 12, 707E17001, 707E17046, 707738, 707740, 707741, 707999003
• CPC Class Codes: G06Q 30/0242 Determining effectiveness of advertisements
• CPC Class Codes: G06Q 30/0243 Comparative campaigns
• CPC Class Codes: G06Q 30/0254 based on statistics

Abstract

Embodiments of the invention present an approach to conversion rate estimation which relies on using past performance observations along user, publisher, and advertiser data hierarchies. More specifically, embodiments of the invention model the conversion event at different select hierarchical levels with separate binomial distributions and estimate the distribution parameters individually. It is shown how to combine these individual estimators using logistic regression to identify conversion events accurately. Embodiments of the invention can also handle many practical issues, such as data imbalance, missing data, and output probability calibration, which render this estimation problem more difficult for a real-world implementation of the approach.

Claims

1. A method comprising:

receiving an advertising opportunity call at a real-time bid exchange server of an advertising campaign management system, wherein the advertising opportunity call is generated in response to a request to access a webpage at a web browser application on a client machine;

generating, via the real-time bid exchange server, an advertising opportunity bid request for the advertising opportunity call, the adverting opportunity bid request providing the opportunity to place a bid to transmit an advertisement for presentation at the client machine,

appending, via the real-time bid exchange server, an address of the webpage and a user profile associated with the client machine to the advertising opportunity bid request;

transmitting, via the real-time bid exchange server, the advertising opportunity bid request to a demand side platform of the advertising campaign management system;

transmitting, via a data management platform of the advertising campaign management system, user data and webpage data to the demand side platform;

predicting, via a decision engine of the demand side platform, the likelihood of a conversion event for the advertising opportunity for each of a plurality of advertisements being presented with respect to data hierarchies including one or more of the following;

a user cluster of a plurality of users that include a user corresponding to the user profile specified in the bid request,

a publisher cluster of a plurality of publishers having a same type as a publisher specified by the bid request, and

an advertiser cluster of a plurality of advertisements of a campaign or advertiser to which such each advertisement belongs,

wherein predicting the likelihood of the conversion event for each advertisement is based on a probabilistic distribution based on historical data of conversion events for a plurality of advertisements that were served for all of the users belonging to the user cluster, all of the publishers belonging to the publisher cluster, and all of the advertisements belonging to the advertiser cluster;

for each advertisement of the plurality of advertisements with a predicted likelihood of a conversion event that exceeds a predetermined threshold, transmitting, from the decision engine, an indication of the particular advertisement and an associated bid value amount for such particular advertisement to the real-time bid exchange;

collecting one or more indications and associated bid value amounts at the real-time bid exchange and determining a winning indication corresponding to an advertisement with the highest associated bid value amount;

transmitting the location of an ad creative file to the web browser, wherein the ad creative file is located on a host server and corresponds to the advertisement corresponding to the winning indication; and

retrieving, via the web browser, the ad creative file and rendering the corresponding advertisement on the webpage at the client machine.

2. The method recited in claim 1, wherein each of the respective probabalistic distributions comprises a binomial distribution, and where the likelihood of a conversion event is predicted using a maximum likelihood estimator.

3. The method recited in claim 1, the method further comprising:

calibrating the estimated likelihood of the conversion events with a plurality of actual conversion rate values.

4. The method recited in claim 1, wherein the user cluster, publisher cluster, and advertiser cluster are selected from a plurality of user clusters that each include the user, publisher clusters of the same type as the publisher, and/or advertiser clusters to which the advertisement for which the likelihood is being predicted belongs, respectively using a logistic regression.

5. The method of claim 4, wherein the user cluster, publisher cluster, and advertiser cluster are each formed by combining the respective plurality of user, publisher, and advertiser clusters using a function having a plurality of parameters that are selected to result in a maximized fit between output scores from the function and actual conversion rates for a set of training data and so that the output scores serve to separate conversion and no conversion data.

15. The method of claim 5, wherein a range of output scores from the function are divided into a plurality of equal sized bins that each have a calibration value that is applied to such bin'"'"'s score based on the actual conversion rates.

6. A demand-side platform advertising campaign management system comprising:

a real-time bid exchange server configured to;

receive an advertising opportunity call generated in response to a request to access a webpage at a web browser application on a client machine,

generate an advertising opportunity bid request for the advertising opportunity call, the adverting opportunity bid request providing the opportunity to place a bid to transmit an advertisement for presentation at the client machine, and

append an address of the webpage and a user profile associated with the client machine to the advertising opportunity bid request;

a demand side platform configured to receive the advertising opportunity bid request, the advertising opportunity bid request being transmitted by the real-time bid exchange server;

a decision engine of the demand side platform, wherein the decision engine is operable to;

predict the likelihood of a conversion event for the advertising opportunity for each of a plurality of advertisements being presented with respect to data hierarchies including one or more of the following;

a user cluster of a plurality of users that include a user corresponding to the user profile specified in the bid request,

a publisher cluster of a plurality of publishers having a same type as a publisher specified by the bid request, and

