Accurately estimating advertisement performance

• US 7,818,208 B1
• Filed: 06/28/2005
• Issued: 10/19/2010
• Est. Priority Date: 06/28/2005
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Veach, Eric Hugh
• Primary Examiner(s): MYHRE, JAMES W
• Application Number:
  US11/167,581
• Time in Patent Office: 1,939 Days
• Field of Search: 705/14.4, 705/14.49
• US Class Current: 705/14.49
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0241 Advertisements
• CPC Class Codes: G06Q 30/0251 Targeted advertisements
• CPC Class Codes: G06Q 30/0252 based on events or environment, e.g. weather or festivals
• CPC Class Codes: G06Q 30/0253 During e-commerce, i.e. online transactions
• CPC Class Codes: G06Q 30/0254 based on statistics
• CPC Class Codes: G06Q 30/0255 based on user history
• CPC Class Codes: G06Q 30/0256 User search
• CPC Class Codes: G06Q 30/0257 User requested

Abstract

An on-line advertising system provides advertisements to users. At least one candidate advertisement is identified in response to an advertisement trigger. At least one key is identified relating to features of the candidate advertisement, the advertisement trigger, or a combination of the candidate advertisement and the advertisement trigger. Candidate advertisement performance is estimated based on historical data relating to the key.

Claims

1. A method for estimating on-line advertisement performance, the method comprising:

identifying, using a processor associated with a server device, a candidate advertisement in response to an advertisement trigger;

identifying, using a processor associated with a server device, a set of at least one key pair relating to features of the candidate advertisement and the advertisement trigger, or a combination of the candidate advertisement and the advertisement trigger, where the at least one key pair identifies a type of the features and a value of the features; and

estimating, using a processor associated with a server device, candidate advertisement performance based on historical data relating to the at least one key pair, where estimating advertisement performance includes;

identifying, using a processor associated with a server device, a prior likelihood that a defined user action associated with the candidate advertisement will be observed;

calculating, using a processor associated with a server device, for each key pair in the set of at least one key pair, a parameter representing a statistical impact of the key pair on the likelihood that a defined user action associated with the candidate advertisement will be observed, where the at least one parameter is based on the historical data associated with the key pair.

2. The method of claim 1, where the advertisement trigger includes a received search query.

3. The method of claim 1, where the advertisement trigger includes a web document.

4. The method of claim 1, further comprising:

identifying the at least one key pair based on properties of the advertisement trigger.

5. The method of claim 1, further comprising:

filtering a key from the at least one key pair based on predetermined criteria.

6. The method of claim 5, where the predetermined criteria includes a time since an impression relating to the at least one key was observed.

7. The method of claim 6, where the predetermined criteria includes a number of impressions relating to the at least one key within a predetermined time period.

8. The method of claim 1, where estimating candidate advertisement performance, further comprises:

estimating a probability that the candidate advertisement will be clicked.

9. The method of claim 1, where estimating candidate advertisement performance, further comprises:

estimating a probability that a user will make a purchase associated with the candidate advertisement.

10. The method of claim 1, where estimating candidate advertisement performance, further comprises:

estimating a probability that a user action associated with the candidate advertisement will be observed.

11. The method of claim 1, where the estimating a candidate advertisement performance further comprises:

estimating advertisement performance based on the prior likelihood and the parameter for each key in the at least one key pair.

12. The method of claim 11, where the at least one parameter representing the statistical impact of each key pair in the at least one key pair has a multiplicative effect on the odds that the user action will be observed.

13. The method of claim 12, where calculating the parameter for a given key pair in the set of at least one key pair, further comprises:

calculating a self-excluding probability representing an estimated likelihood that the user action associated with the candidate advertisement will be observed when disregarding the parameter associated with the given key pair;

identifying a number of impressions relating to the given key pair and a number of user actions relating to the given key pair based on the historical data; and

setting the parameter for the given key pair based on the self-excluding probability and the number of impressions.

14. The method of claim 13, further comprising:

identifying a confidence interval based on the number of impressions relating to the given key pair and a number of user actions relating to the given key pair.

15. The method of claim 14, further comprising:

determining whether the self-excluding probability falls within the confidence interval;

setting the parameter for the given key pair to 1 if it is determined that the self-excluding probability falls within the confidence interval, thereby rendering the given key pair statistically invisible; and

setting the parameter for the given key pair equal to the minimum adjustment necessary to bring the self-excluding probability within the confidence interval if it is determined that the self-excluding probability does not fall within the confidence interval.

16. The method of claim 14, where calculating the self-excluding probability further comprises:

calculating a moving average of the self-excluding probability.

17. The method of claim 1, where the historical data relating to the at least one key includes log data generated upon observation of prior impressions relating to the at least one key pair.

18. A method, comprising:

receiving, using a processor associated with a server device, an advertisement trigger from a user;

identifying, using a processor associated with a server device, a candidate advertisement for display to the user;

identifying, using a processor associated with a server device, a set of key pairs having characteristics associated with the advertisement trigger and the candidate advertisement;

retrieving, using a processor associated with a server device, historical statistical information for each key in the set of key pairs;

calculating, using a processor associated with a server device, a model parameter for each key pair in the set of key pairs, the model parameter representative of a statistical relevance of the characteristics associated with the key and based on the retrieved historical statistical information; and

calculating, using a processor associated with a server device, an estimated user action rate for the candidate advertisement based on a prior user action probability associated with the candidate advertisement and each of the calculated model parameters.

19. The method of claim 18, where the historical statistical information for each key pair includes a number of impressions observed for the key pair and a number of user actions observed for the key pair.

20. The method of claim 19, where calculating a model parameter for a first key pair in the set of key pairs, includes:

calculating a self-excluding probability for the first key pair representing a likelihood of observing a user action when removing the first key pair from consideration;

calculating a confidence interval based on the number of impressions and the number of user actions historically observed for the first key pair;

determining whether the self-excluding probability falls within the confidence interval;

setting the model parameter for the first key pair equal to 1 when it is determined that the self-excluding probability falls within the confidence interval; and

setting the model parameter for the first key pair equal to a minimum adjustment necessary to bring the self-excluding probability within the confidence interval when it is determined that the self-excluding probability does not fall within the confidence interval.

21. The method of claim 18, where calculating an estimated user action rate further comprises calculating the estimated user action rate as a product of the prior probability and the model parameters for each key pair in the set of key pairs.