Real-time updates to digital marketing forecast models

• US 10,181,130 B2
• Filed: 06/24/2013
• Issued: 01/15/2019
• Est. Priority Date: 06/24/2013
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Schein, Andrew I.
• Primary Examiner(s): Long, Fonya M
• Assistant Examiner(s): Cirnu, Alexanru
• Application Number:
  US13/925,138
• Publication Number:
  US 20140379460A1
• Time in Patent Office: 2,031 Days
• Field of Search: None
• CPC Class Codes: G06Q 30/0244 Optimization

Abstract

Techniques are disclosed for automatically creating or updating predictive models, including digital marketing forecast models. A predictive model is updated in real-time or near real-time using a stochastic gradient descent optimization method based on one or more predictive values associated with an advertising impression that is won in an online advertising auction. Each predictive value, which is obtained from the predictive model, is encoded as an argument in a uniform resource locator (URL) corresponding to the ad impression being auctioned. If and when the ad impression is won, the predictive value(s) and other information can be tracked and immediately available for updating the model using information encoded in the URL.

Claims

1. A computer-implemented method of digital marketing forecasting comprising:

receiving, by a bidding processor, a request to bid on a digital advertising impression via an online auction;

computing, by the bidding processor, a bid to buy the advertising impression based on a predictive model;

computing, by the bidding processor, a forecast value associated with serving the advertising impression to a user based on the predictive model;

encoding, by the bidding processor, the bid amount and the forecast value in a uniform resource locator (URL);

sending, by the bidding processor, the URL to an auction processor;

receiving, by the bidding processor and subsequent to sending the URL to the auction processor, a message from the auction processor, the message including the URL previously sent to the auction processor, the URL having the bid amount and the forecast value encoded therein;

parsing, by the bidding processor, the encoded forecast value from the URL included in the message;

computing, by the bidding processor, a prediction error representing a difference between an actual value associated with the advertising impression and the associated forecast value parsed from the URL; and

updating, by the bidding processor, the predictive model based at least in part on the prediction error using a stochastic gradient descent optimization method.

2. The method of claim 1, wherein the predictive model includes a plurality of predictive dimensions, and wherein the method further comprises updating the predictive model for each of the plurality of predictive dimensions independently.

3. The method of claim 1, wherein the actual value includes actual cost per advertising impression, and wherein the forecast value includes predicted cost per advertising impression.

4. The method of claim 1, wherein the actual value includes actual revenue per advertising impression, and wherein the forecast value includes predicted revenue per advertising impression.

5. The method of claim 1, further comprising computing the forecast value based at least in part on the updated predictive model.

6. A digital marketing forecasting system, comprising:

a storage; and

a processor operatively coupled to the storage and configured to;

receive a request to bid on a digital advertising impression via an online auction;

compute a bid to buy the advertising impression based on a predictive model;

compute a forecast value associated with serving the advertising impression to a user based on the predictive model;

encode the bid amount and the forecast value in a uniform resource locator (URL);

send the URL to an auction processor;

receive, subsequent to sending the URL to the auction processor, a message from the auction processor, the message including the URL previously sent to the auction processor, the URL having the bid amount and the forecast value encoded therein;

parse the encoded forecast value from the URL included in the message;

compute a prediction error representing a difference between an actual value associated with the advertising impression and the associated forecast value parsed from the URL; and

update the predictive model based at least in part on the prediction error using a stochastic gradient descent optimization method.

7. The system of claim 6, wherein the predictive model includes a plurality of predictive dimensions, and wherein the processor is further configured to update the predictive model for each of the plurality of predictive dimensions independently.

8. The system of claim 6, wherein the actual value includes actual cost per advertising impression, and wherein the forecast value includes predicted cost per advertising impression.

9. The system of claim 6, wherein the actual value includes actual revenue per advertising impression, and wherein the forecast value includes predicted revenue per advertising impression.

10. The system of claim 6, wherein the processor is further configured to compute the forecast value based at least in part on the updated predictive model.

11. A non-transient computer-readable medium having instructions encoded thereon that when executed by a processor cause the processor to:

receive a request to bid on a digital advertising impression via an online auction;

compute a bid to buy the advertising impression based on a predictive model;

compute a forecast value associated with serving the advertising impression to a user based on the predictive model;

encode the bid amount and the forecast value in a uniform resource locator (URL);

send the URL to an auction processor;

receive, subsequent to sending the URL to the auction processor, a message from the auction processor, the message including the URL previously sent to the auction processor, the URL having the bid amount and the forecast value encoded therein;

parse the encoded forecast value from the URL included in the message;

compute a prediction error representing a difference between an actual value associated with the advertising impression and the associated forecast value parsed from the URL; and

update the predictive model based at least in part on the prediction error using a stochastic gradient descent optimization method.

12. The computer-readable medium of claim 11, wherein the predictive model includes a plurality of predictive dimensions, and wherein the computer-readable medium further comprises instructions that when executed by the processor cause the processor to update the predictive model for each of the plurality of predictive dimensions independently.

13. The computer-readable medium of claim 11, wherein the predictive model includes a plurality of predictive dimensions, and wherein the computer-readable medium further comprises instructions that when executed by the processor cause the processor to update the predictive model for each of the plurality of predictive dimensions independently.

14. The computer-readable medium of claim 11, wherein the actual value includes actual cost per advertising impression, and wherein the forecast value includes predicted cost per advertising impression.

15. The computer-readable medium of claim 11, wherein the actual value includes actual revenue per advertising impression, and wherein the forecast value includes predicted revenue per advertising impression.