an advertiser cluster of a plurality of advertisements of a campaign or advertiser to which such each advertisement belongs,

wherein predicting the likelihood of the conversion event for each advertisement is based on a probabilistic distribution based on historical data of conversion events for a plurality of advertisements that were served for all of the users belonging to the user cluster, all of the publishers belonging to the publisher cluster, and all of the advertisements belonging to the advertiser cluster;

for each advertisement of the plurality of advertisements with a predicted likelihood of a conversion event that exceeds a predetermined threshold, the decision engine, transmit an indication of the particular advertisement and an associated bid value amount for such particular advertisement to the real-time bid exchange;

wherein the real-time bid exchange is further configured to;

collect one or more indications and associated bid value amounts,

determine a winning indication corresponding to an advertisement with the highest associated bid value amount, and

transmit the location of an ad creative file to the web browser, wherein the ad creative file is located on a host server and corresponds to the advertisement corresponding to the winning indication;

wherein the ad creative file is retrieved by the web browser and the corresponding advertisement is rendered on the webpage at the client machine.

7. The system recited in claim 6, wherein each of the respective probabalistic distributions comprises a binomial distribution, and where the likelihood of a conversion event is predicted using a maximum likelihood estimator.

8. The system recited in claim 6, wherein the processor is further operable to:

calibrate the estimated likelihood of the conversion events with a plurality of actual conversion rate values.

9. The system recited in claim 6, wherein the user cluster, publisher cluster, and advertiser cluster are selected from a plurality of user clusters that each include the user, publisher clusters of the same type as the publisher, and/or advertiser clusters to which the advertisement for which the likelihood is being predicted belongs, respectively using a logistic regression.

10. The system of claim 6, wherein the user cluster, publisher cluster, and advertiser cluster are each formed by combining the respective plurality of user, publisher, and advertiser clusters using a function having a plurality of parameters that are selected to result in a maximized fit between output scores from the function and actual conversion rates for a set of training data and so that the output scores serve to separate conversion and no conversion data.

16. The system of claim 10, wherein a range of output scores from the function are divided into a plurality of equal sized bins that each have a calibration value that is applied to such bin'"'"'s score based on the actual conversion rates.

11. One or more non-transitory computer readable media having instructions stored thereon for performing a method, the method comprising:

receiving an advertising opportunity call at a real-time bid exchange server of an advertising campaign management system, wherein the advertising opportunity call is generated in response to a request to access a webpage at a web browser application on a client machine;

generating, via the real-time bid exchange server, an advertising opportunity bid request for the advertising opportunity call, the adverting opportunity bid request providing the opportunity to place a bid to transmit an advertisement for presentation at the client machine,

appending, via the real-time bid exchange server, an address of the webpage and a user profile associated with the client machine to the advertising opportunity bid request;

transmitting, via the real-time bid exchange server, the advertising opportunity bid request to a demand side platform of the advertising campaign management system;

transmitting, via a data management platform of the advertising campaign management system, user data and webpage data to the demand side platform;

predicting, via a decision engine of the demand side platform, the likelihood of a conversion event for the advertising opportunity for each of a plurality of advertisements being presented with respect to data hierarchies including one or more of the following;

a user cluster of a plurality of users that include a user corresponding to the user profile specified in the bid request,

a publisher cluster of a plurality of publishers having a same type as a publisher specified by the bid request, and

an advertiser cluster of a plurality of advertisements of a campaign or advertiser to which such each advertisement belongs,

wherein predicting the likelihood of the conversion event for each advertisement is based on a probabilistic distribution based on historical data of conversion events for a plurality of advertisements that were served for all of the users belonging to the user cluster, all of the publishers belonging to the publisher cluster, and all of the advertisements belonging to the advertiser cluster;

for each advertisement of the plurality of advertisements with a predicted likelihood of a conversion event that exceeds a predetermined threshold, transmitting, from the decision engine, an indication of the particular advertisement and an associated bid value amount for such particular advertisement to the real-time bid exchange;

collecting one or more indications and associated bid value amounts at the real-time bid exchange and determining a winning indication corresponding to an advertisement with the highest associated bid value amount;

transmitting the location of an ad creative file to the web browser, wherein the ad creative file is located on a host server and corresponds to the advertisement corresponding to the winning indication; and

retrieving, via the web browser, the ad creative file and rendering the corresponding advertisement on the webpage at the client machine.

12. The one or more non-transitory computer readable media recited in claim 11, wherein each of the respective probabalistic distributions comprises a binomial distribution, and where the likelihood of a conversion event is predicted using a maximum likelihood estimator.

13. The one or more non-transitory computer readable media recited in claim 11, the method further comprising:

calibrating the estimated likelihood of the conversion events with a plurality of actual conversion rate values.

14. The one or more non-transitory computer readable media recited in claim 11, wherein the user cluster, publisher cluster, and advertiser cluster are selected from a plurality of user clusters that each include the user, publisher clusters of the same type as the publisher, and/or advertiser clusters to which the advertisement for which the likelihood is being predicted belongs, respectively using a logistic regression.