Estimating ad quality from observed user behavior

• US 20070156514A1
• Filed: 12/30/2005
• Published: 07/05/2007
• Est. Priority Date: 12/30/2005
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Pregibon, Daryl, Wright, Daniel, Tang, Diane
• Granted Patent: US 8,065,184 B2
• Time in Patent Office: Days
• US Class Current: 705/14.410
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0242 Determining effectiveness of advertisements
• 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 30/0257 User requested
• CPC Class Codes: G06Q 30/0272 Period of advertisement exposure

Abstract

A system obtains ratings associated with a first set of advertisements hosted by one or more servers, where the ratings indicate a quality of the first set of advertisements. The system observes multiple different first user actions associated with user selection of advertisements of the first set of advertisements and derives a statistical model using the observed first user actions and the obtained ratings. The system further observes second user actions associated with user selection of a second advertisement hosted by the one or more servers and uses the statistical model and the second user actions to estimate a quality of the second advertisement.

Claims

1. A method, comprising:

obtaining ratings associated with a first set of advertisements hosted by one or more servers, where the ratings indicate a quality of the first set of advertisements;

observing multiple different first user actions associated with user selection of advertisements of the first set of advertisements;

deriving a statistical model using the observed first user actions and the obtained ratings;

observing second user actions associated with user selection of a second advertisement hosted by the one or more servers; and

using the statistical model and the second user actions to estimate a quality of the second advertisement.

2. The method of claim 1, wherein observing the multiple different first user actions comprises:

logging the multiple different first user actions.

3. The method of claim 1, wherein the multiple different first user actions include user actions other than, or in addition to, a click through rate (CTR).

4. The method of claim 1, wherein deriving the statistical model comprises:

using at least one of logistic regression, regression trees or boosted stumps to generate the statistical model.

5. The method of claim 1, wherein deriving the statistical model comprises:

testing a fit of the statistical model to determine which of the observed first user actions are correlated with good or bad advertisements.

6. The method of claim 1, wherein using the statistical model and the second user actions comprises:

determining, using the statistical model, a probability that the second advertisement is a good advertisement given the second user actions; and

estimating a quality of the second advertisement based on the determined probability.

7. The method of claim 6, wherein using the statistical model and the second user actions comprises:

determining a score associated with the second advertisement using the probability that the second advertisement is a good advertisement; and

estimating a quality of the second advertisement based on the determined score.

8. The method of claim 7, wherein determining a score associated with the second advertisement comprises:

applying a function to the probability that the second advertisement is a good advertisement to determine the score.

9. The method of claim 1, wherein the multiple different first user actions comprise at least one of the following:

a duration of an ad selection;

a number of selections on other ads before and/or after a given ad selection;

a number of selections of search results before and/or after a given ad selection;

a number of selections of other types of results before and/or after a given ad selection;

a number of document views before and/or after a given ad selection;

a number of search queries before and/or after a given ad selection;

a number of search queries associated with a user session that show ads;

a number of repeat selections on a same given ad;

or

an indication of whether a given ad selection was the last ad selection for a given query or the last selection in a user session.

10. A method, comprising:

observing first user behavior associated with user selection of a first advertisement hosted by a server, where the first observed user behavior includes user behavior other than, or in addition to, a click through rate (CTR); and

estimating a quality of the first advertisement based on the observed first user behavior.

11. The method of claim 10, wherein the observed user behavior includes at least one of the following observed user behavior:

a duration of an ad selection;

a number of selections on other ads before and/or after a given ad selection;

a number of selections of search results before and/or after a given ad selection;

a number of selections of other types of results before and/or after a given ad selection;

a number of document views before and/or after a given ad selection;

a number of search queries before and/or after a given ad selection;

a number of queries associated with a user session that show ads;

a number of repeat selections on a same given ad;

or

an indication of whether a given ad selection was the last ad selection for a given query or the last selection in a user session.

12. The method of claim 10, further comprising:

obtaining ratings associated with a set of ads, where the ratings indicate a quality of each ad in the set of ads;

observing second user behavior associated with user selection of ads in the set of ads; and

constructing a statistical model using the ratings and observed second user behavior,

wherein estimating the quality of the first advertisement comprises;

using the statistical model and the observed first user behavior to estimate the quality of the first ad.

13. The method of claim 12, wherein constructing the statistical model comprises:

using at least one of logistic regression, regression trees or boosted stumps to generate the statistical model.

14. The method of claim 12, wherein using the statistical model and the observed first user behavior comprises:

determining, using the statistical model, a probability that the first ad is a good ad given the observed first user behavior; and

determining a score associated with the first ad based on the determined probability.

15. The method of claim 14, wherein using the statistical model and the observed first user behavior comprises:

estimating a quality of the first ad based on the determined score.

16. A computer-readable medium that stores computer-executable instructions, comprising:

instructions for identifying user behavior associated with user selection of a first advertisement hosted by a server, where the observed user behavior includes user behavior other than, or in addition to, a click through rate (CTR); and

instructions for estimating a quality of the first advertisement based on the user behavior.

17. A method, comprising:

observing multiple user actions associated with user selections of first advertisements hosted by one or more servers;

correlating known qualities associated with the first advertisements with certain ones of the multiple user actions;

observing the presence of the certain ones of the multiple user actions associated with user selection of a second advertisement; and

estimating a quality of the second advertisement based on the presence of the certain ones of the multiple user actions.

18. The method of claim 17, wherein the multiple user actions include user actions other than, or in addition to, a click through rate (CTR).

19. The method of claim 17, wherein observing the multiple user actions comprises:

logging the multiple user actions associated with user selections of the first advertisements hosted by one or more servers.

20. The method of claim 17, wherein the multiple user actions comprise at least one of the following:

a duration of an ad selection;

a number of selections on other ads before and/or after a given ad selection;

a number of selections of search results before and/or after a given ad selection;

a number of selections of other types of results before and/or after a given ad selection;

a number of document views before and/or after a given ad selection;

a number of search queries before and/or after a given ad selection;

a number of search queries associated with a user session that show ads;

a number of repeat selections on a same given ad;

or

an indication of whether a given ad selection was the last ad selection for a given query or the last selection in a user session.

21. The method of claim 17, wherein correlating known qualities associated with the first advertisements with certain ones of the multiple user actions comprises:

using at least one of logistic regression, regression trees or boosted stumps to correlate the known qualities.

22. The method of claim 21, wherein correlating known qualities associated with the first advertisements comprises:

constructing a statistical model that determines a probability that advertisements are good given the multiple user actions; and

wherein determining the quality of the second advertisement comprises;

estimating the quality of the second advertisement using the statistical model and the presence of the certain ones of the multiple user actions.

23. The method of claim 22, wherein correlating known qualities associated with the first advertisements comprises:

testing a fit of the statistical model to determine that the certain ones of the multiple user actions are correlated with good ads.

24. A system, comprising:

means for obtaining ratings associated with a first set of ads hosted by one or more servers, where the ratings indicate a quality of the first set of advertisements;

means for identifying multiple different first user actions associated with user selection of ads of the first set of advertisements;

means for deriving a statistical model based on the first user actions and the obtained ratings;

means for identifying second user actions associated with user selection of a second advertisement hosted by the one or more servers; and

means for estimating a quality of the second advertisement based on the statistical model and the second user actions.

25. A method, comprising:

logging user actions associated with user selection of a group of advertisements, where the group of advertisements is associated with quality ratings and where the user actions include user actions other than, or in addition to, a click through rate (CTR);

generating a statistical model using the logged user actions and the quality ratings; and

estimating qualities of advertisements not included in the group of advertisements using the statistical model.